@agentv/core 4.6.0 → 4.7.0

package/dist/index.cjs

@@ -31,12 +31,9 @@ var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__ge
 var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
 
 // ../../node_modules/.bun/tsup@8.3.5+19811ebab77a7b1c/node_modules/tsup/assets/cjs_shims.js
-var getImportMetaUrl, importMetaUrl;
 var init_cjs_shims = __esm({
   "../../node_modules/.bun/tsup@8.3.5+19811ebab77a7b1c/node_modules/tsup/assets/cjs_shims.js"() {
     "use strict";
-    getImportMetaUrl = () => typeof document === "undefined" ? new URL(`file:${__filename}`).href : document.currentScript && document.currentScript.src || new URL("main.js", document.baseURI).href;
-    importMetaUrl = /* @__PURE__ */ getImportMetaUrl();
   }
 });
 
@@ -1435,6 +1432,7 @@ __export(index_exports, {
   DEFAULT_EVALUATOR_TEMPLATE: () => DEFAULT_EVALUATOR_TEMPLATE,
   DEFAULT_EVAL_PATTERNS: () => DEFAULT_EVAL_PATTERNS,
   DEFAULT_EXPLORATION_TOOLS: () => DEFAULT_EXPLORATION_TOOLS,
+  DEFAULT_THRESHOLD: () => DEFAULT_THRESHOLD,
   DeterministicAssertionEvaluator: () => DeterministicAssertionEvaluator,
   EvaluatorRegistry: () => EvaluatorRegistry,
   ExecutionMetricsEvaluator: () => ExecutionMetricsEvaluator,
@@ -1456,6 +1454,7 @@ __export(index_exports, {
   TemplateNotFoundError: () => TemplateNotFoundError,
   TokenUsageEvaluator: () => TokenUsageEvaluator,
   ToolTrajectoryEvaluator: () => ToolTrajectoryEvaluator,
+  TranscriptProvider: () => TranscriptProvider,
   WorkspaceCreationError: () => WorkspaceCreationError,
   WorkspacePoolManager: () => WorkspacePoolManager,
   addProject: () => addProject,
@@ -1492,6 +1491,7 @@ __export(index_exports, {
   detectFormat: () => detectFormat,
   discoverAssertions: () => discoverAssertions,
   discoverClaudeSessions: () => discoverClaudeSessions,
+  discoverCodexSessions: () => discoverCodexSessions,
   discoverCopilotSessions: () => discoverCopilotSessions,
   discoverGraders: () => discoverGraders,
   discoverJudges: () => discoverGraders,
@@ -1552,6 +1552,8 @@ __export(index_exports, {
   normalizeLineEndings: () => normalizeLineEndings,
   parseAgentSkillsEvals: () => parseAgentSkillsEvals,
   parseClaudeSession: () => parseClaudeSession,
+  parseCodexSession: () => parseCodexSession,
+  parseCopilotEvents: () => parseCopilotEvents,
   parseJsonFromText: () => parseJsonFromText,
   parseJsonSafe: () => parseJsonSafe,
   readJsonFile: () => readJsonFile,
@@ -1559,8 +1561,10 @@ __export(index_exports, {
   readTestSuiteMetadata: () => readTestSuiteMetadata,
   readTextFile: () => readTextFile,
   readTranscriptFile: () => readTranscriptFile,
+  readTranscriptJsonl: () => readTranscriptJsonl,
   removeProject: () => removeProject,
   resolveAndCreateProvider: () => resolveAndCreateProvider,
+  resolveDelegatedTargetDefinition: () => resolveDelegatedTargetDefinition,
   resolveFileReference: () => resolveFileReference3,
   resolveTargetDefinition: () => resolveTargetDefinition,
   resolveWorkspaceTemplate: () => resolveWorkspaceTemplate,
@@ -1590,6 +1594,7 @@ __export(index_exports, {
   substituteVariables: () => substituteVariables,
   toCamelCaseDeep: () => toCamelCaseDeep,
   toSnakeCaseDeep: () => toSnakeCaseDeep,
+  toTranscriptJsonLine: () => toTranscriptJsonLine,
   tokensPerTool: () => tokensPerTool,
   touchProject: () => touchProject,
   transpileEvalYaml: () => transpileEvalYaml,
@@ -2674,8 +2679,13 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
     const negate = rawEvaluator.negate === true ? true : void 0;
     if (isCustomType) {
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
-      const knownProps2 = /* @__PURE__ */ new Set(["name", "type", "weight", "required", "negate"]);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
+      const knownProps2 = /* @__PURE__ */ new Set(["name", "type", "weight", "required", "min_score", "negate"]);
       const config2 = {};
       for (const [key, value] of Object.entries(rawEvaluator)) {
         if (!knownProps2.has(key) && value !== void 0) {
@@ -2687,6 +2697,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         type: customTypeName,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {},
         ...Object.keys(config2).length > 0 ? { config: config2 } : {}
       });
@@ -2756,7 +2767,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
           );
         }
       }
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       const knownProps2 = /* @__PURE__ */ new Set([
         "name",
         "type",
@@ -2782,6 +2798,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         resolvedCwd,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {},
         ...Object.keys(config2).length > 0 ? { config: config2 } : {},
         ...targetConfig !== void 0 ? { target: targetConfig } : {}
@@ -2910,7 +2927,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         };
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "composite",
@@ -2918,6 +2940,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         aggregator,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3028,7 +3051,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       const config2 = {
         name,
         type: "tool-trajectory",
@@ -3037,6 +3065,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         ...expected ? { expected } : {},
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {},
         ...argsMatch2 !== void 0 ? { argsMatch: argsMatch2 } : {}
       };
@@ -3099,7 +3128,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
       const aggregation = asString(rawEvaluator.aggregation);
       const validAggregation = isValidFieldAggregationType(aggregation) ? aggregation : void 0;
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "field-accuracy",
@@ -3107,6 +3141,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         ...validAggregation ? { aggregation: validAggregation } : {},
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3120,13 +3155,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "latency",
         threshold,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3140,13 +3181,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "cost",
         budget,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3178,13 +3225,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "token-usage",
         ...validLimits,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3230,13 +3283,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "execution-metrics",
         ...validThresholds,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3250,7 +3309,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
       const rawShouldTrigger = rawEvaluator.should_trigger;
       const shouldTrigger = typeof rawShouldTrigger === "boolean" ? rawShouldTrigger : void 0;
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "skill-trigger",
@@ -3258,6 +3322,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         ...shouldTrigger !== void 0 ? { should_trigger: shouldTrigger } : {},
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3269,13 +3334,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "contains",
         value,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3289,13 +3360,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: typeValue,
         value,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3307,13 +3384,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "icontains",
         value,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3327,13 +3410,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: typeValue,
         value,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3345,13 +3434,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: typeValue,
         value,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3364,7 +3459,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
       }
       const flags = asString(rawEvaluator.flags);
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "regex",
@@ -3372,18 +3472,25 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         ...flags !== void 0 ? { flags } : {},
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
     }
     if (typeValue === "is-json") {
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "is-json",
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3395,13 +3502,19 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "equals",
         value,
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3437,7 +3550,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "llm-grader",
@@ -3445,6 +3563,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         ...graderTargetName ? { target: graderTargetName } : {},
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
@@ -3514,7 +3633,12 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         continue;
       }
       const weight2 = validateWeight(rawEvaluator.weight, name, evalId);
-      const required2 = parseRequired(rawEvaluator.required);
+      const { required: required2, min_score: min_score2 } = parseRequiredAndMinScore(
+        rawEvaluator.required,
+        rawEvaluator.min_score,
+        name,
+        evalId
+      );
       evaluators.push({
         name,
         type: "llm-grader",
@@ -3522,12 +3646,18 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
         ...graderTargetName ? { target: graderTargetName } : {},
         ...weight2 !== void 0 ? { weight: weight2 } : {},
         ...required2 !== void 0 ? { required: required2 } : {},
+        ...min_score2 !== void 0 ? { min_score: min_score2 } : {},
         ...negate !== void 0 ? { negate } : {}
       });
       continue;
     }
     const weight = validateWeight(rawEvaluator.weight, name, evalId);
-    const required = parseRequired(rawEvaluator.required);
+    const { required, min_score } = parseRequiredAndMinScore(
+      rawEvaluator.required,
+      rawEvaluator.min_score,
+      name,
+      evalId
+    );
     const knownProps = /* @__PURE__ */ new Set([
       "name",
       "type",
@@ -3538,6 +3668,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
       "weight",
       "config",
       "required",
+      "min_score",
       "negate",
       "max_steps",
       "maxSteps",
@@ -3567,6 +3698,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
       ...graderTargetName ? { target: graderTargetName } : {},
       ...weight !== void 0 ? { weight } : {},
       ...required !== void 0 ? { required } : {},
+      ...min_score !== void 0 ? { min_score } : {},
       ...negate !== void 0 ? { negate } : {},
       ...finalConfig ? { config: finalConfig } : {},
       ...llmMaxSteps !== void 0 ? { max_steps: llmMaxSteps } : {},
@@ -3698,10 +3830,23 @@ ${detailBlock}${ANSI_RESET5}`);
     console.warn(`${ANSI_YELLOW4}Warning: ${message}${ANSI_RESET5}`);
   }
 }
-function parseRequired(value) {
-  if (value === true) return true;
-  if (typeof value === "number" && value > 0 && value <= 1) return value;
-  return void 0;
+function parseRequiredAndMinScore(rawRequired, rawMinScore, evaluatorName, evalId) {
+  const result = {};
+  if (typeof rawMinScore === "number" && rawMinScore > 0 && rawMinScore <= 1) {
+    result.min_score = rawMinScore;
+  }
+  if (rawRequired === true) {
+    result.required = true;
+  } else if (typeof rawRequired === "number" && rawRequired > 0 && rawRequired <= 1) {
+    if (result.min_score === void 0) {
+      result.min_score = rawRequired;
+    }
+    result.required = rawRequired;
+    logWarning2(
+      `Evaluator '${evaluatorName}' in '${evalId}': 'required: ${rawRequired}' is deprecated. Use 'required: true' + 'min_score: ${rawRequired}' instead.`
+    );
+  }
+  return result;
 }
 function validateWeight(rawWeight, evaluatorName, evalId) {
   if (rawWeight === void 0) {
@@ -3744,16 +3889,30 @@ function parseRubricItems(rawRubrics, evaluatorName, evalId) {
     const id = asString(rawRubric.id) ?? `rubric-${index + 1}`;
     const expectedOutcome = asString(rawRubric.outcome) ?? "";
     const weight = typeof rawRubric.weight === "number" ? rawRubric.weight : 1;
+    let minScore;
     let requiredMinScore;
     let required;
-    if (typeof rawRubric.required_min_score === "number") {
-      const minScore = rawRubric.required_min_score;
-      if (!Number.isInteger(minScore) || minScore < 0 || minScore > 10) {
+    if (typeof rawRubric.min_score === "number") {
+      const ms = rawRubric.min_score;
+      if (ms <= 0 || ms > 1) {
         throw new Error(
-          `Invalid required_min_score for rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': must be an integer 0-10 (got ${minScore})`
+          `Invalid min_score for rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': must be in (0, 1] (got ${ms})`
         );
       }
-      requiredMinScore = minScore;
+      minScore = ms;
+      requiredMinScore = Math.round(ms * 10);
+    } else if (typeof rawRubric.required_min_score === "number") {
+      const rms = rawRubric.required_min_score;
+      if (!Number.isInteger(rms) || rms < 0 || rms > 10) {
+        throw new Error(
+          `Invalid required_min_score for rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': must be an integer 0-10 (got ${rms})`
+        );
+      }
+      requiredMinScore = rms;
+      minScore = rms / 10;
+      logWarning2(
+        `Rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': 'required_min_score: ${rms}' is deprecated. Use 'min_score: ${rms / 10}' (0-1 scale) instead.`
+      );
     }
     if (typeof rawRubric.required === "boolean") {
       required = rawRubric.required;
@@ -3773,6 +3932,7 @@ function parseRubricItems(rawRubrics, evaluatorName, evalId) {
         weight,
         ...expectedOutcome.length > 0 ? { outcome: expectedOutcome } : {},
         ...required !== void 0 ? { required } : {},
+        ...minScore !== void 0 ? { min_score: minScore } : {},
         ...requiredMinScore !== void 0 ? { required_min_score: requiredMinScore } : {},
         score_ranges: scoreRanges
       });
@@ -3789,6 +3949,7 @@ function parseRubricItems(rawRubrics, evaluatorName, evalId) {
         weight,
         // Default to required: true if not specified (backward compatibility)
         required: required ?? true,
+        ...minScore !== void 0 ? { min_score: minScore } : {},
         ...requiredMinScore !== void 0 ? { required_min_score: requiredMinScore } : {}
       });
     }
@@ -3917,12 +4078,22 @@ function parseInlineRubrics(rawRubrics) {
       id: asString(rubric.id) ?? `rubric-${index + 1}`,
       weight: typeof rubric.weight === "number" ? rubric.weight : 1
     };
+    let inlineMinScore;
+    let inlineRequiredMinScore;
+    if (typeof rubric.min_score === "number") {
+      inlineMinScore = rubric.min_score;
+      inlineRequiredMinScore = Math.round(inlineMinScore * 10);
+    } else if (typeof rubric.required_min_score === "number") {
+      inlineRequiredMinScore = rubric.required_min_score;
+      inlineMinScore = inlineRequiredMinScore / 10;
+    }
     if (scoreRanges && scoreRanges.length > 0) {
       return {
         ...baseRubric,
         ...expectedOutcome.length > 0 ? { outcome: expectedOutcome } : {},
         ...typeof rubric.required === "boolean" ? { required: rubric.required } : {},
-        ...typeof rubric.required_min_score === "number" ? { required_min_score: rubric.required_min_score } : {},
+        ...inlineMinScore !== void 0 ? { min_score: inlineMinScore } : {},
+        ...inlineRequiredMinScore !== void 0 ? { required_min_score: inlineRequiredMinScore } : {},
         score_ranges: scoreRanges
       };
     }
@@ -3930,7 +4101,8 @@ function parseInlineRubrics(rawRubrics) {
       ...baseRubric,
       outcome: expectedOutcome,
       required: typeof rubric.required === "boolean" ? rubric.required : true,
-      ...typeof rubric.required_min_score === "number" ? { required_min_score: rubric.required_min_score } : {}
+      ...inlineMinScore !== void 0 ? { min_score: inlineMinScore } : {},
+      ...inlineRequiredMinScore !== void 0 ? { required_min_score: inlineRequiredMinScore } : {}
     };
   }).filter((r) => r.outcome && r.outcome.length > 0 || "score_ranges" in r && r.score_ranges);
   if (rubricItems.length === 0) {
@@ -4334,6 +4506,9 @@ function resolveExpectedMessages(raw) {
 var ANSI_YELLOW6 = "\x1B[33m";
 var ANSI_RED2 = "\x1B[31m";
 var ANSI_RESET7 = "\x1B[0m";
+function matchesFilter(id, filter) {
