promptfoo 0.121.1 → 0.121.3

package/dist/src/index.d.cts

@@ -88,7 +88,12 @@ declare const ProviderEnvOverridesSchema: z.ZodObject<{
   OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
   OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
   OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+  CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+  CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+  CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
   CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+  OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+  OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
   OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
   OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
   PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -237,6 +242,13 @@ interface ChatMessage {
   role: 'system' | 'user' | 'assistant' | 'tool' | 'function';
   content: string;
 }
+interface SkillCallEntry {
+  name: string;
+  input?: unknown;
+  path?: string;
+  source?: 'heuristic' | 'tool';
+  is_error?: boolean;
+}
 type ProviderTypeMap = Partial<Record<ProviderType, string | ProviderOptions | ApiProvider>>;
 interface AtomicTestCase$1 {
   description?: string;
@@ -574,7 +586,12 @@ declare const ProvidersSchema: z.ZodUnion<readonly [z.ZodString, z.ZodCustom<Cal
     OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
     OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
     OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+    OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+    OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -701,7 +718,12 @@ declare const ProvidersSchema: z.ZodUnion<readonly [z.ZodString, z.ZodCustom<Cal
     OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
     OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
     OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+    OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+    OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -1646,6 +1668,12 @@ declare const BaseAssertionTypesSchema: z.ZodEnum<{
   "similar:euclidean": "similar:euclidean";
   "starts-with": "starts-with";
   "tool-call-f1": "tool-call-f1";
+  "skill-used": "skill-used";
+  "trajectory:goal-success": "trajectory:goal-success";
+  "trajectory:tool-args-match": "trajectory:tool-args-match";
+  "trajectory:step-count": "trajectory:step-count";
+  "trajectory:tool-sequence": "trajectory:tool-sequence";
+  "trajectory:tool-used": "trajectory:tool-used";
   "trace-error-spans": "trace-error-spans";
   "trace-span-count": "trace-span-count";
   "trace-span-duration": "trace-span-duration";
@@ -1714,13 +1742,19 @@ declare const NotPrefixedAssertionTypesSchema: z.ZodPipe<z.ZodEnum<{
   "similar:euclidean": "similar:euclidean";
   "starts-with": "starts-with";
   "tool-call-f1": "tool-call-f1";
+  "skill-used": "skill-used";
+  "trajectory:goal-success": "trajectory:goal-success";
+  "trajectory:tool-args-match": "trajectory:tool-args-match";
+  "trajectory:step-count": "trajectory:step-count";
+  "trajectory:tool-sequence": "trajectory:tool-sequence";
+  "trajectory:tool-used": "trajectory:tool-used";
   "trace-error-spans": "trace-error-spans";
   "trace-span-count": "trace-span-count";
   "trace-span-duration": "trace-span-duration";
   "search-rubric": "search-rubric";
   webhook: "webhook";
   "word-count": "word-count";
-}>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>;
+}>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>;
 declare const AssertionTypeSchema: z.ZodUnion<readonly [z.ZodEnum<{
   "answer-relevance": "answer-relevance";
   bleu: "bleu";
@@ -1775,6 +1809,12 @@ declare const AssertionTypeSchema: z.ZodUnion<readonly [z.ZodEnum<{
   "similar:euclidean": "similar:euclidean";
   "starts-with": "starts-with";
   "tool-call-f1": "tool-call-f1";
+  "skill-used": "skill-used";
+  "trajectory:goal-success": "trajectory:goal-success";
+  "trajectory:tool-args-match": "trajectory:tool-args-match";
+  "trajectory:step-count": "trajectory:step-count";
+  "trajectory:tool-sequence": "trajectory:tool-sequence";
+  "trajectory:tool-used": "trajectory:tool-used";
   "trace-error-spans": "trace-error-spans";
   "trace-span-count": "trace-span-count";
   "trace-span-duration": "trace-span-duration";
@@ -1835,13 +1875,19 @@ declare const AssertionTypeSchema: z.ZodUnion<readonly [z.ZodEnum<{
   "similar:euclidean": "similar:euclidean";
   "starts-with": "starts-with";
   "tool-call-f1": "tool-call-f1";
+  "skill-used": "skill-used";
+  "trajectory:goal-success": "trajectory:goal-success";
+  "trajectory:tool-args-match": "trajectory:tool-args-match";
+  "trajectory:step-count": "trajectory:step-count";
+  "trajectory:tool-sequence": "trajectory:tool-sequence";
+  "trajectory:tool-used": "trajectory:tool-used";
   "trace-error-spans": "trace-error-spans";
   "trace-span-count": "trace-span-count";
   "trace-span-duration": "trace-span-duration";
   "search-rubric": "search-rubric";
   webhook: "webhook";
   "word-count": "word-count";
-}>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+}>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
   "select-best": "select-best";
   human: "human";
   "max-score": "max-score";
@@ -1904,6 +1950,12 @@ declare const AssertionSetSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
@@ -1964,13 +2016,19 @@ declare const AssertionSetSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
       "search-rubric": "search-rubric";
       webhook: "webhook";
       "word-count": "word-count";
-    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
       "select-best": "select-best";
       human: "human";
       "max-score": "max-score";
@@ -2052,6 +2110,12 @@ declare const AssertionSchema: z.ZodObject<{
     "similar:euclidean": "similar:euclidean";
     "starts-with": "starts-with";
     "tool-call-f1": "tool-call-f1";
+    "skill-used": "skill-used";
+    "trajectory:goal-success": "trajectory:goal-success";
+    "trajectory:tool-args-match": "trajectory:tool-args-match";
+    "trajectory:step-count": "trajectory:step-count";
+    "trajectory:tool-sequence": "trajectory:tool-sequence";
+    "trajectory:tool-used": "trajectory:tool-used";
     "trace-error-spans": "trace-error-spans";
     "trace-span-count": "trace-span-count";
     "trace-span-duration": "trace-span-duration";
@@ -2112,13 +2176,19 @@ declare const AssertionSchema: z.ZodObject<{
     "similar:euclidean": "similar:euclidean";
     "starts-with": "starts-with";
     "tool-call-f1": "tool-call-f1";
