@agentv/core 3.10.3 → 3.11.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/{chunk-VCFYWLFV.js → chunk-AVTN5AB7.js} +17 -12
- package/dist/chunk-AVTN5AB7.js.map +1 -0
- package/dist/evaluation/validation/index.cjs.map +1 -1
- package/dist/evaluation/validation/index.js +1 -1
- package/dist/index.cjs +173 -135
- package/dist/index.cjs.map +1 -1
- package/dist/index.d.cts +11 -5
- package/dist/index.d.ts +11 -5
- package/dist/index.js +158 -125
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
- package/dist/chunk-VCFYWLFV.js.map +0 -1
package/dist/index.d.cts
CHANGED
|
@@ -1018,7 +1018,6 @@ interface EvalTest {
|
|
|
1018
1018
|
readonly conversation_id?: string;
|
|
1019
1019
|
readonly question: string;
|
|
1020
1020
|
readonly input: readonly TestMessage[];
|
|
1021
|
-
readonly input_segments: readonly JsonObject[];
|
|
1022
1021
|
readonly expected_output: readonly JsonObject[];
|
|
1023
1022
|
readonly reference_answer?: string;
|
|
1024
1023
|
readonly file_paths: readonly string[];
|
|
@@ -1254,6 +1253,10 @@ type ExecutionDefaults = {
|
|
|
1254
1253
|
readonly trace_file?: string;
|
|
1255
1254
|
readonly keep_workspaces?: boolean;
|
|
1256
1255
|
readonly otel_file?: string;
|
|
1256
|
+
readonly export_otel?: boolean;
|
|
1257
|
+
readonly otel_backend?: string;
|
|
1258
|
+
readonly otel_capture_content?: boolean;
|
|
1259
|
+
readonly otel_group_turns?: boolean;
|
|
1257
1260
|
readonly pool_workspaces?: boolean;
|
|
1258
1261
|
readonly pool_slots?: number;
|
|
1259
1262
|
};
|
|
@@ -1689,7 +1692,7 @@ interface CopilotSdkResolvedConfig {
|
|
|
1689
1692
|
}
|
|
1690
1693
|
interface PiCodingAgentResolvedConfig {
|
|
1691
1694
|
readonly executable: string;
|
|
1692
|
-
readonly
|
|
1695
|
+
readonly subprovider?: string;
|
|
1693
1696
|
readonly model?: string;
|
|
1694
1697
|
readonly apiKey?: string;
|
|
1695
1698
|
readonly tools?: string;
|
|
@@ -1703,7 +1706,7 @@ interface PiCodingAgentResolvedConfig {
|
|
|
1703
1706
|
readonly systemPrompt?: string;
|
|
1704
1707
|
}
|
|
1705
1708
|
interface PiAgentSdkResolvedConfig {
|
|
1706
|
-
readonly
|
|
1709
|
+
readonly subprovider?: string;
|
|
1707
1710
|
readonly model?: string;
|
|
1708
1711
|
readonly apiKey?: string;
|
|
1709
1712
|
readonly timeoutMs?: number;
|
|
@@ -2055,7 +2058,7 @@ interface EvaluationScore {
|
|
|
2055
2058
|
readonly expectedAspectCount: number;
|
|
2056
2059
|
readonly evaluatorRawRequest?: JsonObject;
|
|
2057
2060
|
readonly scores?: readonly ChildEvaluatorResult[];
|
|
2058
|
-
/** Optional structured details from
|
|
2061
|
+
/** Optional structured details from evaluators (e.g., TP/TN/FP/FN counts, alignments, per-turn scores). */
|
|
2059
2062
|
readonly details?: JsonObject;
|
|
2060
2063
|
/** Token usage from LLM calls made by this evaluator (optional). */
|
|
2061
2064
|
readonly tokenUsage?: TokenUsage;