+  return typeof filter === "string" ? import_micromatch.default.isMatch(id, filter) : filter.some((pattern) => import_micromatch.default.isMatch(id, pattern));
+}
 function detectFormat(filePath) {
   const ext = import_node_path7.default.extname(filePath).toLowerCase();
   if (ext === ".jsonl") return "jsonl";
@@ -4401,40 +4576,40 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
   const sidecar = await loadSidecarMetadata(absoluteTestPath, verbose);
   const rawFile = await (0, import_promises7.readFile)(absoluteTestPath, "utf8");
   const rawCases = parseJsonlContent(rawFile, evalFilePath);
-  const fallbackEvalSet = import_node_path7.default.basename(absoluteTestPath, ".jsonl") || "eval";
-  const evalSetName = sidecar.name && sidecar.name.trim().length > 0 ? sidecar.name : fallbackEvalSet;
+  const fallbackSuiteName = import_node_path7.default.basename(absoluteTestPath, ".jsonl") || "eval";
+  const suiteName = sidecar.name && sidecar.name.trim().length > 0 ? sidecar.name : fallbackSuiteName;
   const globalEvaluator = coerceEvaluator(sidecar.evaluator, "sidecar") ?? "llm-grader";
   const globalExecution = sidecar.execution;
   if (verbose) {
     console.log(`
-[JSONL Dataset: ${evalFilePath}]`);
+[JSONL Suite: ${evalFilePath}]`);
     console.log(`  Cases: ${rawCases.length}`);
-    console.log(`  Eval set: ${evalSetName}`);
+    console.log(`  Suite: ${suiteName}`);
     if (sidecar.description) {
       console.log(`  Description: ${sidecar.description}`);
     }
   }
   const results = [];
   for (let lineIndex = 0; lineIndex < rawCases.length; lineIndex++) {
-    const evalcase = rawCases[lineIndex];
+    const testCaseConfig = rawCases[lineIndex];
     const lineNumber = lineIndex + 1;
-    const id = asString4(evalcase.id);
-    if (filterPattern && (!id || !import_micromatch.default.isMatch(id, filterPattern))) {
+    const id = asString4(testCaseConfig.id);
+    if (filterPattern && (!id || !matchesFilter(id, filterPattern))) {
       continue;
     }
-    const conversationId = asString4(evalcase.conversation_id);
-    let outcome = asString4(evalcase.criteria);
-    if (!outcome && evalcase.expected_outcome !== void 0) {
-      outcome = asString4(evalcase.expected_outcome);
+    const conversationId = asString4(testCaseConfig.conversation_id);
+    let outcome = asString4(testCaseConfig.criteria);
+    if (!outcome && testCaseConfig.expected_outcome !== void 0) {
+      outcome = asString4(testCaseConfig.expected_outcome);
       if (outcome) {
         logWarning4(
-          `Test '${asString4(evalcase.id) ?? "unknown"}': 'expected_outcome' is deprecated. Use 'criteria' instead.`
+          `Test '${asString4(testCaseConfig.id) ?? "unknown"}': 'expected_outcome' is deprecated. Use 'criteria' instead.`
         );
       }
     }
-    const rawInputMessages = resolveInputMessages(evalcase);
-    const expectedMessages = resolveExpectedMessages(evalcase) ?? [];
-    const hasEvaluationSpec = !!outcome || expectedMessages.length > 0 || evalcase.assert !== void 0;
+    const rawInputMessages = resolveInputMessages(testCaseConfig);
+    const expectedMessages = resolveExpectedMessages(testCaseConfig) ?? [];
+    const hasEvaluationSpec = !!outcome || expectedMessages.length > 0 || testCaseConfig.assert !== void 0;
     if (!id || !hasEvaluationSpec || !rawInputMessages || rawInputMessages.length === 0) {
       logError2(
         `Skipping incomplete test at line ${lineNumber}: ${id ?? "unknown"}. Missing required fields: id, input, and at least one of criteria/expected_output/assert`
@@ -4471,18 +4646,23 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
       }
     }
     const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const caseExecution = isJsonObject(evalcase.execution) ? evalcase.execution : void 0;
+    const caseExecution = isJsonObject(testCaseConfig.execution) ? testCaseConfig.execution : void 0;
     const mergedExecution = caseExecution ?? globalExecution;
-    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    const testCaseEvaluatorKind = coerceEvaluator(testCaseConfig.evaluator, id) ?? globalEvaluator;
     let evaluators;
     try {
-      evaluators = await parseEvaluators(evalcase, mergedExecution, searchRoots, id ?? "unknown");
+      evaluators = await parseEvaluators(
+        testCaseConfig,
+        mergedExecution,
+        searchRoots,
+        id ?? "unknown"
+      );
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       logError2(`Skipping test '${id}' at line ${lineNumber}: ${message}`);
       continue;
     }
-    const inlineRubrics = evalcase.rubrics;
+    const inlineRubrics = testCaseConfig.rubrics;
     if (inlineRubrics !== void 0 && Array.isArray(inlineRubrics)) {
       const rubricEvaluator = parseInlineRubrics(inlineRubrics);
       if (rubricEvaluator) {
@@ -4493,7 +4673,7 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
     const userFilePaths = collectResolvedInputFilePaths(inputMessages);
     const testCase = {
       id,
-      dataset: evalSetName,
+      suite: suiteName,
       conversation_id: conversationId,
       question,
       input: inputMessages,
@@ -4501,7 +4681,7 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
       reference_answer: referenceAnswer,
       file_paths: userFilePaths,
       criteria: outcome ?? "",
-      evaluator: evalCaseEvaluatorKind,
+      evaluator: testCaseEvaluatorKind,
       assertions: evaluators
     };
     results.push(testCase);
@@ -4686,6 +4866,9 @@ function buildChatPromptFromSegments(options) {
 var ANSI_YELLOW7 = "\x1B[33m";
 var ANSI_RED3 = "\x1B[31m";
 var ANSI_RESET8 = "\x1B[0m";
+function matchesFilter2(id, filter) {
+  return typeof filter === "string" ? import_micromatch2.default.isMatch(id, filter) : filter.some((pattern) => import_micromatch2.default.isMatch(id, pattern));
+}
 function resolveTests(suite) {
   if (suite.tests !== void 0) return suite.tests;
   if (suite.eval_cases !== void 0) {
@@ -4765,18 +4948,18 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
     throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
   const suite = interpolated;
-  const evalSetNameFromSuite = asString5(suite.name)?.trim();
-  const fallbackEvalSet = import_node_path8.default.basename(absoluteTestPath).replace(/\.eval\.ya?ml$/i, "").replace(/\.ya?ml$/i, "") || "eval";
-  const evalSetName = evalSetNameFromSuite && evalSetNameFromSuite.length > 0 ? evalSetNameFromSuite : fallbackEvalSet;
-  const rawTestcases = resolveTests(suite);
+  const suiteNameFromFile = asString5(suite.name)?.trim();
+  const fallbackSuiteName = import_node_path8.default.basename(absoluteTestPath).replace(/\.eval\.ya?ml$/i, "").replace(/\.ya?ml$/i, "") || "eval";
+  const suiteName = suiteNameFromFile && suiteNameFromFile.length > 0 ? suiteNameFromFile : fallbackSuiteName;
+  const rawTestCases = resolveTests(suite);
   const globalEvaluator = coerceEvaluator(suite.evaluator, "global") ?? "llm-grader";
   const evalFileDir = import_node_path8.default.dirname(absoluteTestPath);
-  let expandedTestcases;
-  if (typeof rawTestcases === "string") {
-    const externalPath = import_node_path8.default.resolve(evalFileDir, rawTestcases);
-    expandedTestcases = await loadCasesFromFile(externalPath);
-  } else if (Array.isArray(rawTestcases)) {
-    expandedTestcases = await expandFileReferences(rawTestcases, evalFileDir);
+  let expandedTestCases;
+  if (typeof rawTestCases === "string") {
+    const externalPath = import_node_path8.default.resolve(evalFileDir, rawTestCases);
+    expandedTestCases = await loadCasesFromFile(externalPath);
+  } else if (Array.isArray(rawTestCases)) {
+    expandedTestCases = await expandFileReferences(rawTestCases, evalFileDir);
   } else {
     throw new Error(`Invalid test file format: ${evalFilePath} - missing 'tests' field`);
   }
@@ -4791,32 +4974,33 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
   }
   const globalExecution = suiteAssertions !== void 0 ? { ...rawGlobalExecution ?? {}, assertions: suiteAssertions } : rawGlobalExecution;
   const results = [];
-  for (const rawEvalcase of expandedTestcases) {
-    if (!isJsonObject(rawEvalcase)) {
+  for (const rawTestCase of expandedTestCases) {
+    if (!isJsonObject(rawTestCase)) {
       logWarning5("Skipping invalid test entry (expected object)");
       continue;
     }
-    const evalcase = rawEvalcase;
-    const id = asString5(evalcase.id);
-    if (filterPattern && (!id || !import_micromatch2.default.isMatch(id, filterPattern))) {
+    const testCaseConfig = rawTestCase;
+    const id = asString5(testCaseConfig.id);
+    if (filterPattern && (!id || !matchesFilter2(id, filterPattern))) {
       continue;
     }
-    const conversationId = asString5(evalcase.conversation_id);
-    let outcome = asString5(evalcase.criteria);
-    if (!outcome && evalcase.expected_outcome !== void 0) {
-      outcome = asString5(evalcase.expected_outcome);
+    const conversationId = asString5(testCaseConfig.conversation_id);
+    let outcome = asString5(testCaseConfig.criteria);
+    if (!outcome && testCaseConfig.expected_outcome !== void 0) {
+      outcome = asString5(testCaseConfig.expected_outcome);
       if (outcome) {
         logWarning5(
-          `Test '${asString5(evalcase.id) ?? "unknown"}': 'expected_outcome' is deprecated. Use 'criteria' instead.`
+          `Test '${asString5(testCaseConfig.id) ?? "unknown"}': 'expected_outcome' is deprecated. Use 'criteria' instead.`
         );
       }
     }
-    const caseExecution = isJsonObject(evalcase.execution) ? evalcase.execution : void 0;
+    const caseExecution = isJsonObject(testCaseConfig.execution) ? testCaseConfig.execution : void 0;
     const skipDefaults = caseExecution?.skip_defaults === true;
+    const caseThreshold = typeof caseExecution?.threshold === "number" && caseExecution.threshold >= 0 && caseExecution.threshold <= 1 ? caseExecution.threshold : void 0;
     const effectiveSuiteInputFiles = suiteInputFiles && !skipDefaults ? suiteInputFiles : void 0;
-    const testInputMessages = resolveInputMessages(evalcase, effectiveSuiteInputFiles);
-    const expectedMessages = resolveExpectedMessages(evalcase) ?? [];
-    const hasEvaluationSpec = !!outcome || expectedMessages.length > 0 || evalcase.assertions !== void 0 || evalcase.assert !== void 0;
+    const testInputMessages = resolveInputMessages(testCaseConfig, effectiveSuiteInputFiles);
+    const expectedMessages = resolveExpectedMessages(testCaseConfig) ?? [];
+    const hasEvaluationSpec = !!outcome || expectedMessages.length > 0 || testCaseConfig.assertions !== void 0 || testCaseConfig.assert !== void 0;
     if (!id || !hasEvaluationSpec || !testInputMessages || testInputMessages.length === 0) {
       logError3(
         `Skipping incomplete test: ${id ?? "unknown"}. Missing required fields: id, input, and at least one of criteria/expected_output/assertions`
@@ -4863,16 +5047,21 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       }
     }
     const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    const testCaseEvaluatorKind = coerceEvaluator(testCaseConfig.evaluator, id) ?? globalEvaluator;
     let evaluators;
     try {
-      evaluators = await parseEvaluators(evalcase, globalExecution, searchRoots, id ?? "unknown");
+      evaluators = await parseEvaluators(
+        testCaseConfig,
+        globalExecution,
+        searchRoots,
+        id ?? "unknown"
+      );
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       logError3(`Skipping test '${id}': ${message}`);
       continue;
     }
-    const inlineRubrics = evalcase.rubrics;
+    const inlineRubrics = testCaseConfig.rubrics;
     if (inlineRubrics !== void 0 && Array.isArray(inlineRubrics)) {
       const rubricEvaluator = parseInlineRubrics(inlineRubrics);
       if (rubricEvaluator) {
@@ -4881,13 +5070,13 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
     }
     warnUnconsumedCriteria(outcome, evaluators, id ?? "unknown");
     const userFilePaths = collectResolvedInputFilePaths(inputMessages);
-    const caseWorkspace = await resolveWorkspaceConfig(evalcase.workspace, evalFileDir);
+    const caseWorkspace = await resolveWorkspaceConfig(testCaseConfig.workspace, evalFileDir);
     const mergedWorkspace = mergeWorkspaceConfigs(suiteWorkspace, caseWorkspace);
-    const metadata = isJsonObject(evalcase.metadata) ? evalcase.metadata : void 0;
-    const caseTargets = extractTargetsFromTestCase(evalcase);
+    const metadata = isJsonObject(testCaseConfig.metadata) ? testCaseConfig.metadata : void 0;
+    const caseTargets = extractTargetsFromTestCase(testCaseConfig);
     const testCase = {
       id,
-      dataset: evalSetName,
+      suite: suiteName,
       category: options?.category,
       conversation_id: conversationId,
       question,
@@ -4896,11 +5085,12 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       reference_answer: referenceAnswer,
       file_paths: userFilePaths,
       criteria: outcome ?? "",
-      evaluator: evalCaseEvaluatorKind,
+      evaluator: testCaseEvaluatorKind,
       assertions: evaluators,
       workspace: mergedWorkspace,
       metadata,
-      targets: caseTargets
+      targets: caseTargets,
+      ...caseThreshold !== void 0 ? { threshold: caseThreshold } : {}
     };
     results.push(testCase);
   }
@@ -5566,7 +5756,7 @@ var AzureProvider = class {
     };
     this.retryConfig = config.retry;
     const azure = (0, import_azure2.createAzure)(buildAzureOptions(config));
-    this.model = azure.chat(config.deploymentName);
+    this.model = config.apiFormat === "responses" ? azure(config.deploymentName) : azure.chat(config.deploymentName);
   }
   id;
   kind = "azure";
@@ -5692,7 +5882,9 @@ function buildAzureOptions(config) {
   const options = {
     apiKey: config.apiKey,
     apiVersion: config.version,
-    useDeploymentBasedUrls: true
+    // Chat completions still use deployment-scoped Azure URLs for compatibility
+    // with existing deployments. Responses API should use the SDK's v1 path.
+    useDeploymentBasedUrls: config.apiFormat !== "responses"
   };
   const baseURL = normalizeAzureBaseUrl(config.resourceName);
   if (baseURL) {
@@ -7169,15 +7361,16 @@ var CliProvider = class {
       outputFilePath
     );
     const renderedCommand = renderTemplate(this.config.command, templateValues);
+    const effectiveCwd = requests[0]?.cwd ?? this.config.cwd;
     if (this.verbose) {
       console.log(
-        `[cli-provider:${this.targetName}] (batch size=${requests.length}) cwd=${this.config.cwd ?? ""} command=${renderedCommand}`
+        `[cli-provider:${this.targetName}] (batch size=${requests.length}) cwd=${effectiveCwd ?? ""} command=${renderedCommand}`
       );
     }
     try {
       const startTime = Date.now();
       const result = await this.runCommand(renderedCommand, {
-        cwd: this.config.cwd,
+        cwd: effectiveCwd,
         env: process.env,
         timeoutMs: this.config.timeoutMs,
         signal: controller.signal
@@ -7210,7 +7403,7 @@ var CliProvider = class {
               command: renderedCommand,
               stderr: result.stderr,
               exitCode: result.exitCode ?? 0,
-              cwd: this.config.cwd,
+              cwd: effectiveCwd,
               outputFile: outputFilePath
             }
           };
@@ -7228,7 +7421,7 @@ var CliProvider = class {
               command: renderedCommand,
               stderr: result.stderr,
               exitCode: result.exitCode ?? 0,
-              cwd: this.config.cwd,
+              cwd: effectiveCwd,
               outputFile: outputFilePath,
               error: errorMessage
             }
@@ -7243,7 +7436,7 @@ var CliProvider = class {
             command: renderedCommand,
             stderr: result.stderr,
             exitCode: result.exitCode ?? 0,
-            cwd: this.config.cwd,
+            cwd: effectiveCwd,
             outputFile: outputFilePath,
             recordId: evalCaseId
           }
@@ -9267,6 +9460,76 @@ function subscribeToPiLogEntries(listener) {
   };
 }
 
+// src/evaluation/providers/pi-provider-aliases.ts
+init_cjs_shims();
+var SUBPROVIDER_ALIASES = {
+  azure: "azure-openai-responses"
+};
+var SUBPROVIDER_ALIASES_WITH_BASE_URL = {
+  // Azure v1 endpoints are OpenAI-compatible; use the standard client
+  // to avoid AzureOpenAI adding api-version query params.