+    "skill-used": "skill-used";
+    "trajectory:goal-success": "trajectory:goal-success";
+    "trajectory:tool-args-match": "trajectory:tool-args-match";
+    "trajectory:step-count": "trajectory:step-count";
+    "trajectory:tool-sequence": "trajectory:tool-sequence";
+    "trajectory:tool-used": "trajectory:tool-used";
     "trace-error-spans": "trace-error-spans";
     "trace-span-count": "trace-span-count";
     "trace-span-duration": "trace-span-duration";
     "search-rubric": "search-rubric";
     webhook: "webhook";
     "word-count": "word-count";
-  }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+  }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
     "select-best": "select-best";
     human: "human";
     "max-score": "max-score";
@@ -2201,6 +2271,12 @@ declare const AssertionOrSetSchema: z.ZodUnion<readonly [z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
@@ -2261,13 +2337,19 @@ declare const AssertionOrSetSchema: z.ZodUnion<readonly [z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
       "search-rubric": "search-rubric";
       webhook: "webhook";
       "word-count": "word-count";
-    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
       "select-best": "select-best";
       human: "human";
       "max-score": "max-score";
@@ -2347,6 +2429,12 @@ declare const AssertionOrSetSchema: z.ZodUnion<readonly [z.ZodObject<{
     "similar:euclidean": "similar:euclidean";
     "starts-with": "starts-with";
     "tool-call-f1": "tool-call-f1";
+    "skill-used": "skill-used";
+    "trajectory:goal-success": "trajectory:goal-success";
+    "trajectory:tool-args-match": "trajectory:tool-args-match";
+    "trajectory:step-count": "trajectory:step-count";
+    "trajectory:tool-sequence": "trajectory:tool-sequence";
+    "trajectory:tool-used": "trajectory:tool-used";
     "trace-error-spans": "trace-error-spans";
     "trace-span-count": "trace-span-count";
     "trace-span-duration": "trace-span-duration";
@@ -2407,13 +2495,19 @@ declare const AssertionOrSetSchema: z.ZodUnion<readonly [z.ZodObject<{
     "similar:euclidean": "similar:euclidean";
     "starts-with": "starts-with";
     "tool-call-f1": "tool-call-f1";
+    "skill-used": "skill-used";
+    "trajectory:goal-success": "trajectory:goal-success";
+    "trajectory:tool-args-match": "trajectory:tool-args-match";
+    "trajectory:step-count": "trajectory:step-count";
+    "trajectory:tool-sequence": "trajectory:tool-sequence";
+    "trajectory:tool-used": "trajectory:tool-used";
     "trace-error-spans": "trace-error-spans";
     "trace-span-count": "trace-span-count";
     "trace-span-duration": "trace-span-duration";
     "search-rubric": "search-rubric";
     webhook: "webhook";
     "word-count": "word-count";
-  }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+  }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
     "select-best": "select-best";
     human: "human";
     "max-score": "max-score";
@@ -2621,7 +2715,12 @@ declare const TestCaseSchema: z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -2740,6 +2839,12 @@ declare const TestCaseSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -2800,13 +2905,19 @@ declare const TestCaseSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -2886,6 +2997,12 @@ declare const TestCaseSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
@@ -2946,13 +3063,19 @@ declare const TestCaseSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
       "search-rubric": "search-rubric";
       webhook: "webhook";
       "word-count": "word-count";
-    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
       "select-best": "select-best";
       human: "human";
       "max-score": "max-score";
@@ -3166,7 +3289,12 @@ declare const TestCaseWithVarsFileSchema: z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -3285,6 +3413,12 @@ declare const TestCaseWithVarsFileSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -3345,13 +3479,19 @@ declare const TestCaseWithVarsFileSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -3431,6 +3571,12 @@ declare const TestCaseWithVarsFileSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
@@ -3491,13 +3637,19 @@ declare const TestCaseWithVarsFileSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
       "search-rubric": "search-rubric";
       webhook: "webhook";
       "word-count": "word-count";
-    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
       "select-best": "select-best";
       human: "human";
       "max-score": "max-score";
@@ -3710,7 +3862,12 @@ declare const TestCasesWithMetadataSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -3829,6 +3986,12 @@ declare const TestCasesWithMetadataSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -3889,13 +4052,19 @@ declare const TestCasesWithMetadataSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -3975,6 +4144,12 @@ declare const TestCasesWithMetadataSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -4035,13 +4210,19 @@ declare const TestCasesWithMetadataSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -4312,7 +4493,12 @@ declare const ScenarioSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -4431,6 +4617,12 @@ declare const ScenarioSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -4491,13 +4683,19 @@ declare const ScenarioSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -4577,6 +4775,12 @@ declare const ScenarioSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -4637,13 +4841,19 @@ declare const ScenarioSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -4852,7 +5062,12 @@ declare const ScenarioSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -4971,6 +5186,12 @@ declare const ScenarioSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -5031,13 +5252,19 @@ declare const ScenarioSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -5117,6 +5344,12 @@ declare const ScenarioSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -5177,13 +5410,19 @@ declare const ScenarioSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -5393,7 +5632,12 @@ declare const AtomicTestCaseSchema: z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -5512,6 +5756,12 @@ declare const AtomicTestCaseSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -5572,13 +5822,19 @@ declare const AtomicTestCaseSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -5658,6 +5914,12 @@ declare const AtomicTestCaseSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
@@ -5718,13 +5980,19 @@ declare const AtomicTestCaseSchema: z.ZodObject<{
       "similar:euclidean": "similar:euclidean";