|
|
@@ -2071,7 +2074,7 @@ interface ChildEvaluatorResult {
|
|
|
2071
2074
|
readonly assertions: readonly AssertionEntry[];
|
|
2072
2075
|
readonly evaluatorRawRequest?: JsonObject;
|
|
2073
2076
|
readonly scores?: readonly ChildEvaluatorResult[];
|
|
2074
|
-
/** Optional structured details from
|
|
2077
|
+
/** Optional structured details from evaluators (e.g., TP/TN/FP/FN counts, alignments, per-turn scores). */
|
|
2075
2078
|
readonly details?: JsonObject;
|
|
2076
2079
|
/** Token usage from LLM calls made by this evaluator (optional). */
|
|
2077
2080
|
readonly tokenUsage?: TokenUsage;
|
|
@@ -2258,6 +2261,7 @@ declare const freeformEvaluationSchema: z.ZodObject<{
|
|
|
2258
2261
|
passed: boolean;
|
|
2259
2262
|
evidence?: string | undefined;
|
|
2260
2263
|
}>, "many">>;
|
|
2264
|
+
details: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
2261
2265
|
}, "strip", z.ZodTypeAny, {
|
|
2262
2266
|
score: number;
|
|
2263
2267
|
assertions?: {
|
|
@@ -2265,6 +2269,7 @@ declare const freeformEvaluationSchema: z.ZodObject<{
|
|
|
2265
2269
|
passed: boolean;
|
|
2266
2270
|
evidence?: string | undefined;
|
|
2267
2271
|
}[] | undefined;
|
|
2272
|
+
details?: Record<string, unknown> | undefined;
|
|
2268
2273
|
}, {
|
|
2269
2274
|
score: number;
|
|
2270
2275
|
assertions?: {
|
|
@@ -2272,6 +2277,7 @@ declare const freeformEvaluationSchema: z.ZodObject<{
|
|
|
2272
2277
|
passed: boolean;
|
|
2273
2278
|
evidence?: string | undefined;
|
|
2274
2279
|
}[] | undefined;
|
|
2280
|
+
details?: Record<string, unknown> | undefined;
|
|
2275
2281
|
}>;
|
|
2276
2282
|
declare const rubricEvaluationSchema: z.ZodObject<{
|
|
2277
2283
|
checks: z.ZodArray<z.ZodObject<{
|
package/dist/index.d.ts
CHANGED
|
@@ -1018,7 +1018,6 @@ interface EvalTest {
|
|
|
1018
1018
|
readonly conversation_id?: string;
|
|
1019
1019
|
readonly question: string;
|
|
1020
1020
|
readonly input: readonly TestMessage[];
|
|
1021
|
-
readonly input_segments: readonly JsonObject[];
|
|
1022
1021
|
readonly expected_output: readonly JsonObject[];
|
|
1023
1022
|
readonly reference_answer?: string;
|
|
1024
1023
|
readonly file_paths: readonly string[];
|
|
@@ -1254,6 +1253,10 @@ type ExecutionDefaults = {
|
|
|
1254
1253
|
readonly trace_file?: string;
|
|
1255
1254
|
readonly keep_workspaces?: boolean;
|
|
1256
1255
|
readonly otel_file?: string;
|
|
1256
|
+
readonly export_otel?: boolean;
|
|
1257
|
+
readonly otel_backend?: string;
|
|
1258
|
+
readonly otel_capture_content?: boolean;
|
|
1259
|
+
readonly otel_group_turns?: boolean;
|
|
1257
1260
|
readonly pool_workspaces?: boolean;
|
|
1258
1261
|
readonly pool_slots?: number;
|
|
1259
1262
|
};
|
|
@@ -1689,7 +1692,7 @@ interface CopilotSdkResolvedConfig {
|
|
|
1689
1692
|
}
|
|
1690
1693
|
interface PiCodingAgentResolvedConfig {
|
|
1691
1694
|
readonly executable: string;