+  azure: "openai-responses"
+};
+var ENV_KEY_MAP = {
+  google: "GEMINI_API_KEY",
+  gemini: "GEMINI_API_KEY",
+  anthropic: "ANTHROPIC_API_KEY",
+  openai: "OPENAI_API_KEY",
+  groq: "GROQ_API_KEY",
+  xai: "XAI_API_KEY",
+  openrouter: "OPENROUTER_API_KEY",
+  azure: "AZURE_OPENAI_API_KEY"
+};
+var ENV_BASE_URL_MAP = {
+  openai: "OPENAI_BASE_URL",
+  azure: "AZURE_OPENAI_BASE_URL",
+  openrouter: "OPENROUTER_BASE_URL"
+};
+function resolveSubprovider(name, hasBaseUrl = false) {
+  const lower = name.toLowerCase();
+  if (hasBaseUrl) {
+    const alias = SUBPROVIDER_ALIASES_WITH_BASE_URL[lower];
+    if (alias) return alias;
+  }
+  return SUBPROVIDER_ALIASES[lower] ?? name;
+}
+function resolveCliProvider(name) {
+  const lower = name.toLowerCase();
+  if (lower === "azure") return "azure-openai-responses";
+  return name;
+}
+function resolveEnvKeyName(provider, hasBaseUrl = false) {
+  const lower = provider.toLowerCase();
+  if (hasBaseUrl && lower === "azure") return "OPENAI_API_KEY";
+  return ENV_KEY_MAP[lower];
+}
+function resolveEnvBaseUrlName(provider, hasBaseUrl = false) {
+  const lower = provider.toLowerCase();
+  if (hasBaseUrl && lower === "azure") return "OPENAI_BASE_URL";
+  return ENV_BASE_URL_MAP[lower];
+}
+function extractAzureResourceName(baseUrl) {
+  const urlMatch = baseUrl.match(/^https?:\/\/([^./]+)/);
+  if (urlMatch) return urlMatch[1];
+  return baseUrl;
+}
+function normalizeAzureSdkBaseUrl(baseUrl) {
+  const trimmed = baseUrl.trim().replace(/\/+$/, "");
+  if (!trimmed) {
+    return trimmed;
+  }
+  if (!/^https?:\/\//i.test(trimmed)) {
+    return `https://${trimmed}.openai.azure.com/openai/v1`;
+  }
+  if (/\/openai\/v1$/i.test(trimmed)) {
+    return trimmed;
+  }
+  if (/\/openai$/i.test(trimmed)) {
+    return `${trimmed}/v1`;
+  }
+  return `${trimmed}/openai/v1`;
+}
+
 // src/evaluation/providers/pi-utils.ts
 init_cjs_shims();
 function extractPiTextContent(content) {
@@ -9426,12 +9689,12 @@ var PiCliProvider = class {
   buildPiArgs(prompt, inputFiles) {
     const args = [];
     if (this.config.subprovider) {
-      args.push("--provider", this.config.subprovider);
+      args.push("--provider", resolveCliProvider(this.config.subprovider));
     }
     if (this.config.model) {
       args.push("--model", this.config.model);
     }
-    if (this.config.apiKey) {
+    if (this.config.apiKey && this.config.subprovider?.toLowerCase() !== "azure") {
       args.push("--api-key", this.config.apiKey);
     }
     args.push("--mode", "json");
@@ -9483,35 +9746,35 @@ ${prompt}` : prompt;
   }
   buildEnv() {
     const env = { ...process.env };
-    if (this.config.apiKey) {
-      const provider = this.config.subprovider?.toLowerCase() ?? "google";
-      const ENV_KEY_MAP = {
-        google: "GEMINI_API_KEY",
-        gemini: "GEMINI_API_KEY",
-        anthropic: "ANTHROPIC_API_KEY",
-        openai: "OPENAI_API_KEY",
-        groq: "GROQ_API_KEY",
-        xai: "XAI_API_KEY",
-        openrouter: "OPENROUTER_API_KEY"
-      };
-      const envKey = ENV_KEY_MAP[provider];
-      if (envKey) {
-        env[envKey] = this.config.apiKey;
+    const provider = this.config.subprovider?.toLowerCase() ?? "google";
+    if (provider === "azure") {
+      if (this.config.apiKey) {
+        env.AZURE_OPENAI_API_KEY = this.config.apiKey;
+      }
+      if (this.config.baseUrl) {
+        env.AZURE_OPENAI_RESOURCE_NAME = extractAzureResourceName(this.config.baseUrl);
+      }
+    } else {
+      if (this.config.apiKey) {
+        const envKey = resolveEnvKeyName(provider);
+        if (envKey) {
+          env[envKey] = this.config.apiKey;
+        }
       }
     }
     if (this.config.subprovider) {
-      const provider = this.config.subprovider.toLowerCase();
+      const resolvedProvider = resolveCliProvider(this.config.subprovider);
       const PROVIDER_OWN_PREFIXES = {
         openrouter: ["OPENROUTER_"],
         anthropic: ["ANTHROPIC_"],
         openai: ["OPENAI_"],
-        azure: ["AZURE_OPENAI_"],
+        "azure-openai-responses": ["AZURE_OPENAI_"],
         google: ["GEMINI_", "GOOGLE_GENERATIVE_AI_"],
         gemini: ["GEMINI_", "GOOGLE_GENERATIVE_AI_"],
         groq: ["GROQ_"],
         xai: ["XAI_"]
       };
-      const ownPrefixes = PROVIDER_OWN_PREFIXES[provider] ?? [];
+      const ownPrefixes = PROVIDER_OWN_PREFIXES[resolvedProvider] ?? [];
       const allOtherPrefixes = Object.entries(PROVIDER_OWN_PREFIXES).filter(([key]) => key !== provider).flatMap(([, prefixes]) => prefixes);
       for (const key of Object.keys(env)) {
         if (allOtherPrefixes.some((prefix) => key.startsWith(prefix)) && !ownPrefixes.some((prefix) => key.startsWith(prefix))) {
@@ -9802,6 +10065,24 @@ function extractMessages(events) {
       }
     }
   }
+  if (messages) {
+    for (let i = messages.length - 1; i >= 0; i--) {
+      if (messages[i].role === "assistant" && !messages[i].content) {
+        for (let j = events.length - 1; j >= 0; j--) {
+          const evt = events[j];
+          if (!evt || evt.type !== "message_end") continue;
+          const msg = evt.message;
+          if (msg?.role !== "assistant") continue;
+          const text = extractPiTextContent(msg.content);
+          if (text) {
+            messages[i] = { ...messages[i], content: text };
+            break;
+          }
+        }
+        break;
+      }
+    }
+  }
   const eventToolCalls = extractToolCallsFromEvents(events);
   if (eventToolCalls.length > 0) {
     injectEventToolCalls(messages, eventToolCalls);
@@ -9986,17 +10267,43 @@ function formatTimeoutSuffix3(timeoutMs) {
   if (!timeoutMs || timeoutMs <= 0) return "";
   return ` after ${Math.ceil(timeoutMs / 1e3)}s`;
 }
+function resolveWindowsCmd(executable) {
+  if (process.platform !== "win32") return [executable, []];
+  const lower = executable.toLowerCase();
+  if (lower.endsWith(".js") || lower.endsWith(".exe")) return [executable, []];
+  let fullPath;
+  try {
+    fullPath = (0, import_node_child_process4.execSync)(`where ${executable}`, { encoding: "utf-8" }).trim().split(/\r?\n/)[0].trim();
+  } catch {
+    return [executable, []];
+  }
+  const cmdPath = fullPath.endsWith(".cmd") ? fullPath : `${fullPath}.cmd`;
+  try {
+    const content = (0, import_node_fs9.readFileSync)(cmdPath, "utf-8");
+    const match = content.match(/"?%_prog%"?\s+"([^"]+\.js)"/);
+    if (match) {
+      const dp0 = import_node_path21.default.dirname(import_node_path21.default.resolve(cmdPath));
+      const scriptPath = match[1].replace(/%dp0%[/\\]?/gi, `${dp0}${import_node_path21.default.sep}`);
+      try {
+        (0, import_node_fs9.accessSync)(scriptPath);
+        return ["node", [scriptPath]];
+      } catch {
+      }
+    }
+  } catch {
+  }
+  return [executable, []];
+}
 async function defaultPiRunner(options) {
   return await new Promise((resolve, reject) => {
     const parts = options.executable.split(/\s+/);
-    const executable = parts[0];
-    const executableArgs = parts.slice(1);
+    const [resolvedExe, prefixArgs] = resolveWindowsCmd(parts[0]);
+    const executableArgs = [...prefixArgs, ...parts.slice(1)];
     const allArgs = [...executableArgs, ...options.args];
-    const child = (0, import_node_child_process4.spawn)(executable, allArgs, {
+    const child = (0, import_node_child_process4.spawn)(resolvedExe, allArgs, {
       cwd: options.cwd,
       env: options.env,
-      stdio: ["pipe", "pipe", "pipe"],
-      shell: false
+      stdio: ["pipe", "pipe", "pipe"]
     });
     let stdout = "";
     let stderr = "";
@@ -10056,9 +10363,40 @@ var import_node_child_process5 = require("child_process");
 var import_node_crypto8 = require("crypto");
 var import_node_fs10 = require("fs");
 var import_promises19 = require("fs/promises");
-var import_node_path22 = __toESM(require("path"), 1);
+var import_node_path23 = __toESM(require("path"), 1);
 var import_node_readline = require("readline");
 var import_node_url3 = require("url");
+
+// src/paths.ts
+init_cjs_shims();
+var import_node_os6 = __toESM(require("os"), 1);
+var import_node_path22 = __toESM(require("path"), 1);
+var logged = false;
+function getAgentvHome() {
+  const envHome = process.env.AGENTV_HOME;
+  if (envHome && envHome !== "undefined") {
+    if (!logged) {
+      logged = true;
+      console.warn(`Using AGENTV_HOME: ${envHome}`);
+    }
+    return envHome;
+  }
+  return import_node_path22.default.join(import_node_os6.default.homedir(), ".agentv");
+}
+function getWorkspacesRoot() {
+  return import_node_path22.default.join(getAgentvHome(), "workspaces");
+}
+function getSubagentsRoot() {
+  return import_node_path22.default.join(getAgentvHome(), "subagents");
+}
+function getTraceStateRoot() {
+  return import_node_path22.default.join(getAgentvHome(), "trace-state");
+}
+function getWorkspacePoolRoot() {
+  return import_node_path22.default.join(getAgentvHome(), "workspace-pool");
+}
+
+// src/evaluation/providers/pi-coding-agent.ts
 var piCodingAgentModule = null;
 var piAiModule = null;
 var loadingPromise = null;
@@ -10076,46 +10414,126 @@ async function promptInstall() {
     rl.close();
   }
 }
-function findAgentvRoot() {
-  const thisFile = (0, import_node_url3.fileURLToPath)(importMetaUrl);
-  let dir = import_node_path22.default.dirname(thisFile);
-  for (let i = 0; i < 10; i++) {
+function findManagedSdkInstallRoot() {
+  return import_node_path23.default.join(getAgentvHome(), "deps", "pi-sdk");
+}
+function resolveGlobalNpmRoot() {
+  try {
+    const root = (0, import_node_child_process5.execSync)("npm root -g", {
+      encoding: "utf-8",
+      stdio: ["ignore", "pipe", "ignore"]
+    }).trim();
+    return root.length > 0 ? root : void 0;
+  } catch {
+    return void 0;
+  }
+}
+function buildGlobalModuleEntry(moduleName, globalNpmRoot) {
+  return import_node_path23.default.join(globalNpmRoot, ...moduleName.split("/"), "dist", "index.js");
+}
+function findAccessiblePath(paths) {
+  for (const candidate of paths) {
     try {
-      const pkg = import_node_path22.default.join(dir, "package.json");
-      (0, import_node_fs10.accessSync)(pkg);
-      return dir;
+      (0, import_node_fs10.accessSync)(candidate);
+      return candidate;
     } catch {
-      const parent = import_node_path22.default.dirname(dir);
-      if (parent === dir) break;
-      dir = parent;
     }
   }
-  return import_node_path22.default.dirname(thisFile);
+  return void 0;
 }
-async function doLoadSdkModules() {
+async function tryImportLocalSdkModules() {
   try {
     [piCodingAgentModule, piAiModule] = await Promise.all([
       import("@mariozechner/pi-coding-agent"),
       import("@mariozechner/pi-ai")
     ]);
+    return true;
   } catch {
-    if (await promptInstall()) {
-      const installDir = findAgentvRoot();
-      console.error(`Installing @mariozechner/pi-coding-agent into ${installDir}...`);
-      (0, import_node_child_process5.execSync)("bun add @mariozechner/pi-coding-agent", {
-        cwd: installDir,
-        stdio: "inherit"
-      });
-      [piCodingAgentModule, piAiModule] = await Promise.all([
-        import("@mariozechner/pi-coding-agent"),
-        import("@mariozechner/pi-ai")
-      ]);
-    } else {
-      throw new Error(
-        "pi-coding-agent SDK is not installed. Install it with:\n  bun add @mariozechner/pi-coding-agent"
-      );
+    return false;
+  }
+}
+async function tryImportManagedSdkModules() {
+  const managedRoot = findManagedSdkInstallRoot();
+  const piCodingAgentEntry = findAccessiblePath([
+    import_node_path23.default.join(managedRoot, "node_modules", "@mariozechner", "pi-coding-agent", "dist", "index.js")
+  ]);
+  const piAiEntry = findAccessiblePath([
+    import_node_path23.default.join(managedRoot, "node_modules", "@mariozechner", "pi-ai", "dist", "index.js"),
+    import_node_path23.default.join(
+      managedRoot,
+      "node_modules",
+      "@mariozechner",
+      "pi-coding-agent",
+      "node_modules",
+      "@mariozechner",
+      "pi-ai",
+      "dist",
+      "index.js"
+    )
+  ]);
+  if (!piCodingAgentEntry || !piAiEntry) return false;
+  try {
+    [piCodingAgentModule, piAiModule] = await Promise.all([
+      import((0, import_node_url3.pathToFileURL)(piCodingAgentEntry).href),
+      import((0, import_node_url3.pathToFileURL)(piAiEntry).href)
+    ]);
+    return true;
+  } catch {
+    return false;
+  }
+}
+async function tryImportGlobalSdkModules() {
+  const globalNpmRoot = resolveGlobalNpmRoot();
+  if (!globalNpmRoot) return false;
+  const piCodingAgentEntry = findAccessiblePath([
+    buildGlobalModuleEntry("@mariozechner/pi-coding-agent", globalNpmRoot)
+  ]);
+  const piAiEntry = findAccessiblePath([
+    buildGlobalModuleEntry("@mariozechner/pi-ai", globalNpmRoot),
+    import_node_path23.default.join(
+      globalNpmRoot,
+      "@mariozechner",
+      "pi-coding-agent",
+      "node_modules",
+      "@mariozechner",
+      "pi-ai",
+      "dist",
+      "index.js"
+    )
+  ]);
+  if (!piCodingAgentEntry || !piAiEntry) return false;
+  try {
+    [piCodingAgentModule, piAiModule] = await Promise.all([
+      import((0, import_node_url3.pathToFileURL)(piCodingAgentEntry).href),
+      import((0, import_node_url3.pathToFileURL)(piAiEntry).href)
+    ]);
+    return true;
+  } catch {
+    return false;
+  }
+}
+function installSdkModules(installDir) {
+  console.error(`Installing @mariozechner/pi-coding-agent into ${installDir} via npm...`);
+  (0, import_node_fs10.mkdirSync)(installDir, { recursive: true });
+  (0, import_node_child_process5.execSync)("npm install --no-save --no-package-lock @mariozechner/pi-coding-agent", {
+    cwd: installDir,
+    stdio: "inherit"
+  });
+}
+async function doLoadSdkModules() {
+  if (await tryImportLocalSdkModules() || await tryImportManagedSdkModules() || await tryImportGlobalSdkModules()) {
+    return;
+  }
+  if (await promptInstall()) {
+    const installDir = findManagedSdkInstallRoot();
+    installSdkModules(installDir);
+    if (await tryImportManagedSdkModules()) {
+      return;
     }
   }
+  throw new Error(
+    "pi-coding-agent SDK is not installed. Install it with:\n  npm install @mariozechner/pi-coding-agent"
+  );
 }
 async function loadSdkModules() {
   if (!piCodingAgentModule || !piAiModule) {
@@ -10143,7 +10561,9 @@ async function loadSdkModules() {
     codingTools: piSdk.codingTools,
     toolMap,
     SessionManager: piSdk.SessionManager,
-    getModel: piAi.getModel
+    getModel: piAi.getModel,
+    // biome-ignore lint/suspicious/noExplicitAny: registerBuiltInApiProviders exists at runtime but not in type defs
+    registerBuiltInApiProviders: piAi.registerBuiltInApiProviders
   };
 }
 var PiCodingAgentProvider = class {
@@ -10165,17 +10585,35 @@ var PiCodingAgentProvider = class {
     const startTime = (/* @__PURE__ */ new Date()).toISOString();
     const startMs = Date.now();
     const sdk = await loadSdkModules();
+    sdk.registerBuiltInApiProviders();
     const logger = await this.createStreamLogger(request).catch(() => void 0);
     try {
       const cwd = this.resolveCwd(request.cwd);
-      const providerName = this.config.subprovider ?? "google";
+      const rawProvider = this.config.subprovider ?? "google";
+      const normalizedBaseUrl = this.normalizeSdkBaseUrl(rawProvider, this.config.baseUrl);
+      const hasBaseUrl = !!normalizedBaseUrl;
+      const providerName = resolveSubprovider(rawProvider, hasBaseUrl);
       const modelId = this.config.model ?? "gemini-2.5-flash";
-      this.setApiKeyEnv(providerName);
-      const model = sdk.getModel(providerName, modelId);
+      this.setApiKeyEnv(rawProvider, hasBaseUrl);
+      this.setBaseUrlEnv(rawProvider, normalizedBaseUrl, hasBaseUrl);
+      let model = sdk.getModel(providerName, modelId);
+      if (model && normalizedBaseUrl) {
+        model = { ...model, baseUrl: normalizedBaseUrl };
+      }
       if (!model) {
-        throw new Error(
-          `pi-coding-agent: getModel('${providerName}', '${modelId}') returned undefined. The model '${modelId}' is not registered for provider '${providerName}' in pi-ai. Check that subprovider and model are correct in your target config.`
-        );
+        const envProvider = providerName.replace(/-responses$/, "");