       "starts-with": "starts-with";
       "tool-call-f1": "tool-call-f1";
+      "skill-used": "skill-used";
+      "trajectory:goal-success": "trajectory:goal-success";
+      "trajectory:tool-args-match": "trajectory:tool-args-match";
+      "trajectory:step-count": "trajectory:step-count";
+      "trajectory:tool-sequence": "trajectory:tool-sequence";
+      "trajectory:tool-used": "trajectory:tool-used";
       "trace-error-spans": "trace-error-spans";
       "trace-span-count": "trace-span-count";
       "trace-span-duration": "trace-span-duration";
       "search-rubric": "search-rubric";
       webhook: "webhook";
       "word-count": "word-count";
-    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+    }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
       "select-best": "select-best";
       human: "human";
       "max-score": "max-score";
@@ -5981,7 +6249,12 @@ declare const TestSuiteSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -6100,6 +6373,12 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -6160,13 +6439,19 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -6246,6 +6531,12 @@ declare const TestSuiteSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -6306,13 +6597,19 @@ declare const TestSuiteSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -6523,7 +6820,12 @@ declare const TestSuiteSchema: z.ZodObject<{
           OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
           OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
           OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+          OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+          OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -6642,6 +6944,12 @@ declare const TestSuiteSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
@@ -6702,13 +7010,19 @@ declare const TestSuiteSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
             "search-rubric": "search-rubric";
             webhook: "webhook";
             "word-count": "word-count";
-          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
             "select-best": "select-best";
             human: "human";
             "max-score": "max-score";
@@ -6788,6 +7102,12 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -6848,13 +7168,19 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -7063,7 +7389,12 @@ declare const TestSuiteSchema: z.ZodObject<{
           OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
           OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
           OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+          OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+          OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -7182,6 +7513,12 @@ declare const TestSuiteSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
@@ -7242,13 +7579,19 @@ declare const TestSuiteSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
             "search-rubric": "search-rubric";
             webhook: "webhook";
             "word-count": "word-count";
-          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
             "select-best": "select-best";
             human: "human";
             "max-score": "max-score";
@@ -7328,6 +7671,12 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -7388,13 +7737,19 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -7605,7 +7960,12 @@ declare const TestSuiteSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -7722,6 +8082,12 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -7782,13 +8148,19 @@ declare const TestSuiteSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -7868,6 +8240,12 @@ declare const TestSuiteSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -7928,13 +8306,19 @@ declare const TestSuiteSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -8134,7 +8518,12 @@ declare const TestSuiteSchema: z.ZodObject<{
     OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
     OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
     OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+    OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+    OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -8303,7 +8692,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -8430,7 +8824,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -8573,7 +8972,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -8692,6 +9096,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -8752,13 +9162,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -8838,6 +9254,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -8898,13 +9320,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -9121,7 +9549,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
           OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
           OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+          OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+          OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -9240,6 +9673,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
@@ -9300,13 +9739,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
             "search-rubric": "search-rubric";
             webhook: "webhook";
             "word-count": "word-count";
-          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
             "select-best": "select-best";
             human: "human";
             "max-score": "max-score";
@@ -9386,6 +9831,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -9446,13 +9897,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -9661,7 +10118,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
           OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
           OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+          OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+          OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -9780,6 +10242,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
@@ -9840,13 +10308,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
             "search-rubric": "search-rubric";
             webhook: "webhook";
             "word-count": "word-count";
-          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
             "select-best": "select-best";
             human: "human";
             "max-score": "max-score";