|
|
1692
|
-
readonly
|
|
1695
|
+
readonly subprovider?: string;
|
|
1693
1696
|
readonly model?: string;
|
|
1694
1697
|
readonly apiKey?: string;
|
|
1695
1698
|
readonly tools?: string;
|
|
@@ -1703,7 +1706,7 @@ interface PiCodingAgentResolvedConfig {
|
|
|
1703
1706
|
readonly systemPrompt?: string;
|
|
1704
1707
|
}
|
|
1705
1708
|
interface PiAgentSdkResolvedConfig {
|
|
1706
|
-
readonly
|
|
1709
|
+
readonly subprovider?: string;
|
|
1707
1710
|
readonly model?: string;
|
|
1708
1711
|
readonly apiKey?: string;
|
|
1709
1712
|
readonly timeoutMs?: number;
|
|
@@ -2055,7 +2058,7 @@ interface EvaluationScore {
|
|
|
2055
2058
|
readonly expectedAspectCount: number;
|
|
2056
2059
|
readonly evaluatorRawRequest?: JsonObject;
|
|
2057
2060
|
readonly scores?: readonly ChildEvaluatorResult[];
|
|
2058
|
-
/** Optional structured details from
|
|
2061
|
+
/** Optional structured details from evaluators (e.g., TP/TN/FP/FN counts, alignments, per-turn scores). */
|
|
2059
2062
|
readonly details?: JsonObject;
|
|
2060
2063
|
/** Token usage from LLM calls made by this evaluator (optional). */
|
|
2061
2064
|
readonly tokenUsage?: TokenUsage;
|
|
@@ -2071,7 +2074,7 @@ interface ChildEvaluatorResult {
|
|
|
2071
2074
|
readonly assertions: readonly AssertionEntry[];
|
|
2072
2075
|
readonly evaluatorRawRequest?: JsonObject;
|
|
2073
2076
|
readonly scores?: readonly ChildEvaluatorResult[];
|
|
2074
|
-
/** Optional structured details from
|
|
2077
|
+
/** Optional structured details from evaluators (e.g., TP/TN/FP/FN counts, alignments, per-turn scores). */
|
|
2075
2078
|
readonly details?: JsonObject;
|
|
2076
2079
|
/** Token usage from LLM calls made by this evaluator (optional). */
|
|
2077
2080
|
readonly tokenUsage?: TokenUsage;
|
|
@@ -2258,6 +2261,7 @@ declare const freeformEvaluationSchema: z.ZodObject<{
|
|
|
2258
2261
|
passed: boolean;
|
|
2259
2262
|
evidence?: string | undefined;
|
|
2260
2263
|
}>, "many">>;
|
|
2264
|
+
details: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
2261
2265
|
}, "strip", z.ZodTypeAny, {
|
|
2262
2266
|
score: number;
|
|
2263
2267
|
assertions?: {
|
|
@@ -2265,6 +2269,7 @@ declare const freeformEvaluationSchema: z.ZodObject<{
|
|
|
2265
2269
|
passed: boolean;
|
|
2266
2270
|
evidence?: string | undefined;
|
|
2267
2271
|
}[] | undefined;
|
|
2272
|
+
details?: Record<string, unknown> | undefined;
|
|
2268
2273
|
}, {
|
|
2269
2274
|
score: number;
|
|
2270
2275
|
assertions?: {
|
|
@@ -2272,6 +2277,7 @@ declare const freeformEvaluationSchema: z.ZodObject<{
|
|
|
2272
2277
|
passed: boolean;
|
|
2273
2278
|
evidence?: string | undefined;
|
|
2274
2279
|
}[] | undefined;
|
|
2280
|
+
details?: Record<string, unknown> | undefined;
|
|
2275
2281
|
}>;
|
|
2276
2282
|
declare const rubricEvaluationSchema: z.ZodObject<{
|
|
2277
2283
|
checks: z.ZodArray<z.ZodObject<{
|