+        model = {
+          id: modelId,
+          name: modelId,
+          api: providerName,
+          provider: envProvider,
+          baseUrl: normalizedBaseUrl ?? "",
+          reasoning: false,
+          input: ["text"],
+          cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+          contextWindow: 128e3,
+          maxTokens: 16384
+        };
       }
       const tools = this.resolveTools(sdk);
       const { session } = await sdk.createAgentSession({
@@ -10328,28 +10766,35 @@ ${fileList}`;
     }
   }
   /** Maps config apiKey to the provider-specific env var the SDK reads. */
-  setApiKeyEnv(providerName) {
+  setApiKeyEnv(providerName, hasBaseUrl = false) {
     if (!this.config.apiKey) return;
-    const ENV_KEY_MAP = {
-      google: "GEMINI_API_KEY",
-      gemini: "GEMINI_API_KEY",
-      anthropic: "ANTHROPIC_API_KEY",
-      openai: "OPENAI_API_KEY",
-      groq: "GROQ_API_KEY",
-      xai: "XAI_API_KEY",
-      openrouter: "OPENROUTER_API_KEY"
-    };
-    const envKey = ENV_KEY_MAP[providerName.toLowerCase()];
+    const envKey = resolveEnvKeyName(providerName, hasBaseUrl);
     if (envKey) {
       process.env[envKey] = this.config.apiKey;
     }
   }
+  /** Maps config baseUrl to the provider-specific env var the SDK reads. */
+  setBaseUrlEnv(providerName, baseUrl = this.config.baseUrl, hasBaseUrl = false) {
+    const normalizedBaseUrl = this.normalizeSdkBaseUrl(providerName, baseUrl);
+    if (!normalizedBaseUrl) return;
+    const envKey = resolveEnvBaseUrlName(providerName, hasBaseUrl);
+    if (envKey) {
+      process.env[envKey] = normalizedBaseUrl;
+    }
+  }
+  normalizeSdkBaseUrl(providerName, baseUrl) {
+    if (!baseUrl) return void 0;
+    if (providerName.toLowerCase() === "azure") {
+      return normalizeAzureSdkBaseUrl(baseUrl);
+    }
+    return baseUrl;
+  }
   resolveCwd(cwdOverride) {
     if (cwdOverride) {
-      return import_node_path22.default.resolve(cwdOverride);
+      return import_node_path23.default.resolve(cwdOverride);
     }
     if (this.config.cwd) {
-      return import_node_path22.default.resolve(this.config.cwd);
+      return import_node_path23.default.resolve(this.config.cwd);
     }
     return process.cwd();
   }
@@ -10368,9 +10813,9 @@ ${fileList}`;
   }
   resolveLogDirectory() {
     if (this.config.logDir) {
-      return import_node_path22.default.resolve(this.config.logDir);
+      return import_node_path23.default.resolve(this.config.logDir);
     }
-    return import_node_path22.default.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
+    return import_node_path23.default.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -10384,7 +10829,7 @@ ${fileList}`;
       console.warn(`Skipping Pi stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = import_node_path22.default.join(logDir, buildLogFilename6(request, this.targetName));
+    const filePath = import_node_path23.default.join(logDir, buildLogFilename6(request, this.targetName));
     try {
       const logger = await PiStreamLogger2.create({
         filePath,
@@ -10599,19 +11044,17 @@ var ProviderRegistry = class {
 
 // src/evaluation/providers/targets.ts
 init_cjs_shims();
-var import_node_path23 = __toESM(require("path"), 1);
+var import_node_path24 = __toESM(require("path"), 1);
 var import_zod3 = require("zod");
 var CliHealthcheckHttpInputSchema = import_zod3.z.object({
   url: import_zod3.z.string().min(1, "healthcheck URL is required"),
-  timeout_seconds: import_zod3.z.number().positive().optional(),
-  timeoutSeconds: import_zod3.z.number().positive().optional()
-});
+  timeout_seconds: import_zod3.z.number().positive().optional()
+}).passthrough();
 var CliHealthcheckCommandInputSchema = import_zod3.z.object({
   command: import_zod3.z.string().min(1, "healthcheck command is required"),
   cwd: import_zod3.z.string().optional(),
-  timeout_seconds: import_zod3.z.number().positive().optional(),
-  timeoutSeconds: import_zod3.z.number().positive().optional()
-});
+  timeout_seconds: import_zod3.z.number().positive().optional()
+}).passthrough();
 var CliHealthcheckInputSchema = import_zod3.z.union([
   CliHealthcheckHttpInputSchema,
   CliHealthcheckCommandInputSchema
@@ -10623,36 +11066,28 @@ var CliTargetInputSchema = import_zod3.z.object({
   command: import_zod3.z.string(),
   // Files format - optional
   files_format: import_zod3.z.string().optional(),
-  filesFormat: import_zod3.z.string().optional(),
   attachments_format: import_zod3.z.string().optional(),
-  attachmentsFormat: import_zod3.z.string().optional(),
   // Working directory - optional
   cwd: import_zod3.z.string().optional(),
   // Workspace template directory - optional (mutually exclusive with cwd)
   workspace_template: import_zod3.z.string().optional(),
-  workspaceTemplate: import_zod3.z.string().optional(),
   // Timeout in seconds - optional
   timeout_seconds: import_zod3.z.number().positive().optional(),
-  timeoutSeconds: import_zod3.z.number().positive().optional(),
   // Healthcheck configuration - optional
   healthcheck: CliHealthcheckInputSchema.optional(),
   // Verbose mode - optional
   verbose: import_zod3.z.boolean().optional(),
   cli_verbose: import_zod3.z.boolean().optional(),
-  cliVerbose: import_zod3.z.boolean().optional(),
   // Keep temp files - optional
   keep_temp_files: import_zod3.z.boolean().optional(),
-  keepTempFiles: import_zod3.z.boolean().optional(),
   keep_output_files: import_zod3.z.boolean().optional(),
-  keepOutputFiles: import_zod3.z.boolean().optional(),
   // Common target fields
   grader_target: import_zod3.z.string().optional(),
   judge_target: import_zod3.z.string().optional(),
   // backward compat
   workers: import_zod3.z.number().int().min(1).optional(),
-  provider_batching: import_zod3.z.boolean().optional(),
-  providerBatching: import_zod3.z.boolean().optional()
-});
+  provider_batching: import_zod3.z.boolean().optional()
+}).passthrough();
 var CliHealthcheckHttpSchema = import_zod3.z.object({
   url: import_zod3.z.string().min(1),
   timeoutMs: import_zod3.z.number().positive().optional()
@@ -10677,7 +11112,7 @@ var CliTargetConfigSchema = import_zod3.z.object({
   keepTempFiles: import_zod3.z.boolean().optional()
 }).strict();
 function normalizeCliHealthcheck(input, env, targetName, evalFilePath) {
-  const timeoutSeconds = input.timeout_seconds ?? input.timeoutSeconds;
+  const timeoutSeconds = input.timeout_seconds;
   const timeoutMs = timeoutSeconds !== void 0 ? Math.floor(timeoutSeconds * 1e3) : void 0;
   if ("url" in input && input.url) {
     const url = resolveString(input.url, env, `${targetName} healthcheck URL`);
@@ -10696,11 +11131,11 @@ function normalizeCliHealthcheck(input, env, targetName, evalFilePath) {
     allowLiteral: true,
     optionalEnv: true
   });
-  if (cwd && evalFilePath && !import_node_path23.default.isAbsolute(cwd)) {
-    cwd = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), cwd);
+  if (cwd && evalFilePath && !import_node_path24.default.isAbsolute(cwd)) {
+    cwd = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), cwd);
   }
   if (!cwd && evalFilePath) {
-    cwd = import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath));
+    cwd = import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath));
   }
   return {
     command,
@@ -10711,9 +11146,9 @@ function normalizeCliHealthcheck(input, env, targetName, evalFilePath) {
 function normalizeCliTargetInput(input, env, evalFilePath) {
   const targetName = input.name;
   const command = resolveString(input.command, env, `${targetName} CLI command`, true);
-  const filesFormatSource = input.files_format ?? input.filesFormat ?? input.attachments_format ?? input.attachmentsFormat;
+  const filesFormatSource = input.files_format ?? input.attachments_format;
   const filesFormat = resolveOptionalLiteralString(filesFormatSource);
-  const workspaceTemplateSource = input.workspace_template ?? input.workspaceTemplate;
+  const workspaceTemplateSource = input.workspace_template;
   let workspaceTemplate = resolveOptionalString(
     workspaceTemplateSource,
     env,
@@ -10723,15 +11158,15 @@ function normalizeCliTargetInput(input, env, evalFilePath) {
       optionalEnv: true
     }
   );
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   let cwd = resolveOptionalString(input.cwd, env, `${targetName} working directory`, {
     allowLiteral: true,
     optionalEnv: true
   });
-  if (cwd && evalFilePath && !import_node_path23.default.isAbsolute(cwd)) {
-    cwd = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), cwd);
+  if (cwd && evalFilePath && !import_node_path24.default.isAbsolute(cwd)) {
+    cwd = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), cwd);
   }
   if (cwd && workspaceTemplate) {
     throw new Error(
@@ -10739,14 +11174,12 @@ function normalizeCliTargetInput(input, env, evalFilePath) {
     );
   }
   if (!cwd && !workspaceTemplate && evalFilePath) {
-    cwd = import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath));
+    cwd = import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath));
   }
-  const timeoutSeconds = input.timeout_seconds ?? input.timeoutSeconds;
+  const timeoutSeconds = input.timeout_seconds;
   const timeoutMs = timeoutSeconds !== void 0 ? Math.floor(timeoutSeconds * 1e3) : void 0;
-  const verbose = resolveOptionalBoolean(input.verbose ?? input.cli_verbose ?? input.cliVerbose);
-  const keepTempFiles = resolveOptionalBoolean(
-    input.keep_temp_files ?? input.keepTempFiles ?? input.keep_output_files ?? input.keepOutputFiles
-  );
+  const verbose = resolveOptionalBoolean(input.verbose ?? input.cli_verbose);
+  const keepTempFiles = resolveOptionalBoolean(input.keep_temp_files ?? input.keep_output_files);
   const healthcheck = input.healthcheck ? normalizeCliHealthcheck(input.healthcheck, env, targetName, evalFilePath) : void 0;
   return {
     command,
@@ -10767,15 +11200,106 @@ var CLI_PLACEHOLDERS = /* @__PURE__ */ new Set([
   "FILES",
   "OUTPUT_FILE"
 ]);
+var DEPRECATED_TARGET_CAMEL_CASE_FIELDS = /* @__PURE__ */ new Map([
+  ["providerBatching", "provider_batching"],
+  ["subagentModeAllowed", "subagent_mode_allowed"],
+  ["fallbackTargets", "fallback_targets"],
+  ["resourceName", "endpoint"],
+  ["baseUrl", "base_url"],
+  ["apiKey", "api_key"],
+  ["deploymentName", "model"],
+  ["thinkingBudget", "thinking_budget"],
+  ["maxTokens", "max_output_tokens"],
+  ["apiFormat", "api_format"],
+  ["timeoutSeconds", "timeout_seconds"],
+  ["logDir", "log_dir"],
+  ["logDirectory", "log_directory"],
+  ["logFormat", "log_format"],
+  ["logOutputFormat", "log_output_format"],
+  ["systemPrompt", "system_prompt"],
+  ["maxTurns", "max_turns"],
+  ["maxBudgetUsd", "max_budget_usd"],
+  ["dryRun", "dry_run"],
+  ["subagentRoot", "subagent_root"],
+  ["filesFormat", "files_format"],
+  ["attachmentsFormat", "attachments_format"],
+  ["cliUrl", "cli_url"],
+  ["cliPath", "cli_path"],
+  ["githubToken", "github_token"],
+  ["sessionDir", "session_dir"],
+  ["sessionId", "session_id"],
+  ["sessionStateDir", "session_state_dir"],
+  ["maxRetries", "max_retries"],
+  ["retryInitialDelayMs", "retry_initial_delay_ms"],
+  ["retryMaxDelayMs", "retry_max_delay_ms"],
+  ["retryBackoffFactor", "retry_backoff_factor"],
+  ["retryStatusCodes", "retry_status_codes"]
+]);
+var DEPRECATED_HEALTHCHECK_CAMEL_CASE_FIELDS = /* @__PURE__ */ new Map([
+  ["timeoutSeconds", "timeout_seconds"]
+]);
+function collectDeprecatedCamelCaseWarnings(value, location, aliases) {
+  if (typeof value !== "object" || value === null || Array.isArray(value)) {
+    return [];
+  }
+  const warnings = [];
+  for (const [camelCaseField, snakeCaseField] of aliases) {
+    if (Object.prototype.hasOwnProperty.call(value, camelCaseField)) {
+      warnings.push({
+        location: `${location}.${camelCaseField}`,
+        message: `camelCase field '${camelCaseField}' is no longer supported in targets.yaml. Use '${snakeCaseField}' instead.`
+      });
+    }
+  }
+  return warnings;
+}
+function assertNoDeprecatedCamelCaseTargetFields(definition) {
+  if (Object.prototype.hasOwnProperty.call(definition, "workspaceTemplate")) {
+    throw new Error(
+      `${definition.name}: target-level workspace_template has been removed. Use eval-level workspace.template.`
+    );
+  }
+  const warning = findDeprecatedCamelCaseTargetWarnings(
+    definition,
+    `target "${definition.name}"`
+  )[0];
+  if (!warning) {
+    return;
+  }
+  const fieldMatch = warning.message.match(/field '([^']+)'/);
+  const replacementMatch = warning.message.match(/Use '([^']+)' instead/);
+  const field = fieldMatch?.[1] ?? "unknown";
+  const replacement = replacementMatch?.[1] ?? "snake_case";
+  throw new Error(
+    `${warning.location}: camelCase field '${field}' is no longer supported in targets.yaml. Use '${replacement}' instead.`
+  );
+}
+function findDeprecatedCamelCaseTargetWarnings(target, location) {
+  const warnings = collectDeprecatedCamelCaseWarnings(
+    target,
+    location,
+    DEPRECATED_TARGET_CAMEL_CASE_FIELDS
+  );
+  if (typeof target !== "object" || target === null || Array.isArray(target)) {
+    return warnings;
+  }
+  const healthcheck = target.healthcheck;
+  warnings.push(
+    ...collectDeprecatedCamelCaseWarnings(
+      healthcheck,
+      `${location}.healthcheck`,
+      DEPRECATED_HEALTHCHECK_CAMEL_CASE_FIELDS
+    )
+  );
+  return warnings;
+}
 var COMMON_TARGET_SETTINGS = [
   "use_target",
   "provider_batching",
-  "providerBatching",
   "subagent_mode_allowed",
-  "subagentModeAllowed",
-  "fallback_targets",
-  "fallbackTargets"
+  "fallback_targets"
 ];
+var USE_TARGET_ENV_PATTERN = /^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/i;
 var BASE_TARGET_SCHEMA = import_zod3.z.object({
   name: import_zod3.z.string().min(1, "target name is required"),
   provider: import_zod3.z.string().optional(),
@@ -10785,43 +11309,40 @@ var BASE_TARGET_SCHEMA = import_zod3.z.object({
   // backward compat
   workers: import_zod3.z.number().int().min(1).optional(),
   workspace_template: import_zod3.z.string().optional(),
-  workspaceTemplate: import_zod3.z.string().optional(),
   subagent_mode_allowed: import_zod3.z.boolean().optional(),
-  fallback_targets: import_zod3.z.array(import_zod3.z.string().min(1)).optional(),
-  fallbackTargets: import_zod3.z.array(import_zod3.z.string().min(1)).optional()
+  fallback_targets: import_zod3.z.array(import_zod3.z.string().min(1)).optional()
 }).passthrough();
 var DEFAULT_AZURE_API_VERSION = "2024-12-01-preview";
+var DEFAULT_AZURE_RESPONSES_API_VERSION = "v1";
 var DEFAULT_OPENAI_BASE_URL = "https://api.openai.com/v1";
-function normalizeAzureApiVersion(value) {
+function normalizeAzureApiVersion(value, apiFormat) {
+  const defaultVersion = apiFormat === "responses" ? DEFAULT_AZURE_RESPONSES_API_VERSION : DEFAULT_AZURE_API_VERSION;
   if (!value) {
-    return DEFAULT_AZURE_API_VERSION;
+    return defaultVersion;
   }
   const trimmed = value.trim();
   if (trimmed.length === 0) {
-    return DEFAULT_AZURE_API_VERSION;
+    return defaultVersion;
   }
   const withoutPrefix = trimmed.replace(/^api[-_]?version\s*=\s*/i, "").trim();
-  return withoutPrefix.length > 0 ? withoutPrefix : DEFAULT_AZURE_API_VERSION;
+  return withoutPrefix.length > 0 ? withoutPrefix : defaultVersion;
 }
 function resolveRetryConfig(target) {
-  const maxRetries = resolveOptionalNumber(
-    target.max_retries ?? target.maxRetries,
-    `${target.name} max retries`
-  );
+  const maxRetries = resolveOptionalNumber(target.max_retries, `${target.name} max retries`);
   const initialDelayMs = resolveOptionalNumber(
-    target.retry_initial_delay_ms ?? target.retryInitialDelayMs,
+    target.retry_initial_delay_ms,
     `${target.name} retry initial delay`
   );
   const maxDelayMs = resolveOptionalNumber(
-    target.retry_max_delay_ms ?? target.retryMaxDelayMs,
+    target.retry_max_delay_ms,
     `${target.name} retry max delay`
   );
   const backoffFactor = resolveOptionalNumber(
-    target.retry_backoff_factor ?? target.retryBackoffFactor,
+    target.retry_backoff_factor,
     `${target.name} retry backoff factor`
   );
   const retryableStatusCodes = resolveOptionalNumberArray(