@@ -9926,6 +10400,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -9986,13 +10466,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -10203,7 +10689,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -10320,6 +10811,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -10380,13 +10877,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -10466,6 +10969,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -10526,13 +11035,19 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -10737,7 +11252,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
     OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
     OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
     OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+    OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+    OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -10883,7 +11403,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -11078,7 +11603,12 @@ declare const TestSuiteConfigSchema: z.ZodObject<{
         OPENAI_API_KEY?: string | undefined;
         OPENAI_BASE_URL?: string | undefined;
         OPENAI_ORGANIZATION?: string | undefined;
+        CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+        CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+        CLAWDBOT_GATEWAY_URL?: string | undefined;
         CODEX_API_KEY?: string | undefined;
+        OPENCLAW_CONFIG_PATH?: string | undefined;
+        OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
         OPENCLAW_GATEWAY_TOKEN?: string | undefined;
         OPENCLAW_GATEWAY_URL?: string | undefined;
         PALM_API_HOST?: string | undefined;
@@ -11286,7 +11816,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -11405,6 +11940,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -11465,13 +12006,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -11551,6 +12098,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -11611,13 +12164,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -11834,7 +12393,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
           OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
           OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+          OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+          OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -11953,6 +12517,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
@@ -12013,13 +12583,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
             "search-rubric": "search-rubric";
             webhook: "webhook";
             "word-count": "word-count";
-          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
             "select-best": "select-best";
             human: "human";
             "max-score": "max-score";
@@ -12099,6 +12675,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -12159,13 +12741,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -12374,7 +12962,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
           OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
           OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+          CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+          OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+          OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
           OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
           PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -12493,6 +13086,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
@@ -12553,13 +13152,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
             "similar:euclidean": "similar:euclidean";
             "starts-with": "starts-with";
             "tool-call-f1": "tool-call-f1";
+            "skill-used": "skill-used";
+            "trajectory:goal-success": "trajectory:goal-success";
+            "trajectory:tool-args-match": "trajectory:tool-args-match";
+            "trajectory:step-count": "trajectory:step-count";
+            "trajectory:tool-sequence": "trajectory:tool-sequence";
+            "trajectory:tool-used": "trajectory:tool-used";
             "trace-error-spans": "trace-error-spans";
             "trace-span-count": "trace-span-count";
             "trace-span-duration": "trace-span-duration";
             "search-rubric": "search-rubric";
             webhook: "webhook";
             "word-count": "word-count";
-          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+          }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
             "select-best": "select-best";
             human: "human";
             "max-score": "max-score";
@@ -12639,6 +13244,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -12699,13 +13310,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -12916,7 +13533,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -13033,6 +13655,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
@@ -13093,13 +13721,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           "similar:euclidean": "similar:euclidean";
           "starts-with": "starts-with";
           "tool-call-f1": "tool-call-f1";
+          "skill-used": "skill-used";
+          "trajectory:goal-success": "trajectory:goal-success";
+          "trajectory:tool-args-match": "trajectory:tool-args-match";
+          "trajectory:step-count": "trajectory:step-count";
+          "trajectory:tool-sequence": "trajectory:tool-sequence";
+          "trajectory:tool-used": "trajectory:tool-used";
           "trace-error-spans": "trace-error-spans";
           "trace-span-count": "trace-span-count";
           "trace-span-duration": "trace-span-duration";
           "search-rubric": "search-rubric";
           webhook: "webhook";
           "word-count": "word-count";
-        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+        }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
           "select-best": "select-best";
           human: "human";
           "max-score": "max-score";
@@ -13179,6 +13813,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
@@ -13239,13 +13879,19 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         "similar:euclidean": "similar:euclidean";
         "starts-with": "starts-with";
         "tool-call-f1": "tool-call-f1";
+        "skill-used": "skill-used";
+        "trajectory:goal-success": "trajectory:goal-success";
+        "trajectory:tool-args-match": "trajectory:tool-args-match";
+        "trajectory:step-count": "trajectory:step-count";
+        "trajectory:tool-sequence": "trajectory:tool-sequence";
+        "trajectory:tool-used": "trajectory:tool-used";
         "trace-error-spans": "trace-error-spans";
         "trace-span-count": "trace-span-count";
         "trace-span-duration": "trace-span-duration";
         "search-rubric": "search-rubric";
         webhook: "webhook";
         "word-count": "word-count";
-      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
+      }>, z.ZodTransform<"not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count", "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count">>, z.ZodEnum<{
         "select-best": "select-best";
         human: "human";
         "max-score": "max-score";
@@ -13450,7 +14096,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
     OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
     OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+    CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+    OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+    OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
     OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
     PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -13596,7 +14247,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
         OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
         OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+        CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+        OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+        OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
         OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
         PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -13791,7 +14447,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY?: string | undefined;
         OPENAI_BASE_URL?: string | undefined;
         OPENAI_ORGANIZATION?: string | undefined;
+        CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+        CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+        CLAWDBOT_GATEWAY_URL?: string | undefined;
         CODEX_API_KEY?: string | undefined;
+        OPENCLAW_CONFIG_PATH?: string | undefined;
+        OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
         OPENCLAW_GATEWAY_TOKEN?: string | undefined;
         OPENCLAW_GATEWAY_URL?: string | undefined;
         PALM_API_HOST?: string | undefined;
@@ -14043,7 +14704,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -14170,7 +14836,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -14306,7 +14977,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -14433,7 +15109,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY: z.ZodOptional<z.ZodString>;
       OPENAI_BASE_URL: z.ZodOptional<z.ZodString>;
       OPENAI_ORGANIZATION: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
+      CLAWDBOT_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       CODEX_API_KEY: z.ZodOptional<z.ZodString>;
+      OPENCLAW_CONFIG_PATH: z.ZodOptional<z.ZodString>;
+      OPENCLAW_GATEWAY_PASSWORD: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_TOKEN: z.ZodOptional<z.ZodString>;
       OPENCLAW_GATEWAY_URL: z.ZodOptional<z.ZodString>;
       PALM_API_HOST: z.ZodOptional<z.ZodString>;
@@ -14585,7 +15266,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY?: string | undefined;
         OPENAI_BASE_URL?: string | undefined;
         OPENAI_ORGANIZATION?: string | undefined;
+        CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+        CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+        CLAWDBOT_GATEWAY_URL?: string | undefined;
         CODEX_API_KEY?: string | undefined;
+        OPENCLAW_CONFIG_PATH?: string | undefined;
+        OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
         OPENCLAW_GATEWAY_TOKEN?: string | undefined;
         OPENCLAW_GATEWAY_URL?: string | undefined;
         PALM_API_HOST?: string | undefined;
@@ -14648,7 +15334,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     prompts?: string[] | undefined;
     providerOutput?: string | Record<string, unknown> | undefined;
     assert?: ({
-      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
       value?: AssertionValue | undefined;
       config?: Record<string, any> | undefined;
       threshold?: number | undefined;
@@ -14664,7 +15350,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     } | {
       type: "assert-set";
       assert: {
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -14836,7 +15522,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           OPENAI_API_KEY?: string | undefined;
           OPENAI_BASE_URL?: string | undefined;
           OPENAI_ORGANIZATION?: string | undefined;
+          CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+          CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+          CLAWDBOT_GATEWAY_URL?: string | undefined;
           CODEX_API_KEY?: string | undefined;
+          OPENCLAW_CONFIG_PATH?: string | undefined;
+          OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
           OPENCLAW_GATEWAY_TOKEN?: string | undefined;
           OPENCLAW_GATEWAY_URL?: string | undefined;
           PALM_API_HOST?: string | undefined;
@@ -14899,7 +15590,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       prompts?: string[] | undefined;
       providerOutput?: string | Record<string, unknown> | undefined;
       assert?: ({
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -14915,7 +15606,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       } | {
         type: "assert-set";
         assert: {
-          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
           value?: AssertionValue | undefined;
           config?: Record<string, any> | undefined;
           threshold?: number | undefined;
@@ -15086,7 +15777,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           OPENAI_API_KEY?: string | undefined;
           OPENAI_BASE_URL?: string | undefined;
           OPENAI_ORGANIZATION?: string | undefined;
+          CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+          CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+          CLAWDBOT_GATEWAY_URL?: string | undefined;
           CODEX_API_KEY?: string | undefined;
+          OPENCLAW_CONFIG_PATH?: string | undefined;
+          OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
           OPENCLAW_GATEWAY_TOKEN?: string | undefined;
           OPENCLAW_GATEWAY_URL?: string | undefined;
           PALM_API_HOST?: string | undefined;
@@ -15149,7 +15845,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       prompts?: string[] | undefined;
       providerOutput?: string | Record<string, unknown> | undefined;
       assert?: ({
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -15165,7 +15861,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       } | {
         type: "assert-set";
         assert: {