-    target.retry_status_codes ?? target.retryStatusCodes,
+    target.retry_status_codes,
     `${target.name} retry status codes`
   );
   if (maxRetries === void 0 && initialDelayMs === void 0 && maxDelayMs === void 0 && backoffFactor === void 0 && retryableStatusCodes === void 0) {
@@ -10835,9 +11356,56 @@ function resolveRetryConfig(target) {
     retryableStatusCodes
   };
 }
-function resolveTargetDefinition(definition, env = process.env, evalFilePath) {
+function resolveDelegatedTargetDefinition(name, definitions, env = process.env) {
+  let definition = definitions.get(name);
+  if (!definition) {
+    return void 0;
+  }
+  const visited = [definition.name];
+  for (let depth = 0; depth < 10; depth++) {
+    const rawUseTarget = typeof definition.use_target === "string" ? definition.use_target.trim() : void 0;
+    if (!rawUseTarget) {
+      return definition;
+    }
+    const envMatch = rawUseTarget.match(USE_TARGET_ENV_PATTERN);
+    const envVarName = envMatch?.[1];
+    const resolvedName = envVarName ? env[envVarName]?.trim() ?? "" : rawUseTarget;
+    if (resolvedName.length === 0) {
+      if (envVarName) {
+        throw new Error(
+          `Target "${definition.name}" uses use_target: \${{ ${envVarName} }}, but ${envVarName} is not set. Set ${envVarName} to the name of a concrete target (for example, "azure") before running the eval.`
+        );
+      }
+      throw new Error(
+        `Target "${definition.name}" has an empty use_target value. Point it at a concrete target name before running the eval.`
+      );
+    }
+    const next = definitions.get(resolvedName);
+    if (!next) {
+      if (envVarName) {
+        throw new Error(
+          `Target "${definition.name}" uses use_target: \${{ ${envVarName} }}, which resolved to "${resolvedName}", but no target named "${resolvedName}" exists.`
+        );
+      }
+      throw new Error(
+        `Target "${definition.name}" uses use_target: "${resolvedName}", but no target named "${resolvedName}" exists.`
+      );
+    }
+    if (visited.includes(next.name)) {
+      const chain = [...visited, next.name].join(" -> ");
+      throw new Error(`Circular use_target reference detected: ${chain}`);
+    }
+    definition = next;
+    visited.push(definition.name);
+  }
+  throw new Error(
+    `Target "${name}" exceeded the maximum use_target resolution depth (10). Check for a delegation loop or overly deep alias chain.`
+  );
+}
+function resolveTargetDefinition(definition, env = process.env, evalFilePath, options) {
+  assertNoDeprecatedCamelCaseTargetFields(definition);
   const parsed = BASE_TARGET_SCHEMA.parse(definition);
-  if (parsed.workspace_template !== void 0 || parsed.workspaceTemplate !== void 0) {
+  if (parsed.workspace_template !== void 0) {
     throw new Error(
       `${parsed.name}: target-level workspace_template has been removed. Use eval-level workspace.template.`
     );
@@ -10853,13 +11421,9 @@ function resolveTargetDefinition(definition, env = process.env, evalFilePath) {
     `${parsed.name} provider`,
     true
   ).toLowerCase();
-  const providerBatching = resolveOptionalBoolean(
-    parsed.provider_batching ?? parsed.providerBatching
-  );
-  const subagentModeAllowed = resolveOptionalBoolean(
-    parsed.subagent_mode_allowed ?? parsed.subagentModeAllowed
-  );
-  const fallbackTargets = parsed.fallback_targets ?? parsed.fallbackTargets;
+  const providerBatching = resolveOptionalBoolean(parsed.provider_batching);
+  const subagentModeAllowed = resolveOptionalBoolean(parsed.subagent_mode_allowed);
+  const fallbackTargets = parsed.fallback_targets;
   const base = {
     name: parsed.name,
     graderTarget: parsed.grader_target ?? parsed.judge_target,
@@ -11009,20 +11573,22 @@ function normalizeOpenAIBaseUrl(value) {
   return trimmed.endsWith("/v1") ? trimmed : `${trimmed}/v1`;
 }
 function resolveAzureConfig(target, env) {
-  const endpointSource = target.endpoint ?? target.resource ?? target.resourceName;
-  const apiKeySource = target.api_key ?? target.apiKey;
-  const deploymentSource = target.deployment ?? target.deploymentName ?? target.model;
+  const endpointSource = target.endpoint ?? target.resource;
+  const apiKeySource = target.api_key;
+  const deploymentSource = target.deployment ?? target.model;
   const versionSource = target.version ?? target.api_version;
   const temperatureSource = target.temperature;
-  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
+  const maxTokensSource = target.max_output_tokens;
   const resourceName = resolveString(endpointSource, env, `${target.name} endpoint`);
   const apiKey = resolveString(apiKeySource, env, `${target.name} api key`);
   const deploymentName = resolveString(deploymentSource, env, `${target.name} deployment`);
+  const apiFormat = resolveApiFormat(target, env, target.name);
   const version = normalizeAzureApiVersion(
     resolveOptionalString(versionSource, env, `${target.name} api version`, {
       allowLiteral: true,
       optionalEnv: true
-    })
+    }),
+    apiFormat
   );
   const temperature = resolveOptionalNumber(temperatureSource, `${target.name} temperature`);
   const maxOutputTokens = resolveOptionalNumber(
@@ -11035,13 +11601,17 @@ function resolveAzureConfig(target, env) {
     deploymentName,
     apiKey,
     version,
+    apiFormat,
     temperature,
     maxOutputTokens,
     retry
   };
 }
-function resolveApiFormat(target, targetName) {
-  const raw = target.api_format ?? target.apiFormat;
+function resolveApiFormat(target, env, targetName) {
+  const raw = resolveOptionalString(target.api_format, env, `${targetName} api format`, {
+    allowLiteral: true,
+    optionalEnv: true
+  });
   if (raw === void 0) return void 0;
   if (raw === "chat" || raw === "responses") return raw;
   throw new Error(
@@ -11049,11 +11619,11 @@ function resolveApiFormat(target, targetName) {
   );
 }
 function resolveOpenAIConfig(target, env) {
-  const endpointSource = target.endpoint ?? target.base_url ?? target.baseUrl;
-  const apiKeySource = target.api_key ?? target.apiKey;
+  const endpointSource = target.endpoint ?? target.base_url;
+  const apiKeySource = target.api_key;
   const modelSource = target.model ?? target.deployment ?? target.variant;
   const temperatureSource = target.temperature;
-  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
+  const maxTokensSource = target.max_output_tokens;
   const baseURL = normalizeOpenAIBaseUrl(
     resolveOptionalString(endpointSource, env, `${target.name} endpoint`, {
       allowLiteral: true,
@@ -11067,17 +11637,17 @@ function resolveOpenAIConfig(target, env) {
     baseURL,
     apiKey,
     model,
-    apiFormat: resolveApiFormat(target, target.name),
+    apiFormat: resolveApiFormat(target, env, target.name),
     temperature: resolveOptionalNumber(temperatureSource, `${target.name} temperature`),
     maxOutputTokens: resolveOptionalNumber(maxTokensSource, `${target.name} max output tokens`),
     retry
   };
 }
 function resolveOpenRouterConfig(target, env) {
-  const apiKeySource = target.api_key ?? target.apiKey;
+  const apiKeySource = target.api_key;
   const modelSource = target.model ?? target.deployment ?? target.variant;
   const temperatureSource = target.temperature;
-  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
+  const maxTokensSource = target.max_output_tokens;
   const retry = resolveRetryConfig(target);
   return {
     apiKey: resolveString(apiKeySource, env, `${target.name} OpenRouter api key`),
@@ -11088,11 +11658,11 @@ function resolveOpenRouterConfig(target, env) {
   };
 }
 function resolveAnthropicConfig(target, env) {
-  const apiKeySource = target.api_key ?? target.apiKey;
+  const apiKeySource = target.api_key;
   const modelSource = target.model ?? target.deployment ?? target.variant;
   const temperatureSource = target.temperature;
-  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
-  const thinkingBudgetSource = target.thinking_budget ?? target.thinkingBudget;
+  const maxTokensSource = target.max_output_tokens;
+  const thinkingBudgetSource = target.thinking_budget;
   const apiKey = resolveString(apiKeySource, env, `${target.name} Anthropic api key`);
   const model = resolveString(modelSource, env, `${target.name} Anthropic model`);
   const retry = resolveRetryConfig(target);
@@ -11106,10 +11676,10 @@ function resolveAnthropicConfig(target, env) {
   };
 }
 function resolveGeminiConfig(target, env) {
-  const apiKeySource = target.api_key ?? target.apiKey;
+  const apiKeySource = target.api_key;
   const modelSource = target.model ?? target.deployment ?? target.variant;
   const temperatureSource = target.temperature;
-  const maxTokensSource = target.max_output_tokens ?? target.maxTokens;
+  const maxTokensSource = target.max_output_tokens;
   const apiKey = resolveString(apiKeySource, env, `${target.name} Google API key`);
   const model = resolveOptionalString(modelSource, env, `${target.name} Gemini model`, {
     allowLiteral: true,
@@ -11129,11 +11699,11 @@ function resolveCodexConfig(target, env, evalFilePath) {
   const executableSource = target.executable ?? target.command ?? target.binary;
   const argsSource = target.args ?? target.arguments;
   const cwdSource = target.cwd;
-  const workspaceTemplateSource = target.workspace_template ?? target.workspaceTemplate;
-  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
-  const logDirSource = target.log_dir ?? target.logDir ?? target.log_directory ?? target.logDirectory;
-  const logFormatSource = target.log_format ?? target.logFormat ?? target.log_output_format ?? target.logOutputFormat ?? env.AGENTV_CODEX_LOG_FORMAT;
-  const systemPromptSource = target.system_prompt ?? target.systemPrompt;
+  const workspaceTemplateSource = target.workspace_template;
+  const timeoutSource = target.timeout_seconds;
+  const logDirSource = target.log_dir ?? target.log_directory;
+  const logFormatSource = target.log_format ?? target.log_output_format ?? env.AGENTV_CODEX_LOG_FORMAT;
+  const systemPromptSource = target.system_prompt;
   const model = resolveOptionalString(modelSource, env, `${target.name} codex model`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11156,8 +11726,8 @@ function resolveCodexConfig(target, env, evalFilePath) {
       optionalEnv: true
     }
   );
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   if (cwd && workspaceTemplate) {
     throw new Error(
@@ -11197,16 +11767,16 @@ function normalizeCodexLogFormat(value) {
   throw new Error("codex log format must be 'summary' or 'json'");
 }
 function resolveCopilotSdkConfig(target, env, evalFilePath) {
-  const cliUrlSource = target.cli_url ?? target.cliUrl;
-  const cliPathSource = target.cli_path ?? target.cliPath;
-  const githubTokenSource = target.github_token ?? target.githubToken;
+  const cliUrlSource = target.cli_url;
+  const cliPathSource = target.cli_path;
+  const githubTokenSource = target.github_token;
   const modelSource = target.model;
   const cwdSource = target.cwd;
-  const workspaceTemplateSource = target.workspace_template ?? target.workspaceTemplate;
-  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
-  const logDirSource = target.log_dir ?? target.logDir ?? target.log_directory ?? target.logDirectory;
-  const logFormatSource = target.log_format ?? target.logFormat;
-  const systemPromptSource = target.system_prompt ?? target.systemPrompt;
+  const workspaceTemplateSource = target.workspace_template;
+  const timeoutSource = target.timeout_seconds;
+  const logDirSource = target.log_dir ?? target.log_directory;
+  const logFormatSource = target.log_format;
+  const systemPromptSource = target.system_prompt;
   const cliUrl = resolveOptionalString(cliUrlSource, env, `${target.name} copilot-sdk cli URL`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11241,8 +11811,8 @@ function resolveCopilotSdkConfig(target, env, evalFilePath) {
       optionalEnv: true
     }
   );
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   if (cwd && workspaceTemplate) {
     throw new Error(
@@ -11279,11 +11849,11 @@ function resolveCopilotCliConfig(target, env, evalFilePath) {
   const modelSource = target.model;
   const argsSource = target.args ?? target.arguments;
   const cwdSource = target.cwd;
-  const workspaceTemplateSource = target.workspace_template ?? target.workspaceTemplate;
-  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
-  const logDirSource = target.log_dir ?? target.logDir ?? target.log_directory ?? target.logDirectory;
-  const logFormatSource = target.log_format ?? target.logFormat;
-  const systemPromptSource = target.system_prompt ?? target.systemPrompt;
+  const workspaceTemplateSource = target.workspace_template;
+  const timeoutSource = target.timeout_seconds;
+  const logDirSource = target.log_dir ?? target.log_directory;
+  const logFormatSource = target.log_format;
+  const systemPromptSource = target.system_prompt;
   const executable = resolveOptionalString(executableSource, env, `${target.name} copilot-cli executable`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11306,8 +11876,8 @@ function resolveCopilotCliConfig(target, env, evalFilePath) {
       optionalEnv: true
     }
   );
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   if (cwd && workspaceTemplate) {
     throw new Error(
@@ -11347,16 +11917,16 @@ function normalizeCopilotLogFormat(value) {
 }
 function resolvePiCodingAgentConfig(target, env, evalFilePath) {
   const subproviderSource = target.subprovider;
-  const modelSource = target.model ?? target.pi_model ?? target.piModel;
-  const apiKeySource = target.api_key ?? target.apiKey;
-  const toolsSource = target.tools ?? target.pi_tools ?? target.piTools;
-  const thinkingSource = target.thinking ?? target.pi_thinking ?? target.piThinking;
+  const modelSource = target.model ?? target.pi_model;
+  const apiKeySource = target.api_key;
+  const toolsSource = target.tools ?? target.pi_tools;
+  const thinkingSource = target.thinking ?? target.pi_thinking;
   const cwdSource = target.cwd;
-  const workspaceTemplateSource = target.workspace_template ?? target.workspaceTemplate;
-  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
-  const logDirSource = target.log_dir ?? target.logDir ?? target.log_directory ?? target.logDirectory;
-  const logFormatSource = target.log_format ?? target.logFormat;
-  const systemPromptSource = target.system_prompt ?? target.systemPrompt;
+  const workspaceTemplateSource = target.workspace_template;
+  const timeoutSource = target.timeout_seconds;
+  const logDirSource = target.log_dir ?? target.log_directory;
+  const logFormatSource = target.log_format;
+  const systemPromptSource = target.system_prompt;
   const subprovider = resolveOptionalString(
     subproviderSource,
     env,
@@ -11374,6 +11944,11 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
     allowLiteral: false,
     optionalEnv: true
   });
+  const baseUrlSource = target.base_url ?? target.endpoint;
+  const baseUrl = resolveOptionalString(baseUrlSource, env, `${target.name} pi base url`, {
+    allowLiteral: true,
+    optionalEnv: true
+  });
   const tools = resolveOptionalString(toolsSource, env, `${target.name} pi tools`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11395,8 +11970,8 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
       optionalEnv: true
     }
   );
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   if (cwd && workspaceTemplate) {
     throw new Error(
@@ -11414,6 +11989,7 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
     subprovider,
     model,
     apiKey,
+    baseUrl,
     tools,
     thinking,
     cwd,
@@ -11427,16 +12003,16 @@ function resolvePiCodingAgentConfig(target, env, evalFilePath) {
 function resolvePiCliConfig(target, env, evalFilePath) {
   const executableSource = target.executable ?? target.command ?? target.binary;
   const subproviderSource = target.subprovider;
-  const modelSource = target.model ?? target.pi_model ?? target.piModel;
-  const apiKeySource = target.api_key ?? target.apiKey;
-  const toolsSource = target.tools ?? target.pi_tools ?? target.piTools;
-  const thinkingSource = target.thinking ?? target.pi_thinking ?? target.piThinking;
+  const modelSource = target.model ?? target.pi_model;
+  const apiKeySource = target.api_key;
+  const toolsSource = target.tools ?? target.pi_tools;
+  const thinkingSource = target.thinking ?? target.pi_thinking;
   const cwdSource = target.cwd;
-  const workspaceTemplateSource = target.workspace_template ?? target.workspaceTemplate;
-  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
-  const logDirSource = target.log_dir ?? target.logDir ?? target.log_directory ?? target.logDirectory;
-  const logFormatSource = target.log_format ?? target.logFormat;
-  const systemPromptSource = target.system_prompt ?? target.systemPrompt;
+  const workspaceTemplateSource = target.workspace_template;
+  const timeoutSource = target.timeout_seconds;
+  const logDirSource = target.log_dir ?? target.log_directory;
+  const logFormatSource = target.log_format;
+  const systemPromptSource = target.system_prompt;
   const executable = resolveOptionalString(executableSource, env, `${target.name} pi-cli executable`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11455,6 +12031,11 @@ function resolvePiCliConfig(target, env, evalFilePath) {
     allowLiteral: false,
     optionalEnv: true
   });
+  const baseUrlSource = target.base_url ?? target.endpoint;
+  const baseUrl = resolveOptionalString(baseUrlSource, env, `${target.name} pi-cli base url`, {
+    allowLiteral: true,
+    optionalEnv: true
+  });
   const tools = resolveOptionalString(toolsSource, env, `${target.name} pi-cli tools`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11475,8 +12056,8 @@ function resolvePiCliConfig(target, env, evalFilePath) {
     `${target.name} pi-cli workspace template`,
     { allowLiteral: true, optionalEnv: true }
   );
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   if (cwd && workspaceTemplate) {
     throw new Error(`${target.name}: 'cwd' and 'workspace_template' are mutually exclusive.`);
@@ -11493,6 +12074,7 @@ function resolvePiCliConfig(target, env, evalFilePath) {
     subprovider,
     model,
     apiKey,
+    baseUrl,
     tools,
     thinking,
     args,
@@ -11507,11 +12089,11 @@ function resolvePiCliConfig(target, env, evalFilePath) {
 function resolveClaudeConfig(target, env, evalFilePath) {
   const modelSource = target.model;
   const cwdSource = target.cwd;
-  const workspaceTemplateSource = target.workspace_template ?? target.workspaceTemplate;
-  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
-  const logDirSource = target.log_dir ?? target.logDir ?? target.log_directory ?? target.logDirectory;
-  const logFormatSource = target.log_format ?? target.logFormat ?? target.log_output_format ?? target.logOutputFormat ?? env.AGENTV_CLAUDE_LOG_FORMAT;
-  const systemPromptSource = target.system_prompt ?? target.systemPrompt;
+  const workspaceTemplateSource = target.workspace_template;
+  const timeoutSource = target.timeout_seconds;
+  const logDirSource = target.log_dir ?? target.log_directory;
+  const logFormatSource = target.log_format ?? target.log_output_format ?? env.AGENTV_CLAUDE_LOG_FORMAT;
+  const systemPromptSource = target.system_prompt;
   const model = resolveOptionalString(modelSource, env, `${target.name} claude model`, {
     allowLiteral: true,
     optionalEnv: true
@@ -11529,8 +12111,8 @@ function resolveClaudeConfig(target, env, evalFilePath) {
       optionalEnv: true
     }
   );
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   if (cwd && workspaceTemplate) {
     throw new Error(
@@ -11544,8 +12126,8 @@ function resolveClaudeConfig(target, env, evalFilePath) {
   });
   const logFormat = normalizeClaudeLogFormat(logFormatSource);
   const systemPrompt = typeof systemPromptSource === "string" && systemPromptSource.trim().length > 0 ? systemPromptSource.trim() : void 0;
-  const maxTurns = typeof target.max_turns === "number" ? target.max_turns : typeof target.maxTurns === "number" ? target.maxTurns : void 0;
-  const maxBudgetUsd = typeof target.max_budget_usd === "number" ? target.max_budget_usd : typeof target.maxBudgetUsd === "number" ? target.maxBudgetUsd : void 0;
+  const maxTurns = typeof target.max_turns === "number" ? target.max_turns : void 0;
+  const maxBudgetUsd = typeof target.max_budget_usd === "number" ? target.max_budget_usd : void 0;
   return {
     model,
     systemPrompt,
@@ -11576,9 +12158,7 @@ function resolveMockConfig(target) {
   return { response };
 }
 function resolveVSCodeConfig(target, env, insiders, evalFilePath) {
-  const workspaceTemplateEnvVar = resolveOptionalLiteralString(
-    target.workspace_template ?? target.workspaceTemplate
-  );
+  const workspaceTemplateEnvVar = resolveOptionalLiteralString(target.workspace_template);
   let workspaceTemplate = workspaceTemplateEnvVar ? resolveOptionalString(
     workspaceTemplateEnvVar,
     env,
@@ -11588,14 +12168,14 @@ function resolveVSCodeConfig(target, env, insiders, evalFilePath) {
       optionalEnv: true
     }
   ) : void 0;
-  if (workspaceTemplate && evalFilePath && !import_node_path23.default.isAbsolute(workspaceTemplate)) {
-    workspaceTemplate = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), workspaceTemplate);
+  if (workspaceTemplate && evalFilePath && !import_node_path24.default.isAbsolute(workspaceTemplate)) {
+    workspaceTemplate = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), workspaceTemplate);
   }
   const executableSource = target.executable;
   const waitSource = target.wait;
-  const dryRunSource = target.dry_run ?? target.dryRun;
-  const subagentRootSource = target.subagent_root ?? target.subagentRoot;
-  const timeoutSource = target.timeout_seconds ?? target.timeoutSeconds;
+  const dryRunSource = target.dry_run;
+  const subagentRootSource = target.subagent_root;
+  const timeoutSource = target.timeout_seconds;
   const defaultCommand = insiders ? "code-insiders" : "code";
   const executable = resolveOptionalString(executableSource, env, `${target.name} vscode executable`, {
     allowLiteral: true,
@@ -11630,8 +12210,8 @@ function resolveCliConfig(target, env, evalFilePath) {
   const parseResult = CliTargetInputSchema.safeParse(target, { errorMap: cliErrorMap });
   if (!parseResult.success) {
     const firstError = parseResult.error.errors[0];
-    const path52 = firstError?.path.join(".") || "";
-    const prefix = path52 ? `${target.name} ${path52}: ` : `${target.name}: `;
+    const path53 = firstError?.path.join(".") || "";
+    const prefix = path53 ? `${target.name} ${path53}: ` : `${target.name}: `;
     throw new Error(`${prefix}${firstError?.message}`);
   }
   const normalized = normalizeCliTargetInput(parseResult.data, env, evalFilePath);
@@ -11646,17 +12226,17 @@ function resolveCliConfig(target, env, evalFilePath) {
 }
 function resolveDiscoveredProviderConfig(target, providerKind, env, evalFilePath) {
   const command = target.command ? resolveString(target.command, env, `${target.name} command`, true) : `bun run .agentv/providers/${providerKind}.ts {PROMPT}`;
-  const timeoutSeconds = target.timeout_seconds ?? target.timeoutSeconds;
+  const timeoutSeconds = target.timeout_seconds;
   const timeoutMs = resolveTimeoutMs(timeoutSeconds, `${target.name} timeout`);
   let cwd = resolveOptionalString(target.cwd, env, `${target.name} working directory`, {
     allowLiteral: true,
     optionalEnv: true
   });
-  if (cwd && evalFilePath && !import_node_path23.default.isAbsolute(cwd)) {
-    cwd = import_node_path23.default.resolve(import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath)), cwd);
+  if (cwd && evalFilePath && !import_node_path24.default.isAbsolute(cwd)) {
+    cwd = import_node_path24.default.resolve(import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath)), cwd);
   }
   if (!cwd && evalFilePath) {
-    cwd = import_node_path23.default.dirname(import_node_path23.default.resolve(evalFilePath));
+    cwd = import_node_path24.default.dirname(import_node_path24.default.resolve(evalFilePath));
   }
   return {
     command,
@@ -11710,10 +12290,10 @@ function resolveDiscover(value, targetName) {
   throw new Error(`Target "${targetName}": discover must be "latest" (got "${String(value)}")`);
 }
 function resolveCopilotLogConfig(target, env) {
-  const sessionDirSource = target.session_dir ?? target.sessionDir;
-  const sessionIdSource = target.session_id ?? target.sessionId;
+  const sessionDirSource = target.session_dir;
+  const sessionIdSource = target.session_id;
   const discoverSource = target.discover;
-  const sessionStateDirSource = target.session_state_dir ?? target.sessionStateDir;
+  const sessionStateDirSource = target.session_state_dir;
   const cwdSource = target.cwd;
   return {
     sessionDir: resolveOptionalString(
@@ -11894,7 +12474,7 @@ var import_node_path33 = __toESM(require("path"), 1);
 init_cjs_shims();
 var import_node_fs11 = require("fs");
 var import_promises20 = require("fs/promises");
-var import_node_path24 = __toESM(require("path"), 1);
+var import_node_path25 = __toESM(require("path"), 1);
 async function pathExists(target) {
   try {
     await (0, import_promises20.access)(target, import_node_fs11.constants.F_OK);
@@ -11910,7 +12490,7 @@ async function readDirEntries(target) {
   const entries = await (0, import_promises20.readdir)(target, { withFileTypes: true });
   return entries.map((entry) => ({
     name: entry.name,
-    absolutePath: import_node_path24.default.join(target, entry.name),
+    absolutePath: import_node_path25.default.join(target, entry.name),
     isDirectory: entry.isDirectory()
   }));
 }
@@ -11926,9 +12506,9 @@ async function removeIfExists(target) {
 
 // src/evaluation/providers/vscode/utils/path.ts
 init_cjs_shims();
-var import_node_path25 = __toESM(require("path"), 1);
+var import_node_path26 = __toESM(require("path"), 1);
 function pathToFileUri2(filePath) {
-  const absolutePath = import_node_path25.default.isAbsolute(filePath) ? filePath : import_node_path25.default.resolve(filePath);
+  const absolutePath = import_node_path26.default.isAbsolute(filePath) ? filePath : import_node_path26.default.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -11938,7 +12518,7 @@ function pathToFileUri2(filePath) {
 
 // src/evaluation/providers/vscode/dispatch/promptBuilder.ts
 init_cjs_shims();
-var import_node_path26 = __toESM(require("path"), 1);
+var import_node_path27 = __toESM(require("path"), 1);
 
 // src/evaluation/providers/vscode/utils/template.ts
 init_cjs_shims();
@@ -12032,8 +12612,8 @@ function createBatchRequestPrompt(userQuery, responseFileTmp, responseFileFinal,
   });
 }
 function createBatchOrchestratorPrompt(requestFiles, responseFiles, templateContent) {
-  const requestLines = requestFiles.map((file, index) => `${index + 1}. messages/${import_node_path26.default.basename(file)}`).join("\n");
-  const responseList = responseFiles.map((file) => `"${import_node_path26.default.basename(file)}"`).join(", ");
+  const requestLines = requestFiles.map((file, index) => `${index + 1}. messages/${import_node_path27.default.basename(file)}`).join("\n");
+  const responseList = responseFiles.map((file) => `"${import_node_path27.default.basename(file)}"`).join(", ");
   return renderTemplate2(templateContent, {
     requestFiles: requestLines,
     responseList
@@ -12043,7 +12623,7 @@ function createBatchOrchestratorPrompt(requestFiles, responseFiles, templateCont
 // src/evaluation/providers/vscode/dispatch/responseWaiter.ts
 init_cjs_shims();
 var import_promises21 = require("fs/promises");
-var import_node_path27 = __toESM(require("path"), 1);
+var import_node_path28 = __toESM(require("path"), 1);
 
 // src/evaluation/providers/vscode/utils/time.ts
 init_cjs_shims();
@@ -12103,7 +12683,7 @@ async function waitForResponseOutput(responseFileFinal, pollInterval = 1e3, sile
 }
 async function waitForBatchResponses(responseFilesFinal, pollInterval = 1e3, silent = false, timeoutMs = DEFAULT_TIMEOUT_MS) {
   if (!silent) {
-    const fileList = responseFilesFinal.map((file) => import_node_path27.default.basename(file)).join(", ");
+    const fileList = responseFilesFinal.map((file) => import_node_path28.default.basename(file)).join(", ");
     console.error(`waiting for ${responseFilesFinal.length} batch response(s): ${fileList}`);
   }
   const deadline = Date.now() + timeoutMs;
@@ -12112,7 +12692,7 @@ async function waitForBatchResponses(responseFilesFinal, pollInterval = 1e3, sil
     while (pending.size > 0) {
       if (Date.now() >= deadline) {
         if (!silent) {
-          const remaining = [...pending].map((f) => import_node_path27.default.basename(f)).join(", ");
+          const remaining = [...pending].map((f) => import_node_path28.default.basename(f)).join(", ");
           console.error(
             `error: timed out after ${Math.round(timeoutMs / 1e3)}s waiting for batch responses. Still pending: ${remaining}`
           );
@@ -12170,37 +12750,6 @@ var import_node_util2 = require("util");
 // src/evaluation/providers/vscode/dispatch/constants.ts
 init_cjs_shims();
 var import_node_path29 = __toESM(require("path"), 1);
-
-// src/paths.ts
-init_cjs_shims();
-var import_node_os6 = __toESM(require("os"), 1);
-var import_node_path28 = __toESM(require("path"), 1);
-var logged = false;
-function getAgentvHome() {
-  const envHome = process.env.AGENTV_HOME;
-  if (envHome && envHome !== "undefined") {
-    if (!logged) {
-      logged = true;
-      console.warn(`Using AGENTV_HOME: ${envHome}`);
-    }
-    return envHome;
-  }
-  return import_node_path28.default.join(import_node_os6.default.homedir(), ".agentv");
-}
-function getWorkspacesRoot() {
-  return import_node_path28.default.join(getAgentvHome(), "workspaces");
-}
-function getSubagentsRoot() {
-  return import_node_path28.default.join(getAgentvHome(), "subagents");
-}
-function getTraceStateRoot() {
-  return import_node_path28.default.join(getAgentvHome(), "trace-state");
-}
-function getWorkspacePoolRoot() {
-  return import_node_path28.default.join(getAgentvHome(), "workspace-pool");
-}
-
-// src/evaluation/providers/vscode/dispatch/constants.ts
 var DEFAULT_LOCK_NAME = "subagent.lock";
 var DEFAULT_ALIVE_FILENAME = ".alive";
 function getDefaultSubagentRoot(vscodeCmd = "code") {
@@ -13353,6 +13902,15 @@ var AGENT_PROVIDER_KINDS = [
   "vscode",
   "vscode-insiders"
 ];
+var LLM_GRADER_CAPABLE_KINDS = [
+  "openai",
+  "openrouter",
+  "azure",
+  "anthropic",
+  "gemini",
+  "agentv",
+  "mock"
+];
 function extractLastAssistantContent(messages) {
   if (!messages || messages.length === 0) {
     return "";
@@ -13506,9 +14064,10 @@ init_cjs_shims();
 
 // src/evaluation/evaluators/scoring.ts
 init_cjs_shims();
-var PASS_THRESHOLD = 0.8;
-function scoreToVerdict(score) {
-  return score >= PASS_THRESHOLD ? "pass" : "fail";
+var DEFAULT_THRESHOLD = 0.8;
+var PASS_THRESHOLD = DEFAULT_THRESHOLD;
+function scoreToVerdict(score, threshold = DEFAULT_THRESHOLD) {
+  return score >= threshold ? "pass" : "fail";
 }
 function clampScore(value) {
   if (Number.isNaN(value) || !Number.isFinite(value)) {
@@ -13699,13 +14258,13 @@ async function execFileWithStdinNode(argv, stdinPayload, options) {
 async function execShellWithStdin(command, stdinPayload, options = {}) {
   const { mkdir: mkdir17, readFile: readFile17, rm: rm6, writeFile: writeFile9 } = await import("fs/promises");
   const { tmpdir: tmpdir3 } = await import("os");
-  const path52 = await import("path");
+  const path53 = await import("path");
   const { randomUUID: randomUUID10 } = await import("crypto");
-  const dir = path52.join(tmpdir3(), `agentv-exec-${randomUUID10()}`);
+  const dir = path53.join(tmpdir3(), `agentv-exec-${randomUUID10()}`);
   await mkdir17(dir, { recursive: true });
-  const stdinPath = path52.join(dir, "stdin.txt");
-  const stdoutPath = path52.join(dir, "stdout.txt");
-  const stderrPath = path52.join(dir, "stderr.txt");
+  const stdinPath = path53.join(dir, "stdin.txt");
+  const stdoutPath = path53.join(dir, "stdout.txt");
+  const stderrPath = path53.join(dir, "stderr.txt");
   await writeFile9(stdinPath, stdinPayload, "utf8");
   const wrappedCommand = process.platform === "win32" ? `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}` : `(${command}) < ${shellEscapePath(stdinPath)} > ${shellEscapePath(stdoutPath)} 2> ${shellEscapePath(stderrPath)}`;