-          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
           value?: AssertionValue | undefined;
           config?: Record<string, any> | undefined;
           threshold?: number | undefined;
@@ -15339,7 +16035,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY?: string | undefined;
         OPENAI_BASE_URL?: string | undefined;
         OPENAI_ORGANIZATION?: string | undefined;
+        CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+        CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+        CLAWDBOT_GATEWAY_URL?: string | undefined;
         CODEX_API_KEY?: string | undefined;
+        OPENCLAW_CONFIG_PATH?: string | undefined;
+        OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
         OPENCLAW_GATEWAY_TOKEN?: string | undefined;
         OPENCLAW_GATEWAY_URL?: string | undefined;
         PALM_API_HOST?: string | undefined;
@@ -15400,7 +16101,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     } | undefined;
     providerOutput?: string | Record<string, unknown> | undefined;
     assert?: ({
-      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
       value?: AssertionValue | undefined;
       config?: Record<string, any> | undefined;
       threshold?: number | undefined;
@@ -15416,7 +16117,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     } | {
       type: "assert-set";
       assert: {
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -15583,7 +16284,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     OPENAI_API_KEY?: string | undefined;
     OPENAI_BASE_URL?: string | undefined;
     OPENAI_ORGANIZATION?: string | undefined;
+    CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+    CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+    CLAWDBOT_GATEWAY_URL?: string | undefined;
     CODEX_API_KEY?: string | undefined;
+    OPENCLAW_CONFIG_PATH?: string | undefined;
+    OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
     OPENCLAW_GATEWAY_TOKEN?: string | undefined;
     OPENCLAW_GATEWAY_URL?: string | undefined;
     PALM_API_HOST?: string | undefined;
@@ -15801,7 +16507,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -15928,7 +16639,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -16060,7 +16776,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -16187,7 +16908,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -16339,7 +17065,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY?: string | undefined;
         OPENAI_BASE_URL?: string | undefined;
         OPENAI_ORGANIZATION?: string | undefined;
+        CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+        CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+        CLAWDBOT_GATEWAY_URL?: string | undefined;
         CODEX_API_KEY?: string | undefined;
+        OPENCLAW_CONFIG_PATH?: string | undefined;
+        OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
         OPENCLAW_GATEWAY_TOKEN?: string | undefined;
         OPENCLAW_GATEWAY_URL?: string | undefined;
         PALM_API_HOST?: string | undefined;
@@ -16402,7 +17133,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     prompts?: string[] | undefined;
     providerOutput?: string | Record<string, unknown> | undefined;
     assert?: ({
-      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
       value?: AssertionValue | undefined;
       config?: Record<string, any> | undefined;
       threshold?: number | undefined;
@@ -16418,7 +17149,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     } | {
       type: "assert-set";
       assert: {
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -16590,7 +17321,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           OPENAI_API_KEY?: string | undefined;
           OPENAI_BASE_URL?: string | undefined;
           OPENAI_ORGANIZATION?: string | undefined;
+          CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+          CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+          CLAWDBOT_GATEWAY_URL?: string | undefined;
           CODEX_API_KEY?: string | undefined;
+          OPENCLAW_CONFIG_PATH?: string | undefined;
+          OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
           OPENCLAW_GATEWAY_TOKEN?: string | undefined;
           OPENCLAW_GATEWAY_URL?: string | undefined;
           PALM_API_HOST?: string | undefined;
@@ -16653,7 +17389,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       prompts?: string[] | undefined;
       providerOutput?: string | Record<string, unknown> | undefined;
       assert?: ({
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -16669,7 +17405,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       } | {
         type: "assert-set";
         assert: {
-          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
           value?: AssertionValue | undefined;
           config?: Record<string, any> | undefined;
           threshold?: number | undefined;
@@ -16840,7 +17576,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
           OPENAI_API_KEY?: string | undefined;
           OPENAI_BASE_URL?: string | undefined;
           OPENAI_ORGANIZATION?: string | undefined;
+          CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+          CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+          CLAWDBOT_GATEWAY_URL?: string | undefined;
           CODEX_API_KEY?: string | undefined;
+          OPENCLAW_CONFIG_PATH?: string | undefined;
+          OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
           OPENCLAW_GATEWAY_TOKEN?: string | undefined;
           OPENCLAW_GATEWAY_URL?: string | undefined;
           PALM_API_HOST?: string | undefined;
@@ -16903,7 +17644,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       prompts?: string[] | undefined;
       providerOutput?: string | Record<string, unknown> | undefined;
       assert?: ({
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -16919,7 +17660,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       } | {
         type: "assert-set";
         assert: {
-          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+          type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
           value?: AssertionValue | undefined;
           config?: Record<string, any> | undefined;
           threshold?: number | undefined;
@@ -17093,7 +17834,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
         OPENAI_API_KEY?: string | undefined;
         OPENAI_BASE_URL?: string | undefined;
         OPENAI_ORGANIZATION?: string | undefined;
+        CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+        CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+        CLAWDBOT_GATEWAY_URL?: string | undefined;
         CODEX_API_KEY?: string | undefined;
+        OPENCLAW_CONFIG_PATH?: string | undefined;
+        OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
         OPENCLAW_GATEWAY_TOKEN?: string | undefined;
         OPENCLAW_GATEWAY_URL?: string | undefined;
         PALM_API_HOST?: string | undefined;
@@ -17154,7 +17900,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     } | undefined;
     providerOutput?: string | Record<string, unknown> | undefined;
     assert?: ({
-      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+      type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
       value?: AssertionValue | undefined;
       config?: Record<string, any> | undefined;
       threshold?: number | undefined;
@@ -17170,7 +17916,7 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     } | {
       type: "assert-set";
       assert: {
-        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
+        type: `promptfoo:redteam:${string}` | "answer-relevance" | "bleu" | "classifier" | "contains" | "contains-all" | "contains-any" | "contains-html" | "contains-json" | "contains-sql" | "contains-xml" | "context-faithfulness" | "context-recall" | "context-relevance" | "conversation-relevance" | "cost" | "equals" | "factuality" | "finish-reason" | "g-eval" | "gleu" | "guardrails" | "icontains" | "icontains-all" | "icontains-any" | "is-html" | "is-json" | "is-refusal" | "is-sql" | "is-valid-function-call" | "is-valid-openai-function-call" | "is-valid-openai-tools-call" | "is-xml" | "javascript" | "latency" | "levenshtein" | "llm-rubric" | "pi" | "meteor" | "model-graded-closedqa" | "model-graded-factuality" | "moderation" | "perplexity" | "perplexity-score" | "python" | "regex" | "rouge-n" | "ruby" | "similar" | "similar:cosine" | "similar:dot" | "similar:euclidean" | "starts-with" | "tool-call-f1" | "skill-used" | "trajectory:goal-success" | "trajectory:tool-args-match" | "trajectory:step-count" | "trajectory:tool-sequence" | "trajectory:tool-used" | "trace-error-spans" | "trace-span-count" | "trace-span-duration" | "search-rubric" | "webhook" | "word-count" | "not-answer-relevance" | "not-bleu" | "not-classifier" | "not-contains" | "not-contains-all" | "not-contains-any" | "not-contains-html" | "not-contains-json" | "not-contains-sql" | "not-contains-xml" | "not-context-faithfulness" | "not-context-recall" | "not-context-relevance" | "not-conversation-relevance" | "not-cost" | "not-equals" | "not-factuality" | "not-finish-reason" | "not-g-eval" | "not-gleu" | "not-guardrails" | "not-icontains" | "not-icontains-all" | "not-icontains-any" | "not-is-html" | "not-is-json" | "not-is-refusal" | "not-is-sql" | "not-is-valid-function-call" | "not-is-valid-openai-function-call" | "not-is-valid-openai-tools-call" | "not-is-xml" | "not-javascript" | "not-latency" | "not-levenshtein" | "not-llm-rubric" | "not-pi" | "not-meteor" | "not-model-graded-closedqa" | "not-model-graded-factuality" | "not-moderation" | "not-perplexity" | "not-perplexity-score" | "not-python" | "not-regex" | "not-rouge-n" | "not-ruby" | "not-similar" | "not-similar:cosine" | "not-similar:dot" | "not-similar:euclidean" | "not-starts-with" | "not-tool-call-f1" | "not-skill-used" | "not-trajectory:goal-success" | "not-trajectory:tool-args-match" | "not-trajectory:step-count" | "not-trajectory:tool-sequence" | "not-trajectory:tool-used" | "not-trace-error-spans" | "not-trace-span-count" | "not-trace-span-duration" | "not-search-rubric" | "not-webhook" | "not-word-count" | "select-best" | "human" | "max-score";
         value?: AssertionValue | undefined;
         config?: Record<string, any> | undefined;
         threshold?: number | undefined;
@@ -17337,7 +18083,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
     OPENAI_API_KEY?: string | undefined;
     OPENAI_BASE_URL?: string | undefined;
     OPENAI_ORGANIZATION?: string | undefined;
+    CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+    CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+    CLAWDBOT_GATEWAY_URL?: string | undefined;
     CODEX_API_KEY?: string | undefined;
+    OPENCLAW_CONFIG_PATH?: string | undefined;
+    OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
     OPENCLAW_GATEWAY_TOKEN?: string | undefined;
     OPENCLAW_GATEWAY_URL?: string | undefined;
     PALM_API_HOST?: string | undefined;
@@ -17555,7 +18306,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -17682,7 +18438,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -17814,7 +18575,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -17941,7 +18707,12 @@ declare const UnifiedConfigSchema: z.ZodPipe<z.ZodObject<{
       OPENAI_API_KEY?: string | undefined;
       OPENAI_BASE_URL?: string | undefined;
       OPENAI_ORGANIZATION?: string | undefined;
+      CLAWDBOT_GATEWAY_PASSWORD?: string | undefined;
+      CLAWDBOT_GATEWAY_TOKEN?: string | undefined;
+      CLAWDBOT_GATEWAY_URL?: string | undefined;
       CODEX_API_KEY?: string | undefined;
+      OPENCLAW_CONFIG_PATH?: string | undefined;
+      OPENCLAW_GATEWAY_PASSWORD?: string | undefined;
       OPENCLAW_GATEWAY_TOKEN?: string | undefined;
       OPENCLAW_GATEWAY_URL?: string | undefined;
       PALM_API_HOST?: string | undefined;
@@ -18157,7 +18928,8 @@ declare function runAssertion({
   vars,
   latencyMs,
   providerResponse,
-  traceId
+  traceId,
+  traceData
 }: {
   prompt?: string;
   provider?: ApiProvider;
@@ -18168,6 +18940,7 @@ declare function runAssertion({
   latencyMs?: number;
   assertIndex?: number;
   traceId?: string;
+  traceData?: TraceData | null;
 }): Promise<GradingResult>;
 declare function runAssertions({
   assertScoringFunction,
@@ -18833,5 +19606,5 @@ declare const _default$1: {
   };
 };
 //#endregion