   const { spawn: spawn5 } = await import("child_process");
@@ -14907,7 +15466,7 @@ ${outputSchema}`;
     parts.push("[[ ## scoring_criteria ## ]]");
     for (const rubric of rubrics) {
       const weightLabel = rubric.weight !== 1 ? ` (weight: ${rubric.weight})` : "";
-      const minScoreLabel = rubric.required_min_score !== void 0 ? ` [REQUIRED: min score ${rubric.required_min_score}]` : "";
+      const minScoreLabel = rubric.min_score !== void 0 ? ` [REQUIRED: min score ${rubric.min_score}]` : rubric.required_min_score !== void 0 ? ` [REQUIRED: min score ${rubric.required_min_score}]` : "";
       parts.push("", `### Criterion: ${rubric.id}${weightLabel}${minScoreLabel}`);
       if (rubric.outcome) {
         parts.push(`Description: ${rubric.outcome}`);
@@ -14961,54 +15520,106 @@ ${outputSchema}`;
   async runWithRetry(options) {
     const { context: context2, graderProvider, systemPrompt, userPrompt, schema, images } = options;
     let lastError;
+    let lastInvalidResponse;
+    let shouldAttemptStructureFix = false;
     for (let attempt = 1; attempt <= 3; attempt++) {
       try {
-        const model = graderProvider.asLanguageModel?.();
-        if (model) {
-          const modelOptions = {
-            ...this.maxOutputTokens ? { maxTokens: this.maxOutputTokens } : {},
-            ...typeof this.temperature === "number" ? { temperature: this.temperature } : {}
-          };
-          const hasImages = images && images.length > 0;
-          const result = hasImages ? await (0, import_ai2.generateText)({
-            model,
-            system: systemPrompt,
-            messages: [
-              {
-                role: "user",
-                content: [
-                  { type: "text", text: userPrompt },
-                  ...toAiSdkImageParts(images)
-                ]
-              }
-            ],
-            ...modelOptions
-          }) : await (0, import_ai2.generateText)({
-            model,
-            system: systemPrompt,
-            prompt: userPrompt,
-            ...modelOptions
-          });
-          const data2 = schema.parse(parseJsonFromText(result.text));
-          const rawUsage = result.usage;
-          const tokenUsage = rawUsage?.inputTokens != null && rawUsage?.outputTokens != null ? { input: rawUsage.inputTokens, output: rawUsage.outputTokens } : void 0;
-          return { data: data2, tokenUsage };
+        const result = await this.generateStructuredResponse({
+          context: context2,
+          graderProvider,
+          systemPrompt,
+          userPrompt,
+          images
+        });
+        const canRepairResponse = result.text.trim().length > 0;
+        lastInvalidResponse = canRepairResponse ? result : void 0;
+        let data;
+        try {
+          data = schema.parse(parseJsonFromText(result.text));
+        } catch (e) {
+          lastError = e instanceof Error ? e : new Error(String(e));
+          shouldAttemptStructureFix = canRepairResponse;
+          continue;
         }
-        const response = await graderProvider.invoke({
-          question: userPrompt,
+        return {
+          data,
+          providerResponse: result.providerResponse,
+          tokenUsage: result.tokenUsage
+        };
+      } catch (e) {
+        lastError = e instanceof Error ? e : new Error(String(e));
+      }
+    }
+    if (shouldAttemptStructureFix && lastInvalidResponse) {
+      try {
+        const repaired = await this.generateStructuredResponse({
+          context: context2,
+          graderProvider,
           systemPrompt,
-          evalCaseId: context2.evalCase.id,
-          attempt: context2.attempt,
-          maxOutputTokens: this.maxOutputTokens,
-          temperature: this.temperature
+          userPrompt: buildStructureRepairPrompt({
+            validationError: lastError?.message ?? "Schema validation failed",
+            invalidResponse: lastInvalidResponse.text
+          })
         });
-        const data = schema.parse(parseJsonFromText(extractLastAssistantContent(response.output)));
-        return { data, providerResponse: response, tokenUsage: response.tokenUsage };
+        const data = schema.parse(parseJsonFromText(repaired.text));
+        return {
+          data,
+          providerResponse: repaired.providerResponse,
+          tokenUsage: sumTokenUsage(lastInvalidResponse.tokenUsage, repaired.tokenUsage)
+        };
       } catch (e) {
         lastError = e instanceof Error ? e : new Error(String(e));
       }
     }
-    throw new Error(`Failed to parse evaluator response after 3 attempts: ${lastError?.message}`);
+    throw new Error(
+      `Failed to parse evaluator response after 3 attempts and 1 structure-fix attempt: ${lastError?.message}`
+    );
+  }
+  async generateStructuredResponse(options) {
+    const { context: context2, graderProvider, systemPrompt, userPrompt, images } = options;
+    const model = graderProvider.asLanguageModel?.();
+    if (model) {
+      const modelOptions = {
+        ...this.maxOutputTokens ? { maxTokens: this.maxOutputTokens } : {},
+        ...typeof this.temperature === "number" ? { temperature: this.temperature } : {}
+      };
+      const hasImages = images && images.length > 0;
+      const result = hasImages ? await (0, import_ai2.generateText)({
+        model,
+        system: systemPrompt,
+        messages: [
+          {
+            role: "user",
+            content: [
+              { type: "text", text: userPrompt },
+              ...toAiSdkImageParts(images)
+            ]
+          }
+        ],
+        ...modelOptions
+      }) : await (0, import_ai2.generateText)({
+        model,
+        system: systemPrompt,
+        prompt: userPrompt,
+        ...modelOptions
+      });
+      const rawUsage = result.usage;
+      const tokenUsage = rawUsage?.inputTokens != null && rawUsage?.outputTokens != null ? { input: rawUsage.inputTokens, output: rawUsage.outputTokens } : void 0;
+      return { text: result.text, tokenUsage };
+    }
+    const response = await graderProvider.invoke({
+      question: userPrompt,
+      systemPrompt,
+      evalCaseId: context2.evalCase.id,
+      attempt: context2.attempt,
+      maxOutputTokens: this.maxOutputTokens,
+      temperature: this.temperature
+    });
+    return {
+      text: extractLastAssistantContent(response.output),
+      providerResponse: response,
+      tokenUsage: response.tokenUsage
+    };
   }
 };
 function buildOutputSchema() {
@@ -15028,6 +15639,29 @@ function buildOutputSchema() {
     "}"
   ].join("\n");
 }
+function buildStructureRepairPrompt(options) {
+  const { validationError, invalidResponse } = options;
+  return [
+    "The following evaluation response has useful grading content but invalid JSON structure.",
+    "Repair it to satisfy the schema in the system prompt.",
+    "Preserve the evaluation meaning, do not re-grade the answer, and return only a single JSON object.",
+    "",
+    "Validation error:",
+    validationError,
+    "",
+    "Invalid response:",
+    invalidResponse
+  ].join("\n");
+}
+function sumTokenUsage(first, second) {
+  if (!first && !second) {
+    return void 0;
+  }
+  return {
+    input: (first?.input ?? 0) + (second?.input ?? 0),
+    output: (first?.output ?? 0) + (second?.output ?? 0)
+  };
+}
 function buildRubricOutputSchema() {
   return `You are an expert evaluator. Evaluate the candidate answer against each rubric item.
 You must return a valid JSON object matching this schema:
@@ -15127,19 +15761,21 @@ function calculateScoreRangeResult(result, rubrics) {
     rawScores[rubric.id] = rawScore;
     totalWeight += rubric.weight;
     weightedScoreSum += normalizedScore * rubric.weight;
-    let requiredMinScore;
-    if (rubric.required_min_score !== void 0) {
-      requiredMinScore = rubric.required_min_score;
+    let minScoreThreshold;
+    if (rubric.min_score !== void 0) {
+      minScoreThreshold = rubric.min_score;
+    } else if (rubric.required_min_score !== void 0) {
+      minScoreThreshold = rubric.required_min_score / 10;
     } else if (rubric.required === true) {
-      requiredMinScore = 10;
+      minScoreThreshold = 1;
     }
     const matchingRange = rubric.score_ranges?.find(
       (r) => rawScore >= r.score_range[0] && rawScore <= r.score_range[1]
     );
     const rangeDescription = matchingRange?.outcome ?? "";
     const criterionLabel = rubric.outcome ?? rubric.id;
-    const passed = !(requiredMinScore !== void 0 && rawScore < requiredMinScore) && rawScore >= 7;
-    if (requiredMinScore !== void 0 && rawScore < requiredMinScore) {
+    const passed = !(minScoreThreshold !== void 0 && normalizedScore < minScoreThreshold) && rawScore >= 7;
+    if (minScoreThreshold !== void 0 && normalizedScore < minScoreThreshold) {
       failedRequired = true;
     }
     assertions.push({
@@ -15216,11 +15852,11 @@ function createFilesystemTools(workspacePath) {
       execute: async (input) => {
         try {
           const resolved = resolveSandboxed(workspacePath, input.path);
-          const stat10 = await import_promises29.default.stat(resolved);
-          if (stat10.isDirectory()) {
+          const stat11 = await import_promises29.default.stat(resolved);
+          if (stat11.isDirectory()) {
             return { error: `'${input.path}' is a directory, not a file` };
           }
-          const buffer = Buffer.alloc(Math.min(stat10.size, MAX_FILE_SIZE));
+          const buffer = Buffer.alloc(Math.min(stat11.size, MAX_FILE_SIZE));
           const fd = await import_promises29.default.open(resolved, "r");
           try {
             await fd.read(buffer, 0, buffer.length, 0);
@@ -15228,8 +15864,8 @@ function createFilesystemTools(workspacePath) {
             await fd.close();
           }
           const content = buffer.toString("utf-8");
-          const truncated = stat10.size > MAX_FILE_SIZE;
-          return { content, truncated, size: stat10.size };
+          const truncated = stat11.size > MAX_FILE_SIZE;
+          return { content, truncated, size: stat11.size };
         } catch (error) {
           return { error: error instanceof Error ? error.message : String(error) };
         }
@@ -15280,8 +15916,8 @@ async function searchDirectory(dirPath, workspacePath, regex, matches) {
       const ext = import_node_path39.default.extname(entry.name).toLowerCase();
       if (BINARY_EXTENSIONS.has(ext)) continue;
       try {
-        const stat10 = await import_promises29.default.stat(fullPath);
-        if (stat10.size > MAX_FILE_SIZE) continue;
+        const stat11 = await import_promises29.default.stat(fullPath);
+        if (stat11.size > MAX_FILE_SIZE) continue;
         const content = await import_promises29.default.readFile(fullPath, "utf-8");
         const lines = content.split("\n");
         for (let i = 0; i < lines.length; i++) {
@@ -15925,115 +16561,115 @@ var FieldAccuracyEvaluator = class {
    * Evaluate a single field against the expected value.
    */
   evaluateField(fieldConfig, candidateData, expectedData) {
-    const { path: path52, match, required = true, weight = 1 } = fieldConfig;
-    const candidateValue = resolvePath(candidateData, path52);
-    const expectedValue = resolvePath(expectedData, path52);
+    const { path: path53, match, required = true, weight = 1 } = fieldConfig;
+    const candidateValue = resolvePath(candidateData, path53);
+    const expectedValue = resolvePath(expectedData, path53);
     if (expectedValue === void 0) {
       return {
-        path: path52,
+        path: path53,
         score: 1,
         // No expected value means no comparison needed
         weight,
         hit: true,
-        message: `${path52}: no expected value`
+        message: `${path53}: no expected value`
       };
     }
     if (candidateValue === void 0) {
       if (required) {
         return {
-          path: path52,
+          path: path53,
           score: 0,
           weight,
           hit: false,
-          message: `${path52} (required, missing)`
+          message: `${path53} (required, missing)`
         };
       }
       return {
-        path: path52,
+        path: path53,
         score: 1,
         // Don't penalize missing optional fields
         weight: 0,
         // Zero weight means it won't affect the score
         hit: true,
-        message: `${path52}: optional field missing`
+        message: `${path53}: optional field missing`
       };
     }
     switch (match) {
       case "exact":
-        return this.compareExact(path52, candidateValue, expectedValue, weight);
+        return this.compareExact(path53, candidateValue, expectedValue, weight);
       case "numeric_tolerance":
         return this.compareNumericTolerance(
-          path52,
+          path53,
           candidateValue,
           expectedValue,
           fieldConfig,
           weight
         );
       case "date":
-        return this.compareDate(path52, candidateValue, expectedValue, fieldConfig, weight);
+        return this.compareDate(path53, candidateValue, expectedValue, fieldConfig, weight);
       default:
         return {
-          path: path52,
+          path: path53,
           score: 0,
           weight,
           hit: false,
-          message: `${path52}: unknown match type "${match}"`
+          message: `${path53}: unknown match type "${match}"`
         };
     }
   }
   /**
    * Exact equality comparison.
    */
-  compareExact(path52, candidateValue, expectedValue, weight) {
+  compareExact(path53, candidateValue, expectedValue, weight) {
     if (deepEqual(candidateValue, expectedValue)) {
       return {
-        path: path52,
+        path: path53,
         score: 1,
         weight,
         hit: true,
-        message: path52
+        message: path53
       };
     }
     if (typeof candidateValue !== typeof expectedValue) {
       return {
-        path: path52,
+        path: path53,
         score: 0,
         weight,
         hit: false,
-        message: `${path52} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
+        message: `${path53} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
       };
     }
     return {
-      path: path52,
+      path: path53,
       score: 0,
       weight,
       hit: false,
-      message: `${path52} (value mismatch)`
+      message: `${path53} (value mismatch)`
     };
   }
   /**
    * Numeric comparison with absolute or relative tolerance.
    */
-  compareNumericTolerance(path52, candidateValue, expectedValue, fieldConfig, weight) {
+  compareNumericTolerance(path53, candidateValue, expectedValue, fieldConfig, weight) {
     const { tolerance = 0, relative = false } = fieldConfig;
     const candidateNum = toNumber(candidateValue);
     const expectedNum = toNumber(expectedValue);
     if (candidateNum === null || expectedNum === null) {
       return {
-        path: path52,
+        path: path53,
         score: 0,
         weight,
         hit: false,
-        message: `${path52} (non-numeric value)`
+        message: `${path53} (non-numeric value)`
       };
     }
     if (!Number.isFinite(candidateNum) || !Number.isFinite(expectedNum)) {
       return {
-        path: path52,
+        path: path53,
         score: 0,
         weight,
         hit: false,
-        message: `${path52} (invalid numeric value)`
+        message: `${path53} (invalid numeric value)`
       };
     }
     const diff = Math.abs(candidateNum - expectedNum);
@@ -16046,61 +16682,61 @@ var FieldAccuracyEvaluator = class {
     }
     if (withinTolerance) {
       return {
-        path: path52,
+        path: path53,
         score: 1,
         weight,
         hit: true,
-        message: `${path52} (within tolerance: diff=${diff.toFixed(2)})`
+        message: `${path53} (within tolerance: diff=${diff.toFixed(2)})`
       };
     }
     return {
-      path: path52,
+      path: path53,
       score: 0,
       weight,
       hit: false,
-      message: `${path52} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
+      message: `${path53} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
     };
   }
   /**
    * Date comparison with format normalization.