-export { type AfterAllExtensionHookContext, type AfterEachExtensionHookContext, AgentCancelledPayload, AgentErrorPayload, ApiClassificationProvider, ApiEmbeddingProvider, ApiModerationProvider, ApiProvider, ApiSimilarityProvider, Assertion, AssertionOrSet, AssertionOrSetSchema, AssertionParams, AssertionSchema, AssertionSet, AssertionSetSchema, AssertionType, AssertionTypeSchema, AssertionValue, AssertionValueFunction, AssertionValueFunctionContext, AssertionValueFunctionResult, AtomicTestCase, AtomicTestCaseSchema, AudioGradingConfig, BaseAssertionTypes, BaseAssertionTypesSchema, BaseRedteamMetadata, BaseTokenUsage, BaseTokenUsageSchema, type BeforeAllExtensionHookContext, type BeforeEachExtensionHookContext, CallApiContextParams, CallApiFunction, CallApiOptionsParams, ChatMessage, CommandLineOptions, CommandLineOptionsSchema, CompletedPrompt, CompletedPromptSchema, CompletionTokenDetails, CompletionTokenDetailsSchema, ConversationMessage, ConversationMessageSchema, CsvRow, DefaultProviders, DerivedMetric, DerivedMetricSchema, type EnvOverrides, EvalConversations, EvalRegisters, EvalResultsFilterMode, EvalSummary, EvalTableDTO, EvalWithMetadata, EvaluateOptions, EvaluateOptionsSchema, EvaluateResult, EvaluateStats, EvaluateSummaryV2, EvaluateSummaryV3, EvaluateTable, EvaluateTableOutput, EvaluateTableRow, EvaluateTestSuite, EvaluateTestSuiteWithEvaluateOptions, type ExtensionHookContextMap, FailedPluginInfo, FilePath, GradingConfig, GradingConfigSchema, GradingResult, GuardrailResponse, ImageOutput, InferSchema, type Inputs, InputsSchema, Intent, Job, LoadApiProviderContext, ModerationFlag, Modifier, NotPrefixedAssertionTypesSchema, NunjucksFilterMap, OutputConfig, OutputConfigSchema, OutputFile, OutputFileExtension, OutputMetadata, PartialGenerationError, PluginActionParams, PluginConfig, PluginConfigSchema, PoliciesById, Policy, PolicyObject, PolicyObjectSchema, Prompt, PromptConfig, PromptContent, PromptFunction, PromptFunctionContext, PromptFunctionResult, PromptMetrics, PromptWithMetadata, ProviderClassificationResponse, ProviderEmbeddingResponse, ProviderFunction, ProviderId, ProviderLabel, ProviderModerationResponse, ProviderOptions, ProviderOptionsMap, ProviderResponse, ProviderSimilarityResponse, ProviderTestResponse, ProviderType, ProviderTypeMap, ProvidersSchema, RateLimitRegistryRef, RedteamAssertionTypes, RedteamCliGenerateOptions, RedteamContext, RedteamFileConfig, RedteamGenerateOptions, RedteamHistoryEntry, RedteamMediaData, RedteamObjectConfig, RedteamPlugin, RedteamPluginObject, RedteamRunOptions, RedteamStrategy, RedteamStrategyObject, ResultFailureReason, ResultLightweight, ResultLightweightWithLabel, ResultSuggestion, ResultsFile, RunEvalOptions, SavedRedteamConfig, Scenario, ScenarioSchema, ScoringFunction, ServerPromptWithMetadata, SharedResults, SpecialAssertionTypes, SpecialAssertionTypesSchema, StrategyConfig, StrategyConfigSchema, SynthesizeOptions, TestCase, TestCaseSchema, TestCaseWithPlugin, TestCaseWithVarsFile, TestCaseWithVarsFileSchema, TestCasesWithMetadata, TestCasesWithMetadataPrompt, TestCasesWithMetadataPromptSchema, TestCasesWithMetadataSchema, TestGeneratorConfig, TestGeneratorConfigSchema, TestSuite, TestSuiteConfig, TestSuiteConfigSchema, TestSuiteSchema, type TokenUsage, TraceData, TraceSpan, TracingConfig, UnifiedConfig, UnifiedConfigSchema, VarMapping, type VarValue, Vars, VarsSchema, ZodLikeSchema, _default as assertions, cache_d_exports as cache, _default$1 as default, evaluate, generateTable, guardrails, isApiProvider, isGradingResult, isProviderOptions, isResultFailureReason, loadApiProvider, redteam };
+export { type AfterAllExtensionHookContext, type AfterEachExtensionHookContext, AgentCancelledPayload, AgentErrorPayload, ApiClassificationProvider, ApiEmbeddingProvider, ApiModerationProvider, ApiProvider, ApiSimilarityProvider, Assertion, AssertionOrSet, AssertionOrSetSchema, AssertionParams, AssertionSchema, AssertionSet, AssertionSetSchema, AssertionType, AssertionTypeSchema, AssertionValue, AssertionValueFunction, AssertionValueFunctionContext, AssertionValueFunctionResult, AtomicTestCase, AtomicTestCaseSchema, AudioGradingConfig, BaseAssertionTypes, BaseAssertionTypesSchema, BaseRedteamMetadata, BaseTokenUsage, BaseTokenUsageSchema, type BeforeAllExtensionHookContext, type BeforeEachExtensionHookContext, CallApiContextParams, CallApiFunction, CallApiOptionsParams, ChatMessage, CommandLineOptions, CommandLineOptionsSchema, CompletedPrompt, CompletedPromptSchema, CompletionTokenDetails, CompletionTokenDetailsSchema, ConversationMessage, ConversationMessageSchema, CsvRow, DefaultProviders, DerivedMetric, DerivedMetricSchema, type EnvOverrides, EvalConversations, EvalRegisters, EvalResultsFilterMode, EvalSummary, EvalTableDTO, EvalWithMetadata, EvaluateOptions, EvaluateOptionsSchema, EvaluateResult, EvaluateStats, EvaluateSummaryV2, EvaluateSummaryV3, EvaluateTable, EvaluateTableOutput, EvaluateTableRow, EvaluateTestSuite, EvaluateTestSuiteWithEvaluateOptions, type ExtensionHookContextMap, FailedPluginInfo, FilePath, GradingConfig, GradingConfigSchema, GradingResult, GuardrailResponse, ImageOutput, InferSchema, type Inputs, InputsSchema, Intent, Job, LoadApiProviderContext, ModerationFlag, Modifier, NotPrefixedAssertionTypesSchema, NunjucksFilterMap, OutputConfig, OutputConfigSchema, OutputFile, OutputFileExtension, OutputMetadata, PartialGenerationError, PluginActionParams, PluginConfig, PluginConfigSchema, PoliciesById, Policy, PolicyObject, PolicyObjectSchema, Prompt, PromptConfig, PromptContent, PromptFunction, PromptFunctionContext, PromptFunctionResult, PromptMetrics, PromptWithMetadata, ProviderClassificationResponse, ProviderEmbeddingResponse, ProviderFunction, ProviderId, ProviderLabel, ProviderModerationResponse, ProviderOptions, ProviderOptionsMap, ProviderResponse, ProviderSimilarityResponse, ProviderTestResponse, ProviderType, ProviderTypeMap, ProvidersSchema, RateLimitRegistryRef, RedteamAssertionTypes, RedteamCliGenerateOptions, RedteamContext, RedteamFileConfig, RedteamGenerateOptions, RedteamHistoryEntry, RedteamMediaData, RedteamObjectConfig, RedteamPlugin, RedteamPluginObject, RedteamRunOptions, RedteamStrategy, RedteamStrategyObject, ResultFailureReason, ResultLightweight, ResultLightweightWithLabel, ResultSuggestion, ResultsFile, RunEvalOptions, SavedRedteamConfig, Scenario, ScenarioSchema, ScoringFunction, ServerPromptWithMetadata, SharedResults, SkillCallEntry, SpecialAssertionTypes, SpecialAssertionTypesSchema, StrategyConfig, StrategyConfigSchema, SynthesizeOptions, TestCase, TestCaseSchema, TestCaseWithPlugin, TestCaseWithVarsFile, TestCaseWithVarsFileSchema, TestCasesWithMetadata, TestCasesWithMetadataPrompt, TestCasesWithMetadataPromptSchema, TestCasesWithMetadataSchema, TestGeneratorConfig, TestGeneratorConfigSchema, TestSuite, TestSuiteConfig, TestSuiteConfigSchema, TestSuiteSchema, type TokenUsage, TraceData, TraceSpan, TracingConfig, UnifiedConfig, UnifiedConfigSchema, VarMapping, type VarValue, Vars, VarsSchema, ZodLikeSchema, _default as assertions, cache_d_exports as cache, _default$1 as default, evaluate, generateTable, guardrails, isApiProvider, isGradingResult, isProviderOptions, isResultFailureReason, loadApiProvider, redteam };
 //# sourceMappingURL=index.d.cts.map