    */
-  compareDate(path52, candidateValue, expectedValue, fieldConfig, weight) {
+  compareDate(path53, candidateValue, expectedValue, fieldConfig, weight) {
     const formats = fieldConfig.formats ?? DEFAULT_DATE_FORMATS;
     const candidateDate = parseDate(String(candidateValue), formats);
     const expectedDate = parseDate(String(expectedValue), formats);
     if (candidateDate === null) {
       return {
-        path: path52,
+        path: path53,
         score: 0,
         weight,
         hit: false,
-        message: `${path52} (unparseable candidate date)`
+        message: `${path53} (unparseable candidate date)`
       };
     }
     if (expectedDate === null) {
       return {
-        path: path52,
+        path: path53,
         score: 0,
         weight,
         hit: false,
-        message: `${path52} (unparseable expected date)`
+        message: `${path53} (unparseable expected date)`
       };
     }
     if (candidateDate.getFullYear() === expectedDate.getFullYear() && candidateDate.getMonth() === expectedDate.getMonth() && candidateDate.getDate() === expectedDate.getDate()) {
       return {
-        path: path52,
+        path: path53,
         score: 1,
         weight,
         hit: true,
-        message: path52
+        message: path53
       };
     }
     return {
-      path: path52,
+      path: path53,
       score: 0,
       weight,
       hit: false,
-      message: `${path52} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
+      message: `${path53} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
     };
   }
   /**
@@ -16133,11 +16769,11 @@ var FieldAccuracyEvaluator = class {
     };
   }
 };
-function resolvePath(obj, path52) {
-  if (!path52 || !obj) {
+function resolvePath(obj, path53) {
+  if (!path53 || !obj) {
     return void 0;
   }
-  const parts = path52.split(/\.|\[|\]/).filter((p) => p.length > 0);
+  const parts = path53.split(/\.|\[|\]/).filter((p) => p.length > 0);
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0) {
@@ -16634,8 +17270,8 @@ var TokenUsageEvaluator = class {
 
 // src/evaluation/evaluators/tool-trajectory.ts
 init_cjs_shims();
-function getNestedValue(obj, path52) {
-  const parts = path52.split(".");
+function getNestedValue(obj, path53) {
+  const parts = path53.split(".");
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0 || typeof current !== "object") {
@@ -18428,7 +19064,7 @@ var WorkspacePoolManager = class {
   }
   /**
    * Reset an existing slot for reuse:
-   * 1. Reset repos (git reset --hard {ref} && git clean -fd per repo)
+   * 1. Reset repos (fetch from origin when resolve=remote, then git reset --hard && git clean per repo)
    * 2. Re-copy template files (skip repo directories)
    */
   async resetSlot(slotPath, templatePath, repos, poolReset = "fast") {
@@ -18441,7 +19077,17 @@ var WorkspacePoolManager = class {
         continue;
       }
       const ref = repo.checkout?.ref ?? "HEAD";
-      await git(["reset", "--hard", ref], { cwd: repoDir });
+      const resolve = repo.checkout?.resolve ?? "remote";
+      if (resolve === "remote") {
+        const fetchArgs = ["fetch", "origin", ref];
+        if (repo.clone?.depth) {
+          fetchArgs.splice(1, 0, "--depth", String(repo.clone.depth));
+        }
+        await git(fetchArgs, { cwd: repoDir });
+        await git(["reset", "--hard", "FETCH_HEAD"], { cwd: repoDir });
+      } else {
+        await git(["reset", "--hard", ref], { cwd: repoDir });
+      }
       const cleanFlag = poolReset === "strict" ? "-fdx" : "-fd";
       await git(["clean", cleanFlag], { cwd: repoDir });
     }
@@ -18741,7 +19387,7 @@ async function executeWorkspaceScript(config, context2, failureMode = "fatal") {
 }
 
 // src/evaluation/orchestrator.ts
-function classifyQualityStatus(score, threshold = PASS_THRESHOLD) {
+function classifyQualityStatus(score, threshold = DEFAULT_THRESHOLD) {
   return score >= threshold ? "ok" : "quality_failure";
 }
 function buildSkippedEvaluatorError(scores) {
@@ -18833,7 +19479,7 @@ async function runEvaluation(options) {
   const filteredEvalCases = filterEvalCases(evalCases, filter);
   if (filteredEvalCases.length === 0) {
     if (filter) {
-      throw new Error(`No tests matched filter '${filter}' in ${evalFilePath}`);
+      throw new Error(`No tests matched filter '${formatFilter(filter)}' in ${evalFilePath}`);
     }
     return [];
   }
@@ -18859,20 +19505,10 @@ async function runEvaluation(options) {
     if (resolvedTargetsByName.has(name)) {
       return resolvedTargetsByName.get(name);
     }
-    let definition = targetDefinitions.get(name);
+    const definition = resolveDelegatedTargetDefinition(name, targetDefinitions, envLookup);
     if (!definition) {
       return void 0;
     }
-    for (let depth = 0; depth < 5; depth++) {
-      const useTarget = definition.use_target;
-      if (typeof useTarget !== "string" || useTarget.trim().length === 0) break;
-      const envMatch = useTarget.trim().match(/^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/i);
-      const resolvedName = envMatch ? envLookup[envMatch[1]] ?? "" : useTarget.trim();
-      if (resolvedName.length === 0) break;
-      const next = targetDefinitions.get(resolvedName);
-      if (!next) break;
-      definition = next;
-    }
     const resolved = resolveTargetDefinition(definition, envLookup, evalFilePath);
     resolvedTargetsByName.set(name, resolved);
     return resolved;
@@ -18895,6 +19531,9 @@ async function runEvaluation(options) {
     const graderName = targetContext.graderTarget ?? targetContext.name;
     const resolvedGrader = resolveTargetByName(graderName);
     if (!resolvedGrader) {
+      if (!LLM_GRADER_CAPABLE_KINDS.includes(targetContext.kind)) {
+        return void 0;
+      }
       return getOrCreateProvider(targetContext);
     }
     return getOrCreateProvider(resolvedGrader);
@@ -19225,7 +19864,7 @@ async function runEvaluation(options) {
           const budgetResult = {
             timestamp: (now ?? (() => /* @__PURE__ */ new Date()))().toISOString(),
             testId: evalCase.id,
-            dataset: evalCase.dataset,
+            suite: evalCase.suite,
             category: evalCase.category,
             score: 0,
             assertions: [],
@@ -19262,7 +19901,7 @@ async function runEvaluation(options) {
           const haltResult = {
             timestamp: (now ?? (() => /* @__PURE__ */ new Date()))().toISOString(),
             testId: evalCase.id,
-            dataset: evalCase.dataset,
+            suite: evalCase.suite,
             category: evalCase.category,
             score: 0,
             assertions: [],
@@ -19574,7 +20213,7 @@ async function runBatchEvaluation(options) {
         targetResolver,
         availableTargets,
         verbose,
-        threshold: batchThreshold
+        threshold: evalCase.threshold ?? batchThreshold
       });
       if (providerError) {
         result = {
@@ -20036,8 +20675,9 @@ async function runEvalCase(options) {
       fileChanges,
       workspacePath,
       verbose,
-      threshold: caseThreshold
+      threshold: evalCase.threshold ?? caseThreshold
     });
+    const effectiveThreshold = evalCase.threshold ?? caseThreshold;
     const totalDurationMs = Date.now() - caseStartMs;
     const graderTokens = aggregateEvaluatorTokenUsage(result.scores);
     const evalRunTokenUsage = tokenUsage || graderTokens ? {
@@ -20051,7 +20691,7 @@ async function runEvalCase(options) {
       ...evalRunTokenUsage ? { tokenUsage: evalRunTokenUsage } : {}
     };
     const skippedEvaluatorError = buildSkippedEvaluatorError(result.scores);
-    const executionStatus = providerError || skippedEvaluatorError ? "execution_error" : classifyQualityStatus(result.score, caseThreshold);
+    const executionStatus = providerError || skippedEvaluatorError ? "execution_error" : classifyQualityStatus(result.score, effectiveThreshold);
     const targetUsedField = targetUsed ? { targetUsed } : {};
     const finalResult = providerError ? {
       ...result,
@@ -20252,7 +20892,8 @@ async function evaluateCandidate(options) {
     targetResolver,
     availableTargets,
     fileChanges,
-    workspacePath
+    workspacePath,
+    threshold: evalThreshold
   });
   const completedAt = nowFn();
   let agentRequest;
@@ -20283,7 +20924,7 @@ async function evaluateCandidate(options) {
   return {
     timestamp: completedAt.toISOString(),
     testId: evalCase.id,
-    dataset: evalCase.dataset,
+    suite: evalCase.suite,
     category: evalCase.category,
     conversationId: evalCase.conversation_id,
     score: score.score,
@@ -20326,7 +20967,8 @@ async function runEvaluatorsForCase(options) {
     targetResolver,
     availableTargets,
     fileChanges,
-    workspacePath
+    workspacePath,
+    threshold
   } = options;
   if (evalCase.assertions && evalCase.assertions.length > 0) {
     return runEvaluatorList({
@@ -20352,7 +20994,8 @@ async function runEvaluatorsForCase(options) {
       targetResolver,
       availableTargets,
       fileChanges,
-      workspacePath
+      workspacePath,
+      threshold
     });
   }
   const evaluatorKind = evalCase.evaluator ?? "llm-grader";
@@ -20454,7 +21097,8 @@ async function runEvaluatorList(options) {
         name: evaluatorConfig.name,
         type: evaluatorConfig.type,
         weight,
-        ...evaluatorConfig.required !== void 0 ? { required: evaluatorConfig.required } : {}
+        ...evaluatorConfig.required !== void 0 ? { required: evaluatorConfig.required } : {},
+        ...evaluatorConfig.min_score !== void 0 ? { min_score: evaluatorConfig.min_score } : {}
       });
       scores.push({
         name: evaluatorConfig.name,
@@ -20489,7 +21133,8 @@ async function runEvaluatorList(options) {
         name: evaluatorConfig.name ?? "unknown",
         type: evaluatorConfig.type ?? "llm-grader",
         weight,
-        ...evaluatorConfig.required !== void 0 ? { required: evaluatorConfig.required } : {}
+        ...evaluatorConfig.required !== void 0 ? { required: evaluatorConfig.required } : {},
+        ...evaluatorConfig.min_score !== void 0 ? { min_score: evaluatorConfig.min_score } : {}
       });
       scores.push({
         name: evaluatorConfig.name ?? "unknown",
@@ -20523,9 +21168,10 @@ async function runEvaluatorList(options) {
       }
     }
   }
+  const effectiveThreshold = options.threshold ?? DEFAULT_THRESHOLD;
   const hasRequiredFailure = scored.some((entry) => {
     if (!entry.required) return false;
-    const minScore = typeof entry.required === "number" ? entry.required : PASS_THRESHOLD;
+    const minScore = entry.min_score ?? (typeof entry.required === "number" ? entry.required : effectiveThreshold);
     return entry.score.score < minScore;
   });
   const scorable = scored.filter((entry) => entry.score.verdict !== "skip");
@@ -20536,17 +21182,23 @@ async function runEvaluatorList(options) {
   const expectedAspectCount = assertions.length || 1;
   const score = {
     score: aggregateScore,
-    verdict: scoreToVerdict(aggregateScore),
+    verdict: scoreToVerdict(aggregateScore, effectiveThreshold),
     assertions,
     expectedAspectCount
   };
   return { score, scores };
 }
+function formatFilter(filter) {
+  return typeof filter === "string" ? filter : filter.join(", ");
+}
+function matchesFilter3(id, filter) {
+  return typeof filter === "string" ? import_micromatch3.default.isMatch(id, filter) : filter.some((pattern) => import_micromatch3.default.isMatch(id, pattern));
+}
 function filterEvalCases(evalCases, filter) {
   if (!filter) {
     return evalCases;
   }
-  return evalCases.filter((evalCase) => import_micromatch3.default.isMatch(evalCase.id, filter));
+  return evalCases.filter((evalCase) => matchesFilter3(evalCase.id, filter));
 }
 function buildEvaluatorRegistry(overrides, resolveGraderProvider) {
   const llmGrader = overrides?.["llm-grader"] ?? new LlmGraderEvaluator({
@@ -20633,7 +21285,7 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
   return {
     timestamp: timestamp.toISOString(),
     testId: evalCase.id,
-    dataset: evalCase.dataset,
+    suite: evalCase.suite,
     category: evalCase.category,
     conversationId: evalCase.conversation_id,
     score: 0,
@@ -20907,6 +21559,7 @@ async function evaluate(config) {
     verbose: config.verbose,
     maxConcurrency: config.workers ?? 3,
     filter: config.filter,
+    threshold: config.threshold,
     evalCases,
     onResult: async (result) => {
       collectedResults.push(result);
@@ -20917,19 +21570,19 @@ async function evaluate(config) {
   const durationMs = Date.now() - startTime;
   return {
     results: allResults,
-    summary: computeSummary(allResults, durationMs)
+    summary: computeSummary(allResults, durationMs, config.threshold)
   };
 }
 function mapAssertionType(type) {
   return type.replace(/_/g, "-");
 }
-function computeSummary(results, durationMs) {
+function computeSummary(results, durationMs, threshold = DEFAULT_THRESHOLD) {
   const total = results.length;
   let passed = 0;
   let scoreSum = 0;
   for (const r of results) {
     scoreSum += r.score;
-    if (r.score >= PASS_THRESHOLD) {
+    if (r.score >= threshold) {
       passed++;
     }
   }
@@ -20960,7 +21613,7 @@ async function discoverDefaultTarget(repoRoot) {
   return null;
 }
 async function loadEnvHierarchy(repoRoot, startPath) {
-  const { readFileSync: readFileSync3 } = await import("fs");
+  const { readFileSync: readFileSync4 } = await import("fs");
   const chain = buildDirectoryChain2(startPath, repoRoot);
   const envFiles = [];
   for (const dir of chain) {
@@ -20969,7 +21622,7 @@ async function loadEnvHierarchy(repoRoot, startPath) {
   }
   for (let i = 0; i < envFiles.length; i++) {
     try {
-      const content = readFileSync3(envFiles[i], "utf8");
+      const content = readFileSync4(envFiles[i], "utf8");
       for (const line of content.split("\n")) {
         const trimmed = line.trim();
         if (!trimmed || trimmed.startsWith("#")) continue;
@@ -21043,7 +21696,7 @@ var CONFIG_FILE_NAMES = [
 ];
 async function loadTsConfig(projectRoot) {
   const { existsSync: existsSync7 } = await import("fs");
-  const { pathToFileURL } = await import("url");
+  const { pathToFileURL: pathToFileURL2 } = await import("url");
   const { join: join2 } = await import("path");
   for (const fileName of CONFIG_FILE_NAMES) {
     const filePath = join2(projectRoot, fileName);
@@ -21051,7 +21704,7 @@ async function loadTsConfig(projectRoot) {
       continue;
     }
     try {
-      const fileUrl = pathToFileURL(filePath).href;
+      const fileUrl = pathToFileURL2(filePath).href;
       const mod = await import(fileUrl);
       const config = mod.default ?? mod;
       return AgentVConfigSchema.parse(config);
@@ -21492,7 +22145,7 @@ var OtelTraceExporter = class {
         rootSpan.setAttribute("gen_ai.system", "agentv");
         rootSpan.setAttribute("agentv.test_id", result.testId);
         rootSpan.setAttribute("agentv.target", result.target);
-        if (result.dataset) rootSpan.setAttribute("agentv.dataset", result.dataset);
+        if (result.suite) rootSpan.setAttribute("agentv.suite", result.suite);
         rootSpan.setAttribute("agentv.score", result.score);
         if (captureContent && result.output.length > 0) {
           const lastMsg = result.output[result.output.length - 1];
@@ -21701,7 +22354,7 @@ var OtelStreamingObserver = class {
     this.rootSpan.setAttribute("gen_ai.system", "agentv");
     this.rootSpan.setAttribute("agentv.test_id", testId);
     this.rootSpan.setAttribute("agentv.target", target);
-    if (evalSet) this.rootSpan.setAttribute("agentv.dataset", evalSet);
+    if (evalSet) this.rootSpan.setAttribute("agentv.suite", evalSet);
     this.rootCtx = this.api.trace.setSpan(this.api.context.active(), this.rootSpan);
   }
   /** Create and immediately export a tool span */
@@ -22057,12 +22710,244 @@ function extractToolResultContent(content) {
   return parts.length > 0 ? parts.join("") : void 0;
 }
 
-// src/import/session-discovery.ts
+// src/import/codex-parser.ts
+init_cjs_shims();
+function parseCodexSession(jsonl) {
+  const messages = [];
+  let sessionId = "";
+  let cwd;
+  let model;
+  let version;
+  let startTimestamp;
+  let endTimestamp;
+  const pendingCalls = /* @__PURE__ */ new Map();
+  const lines = jsonl.split("\n").filter((l) => l.trim().length > 0);
+  for (const line of lines) {
+    let entry;
+    try {
+      entry = JSON.parse(line);
+    } catch {
+      continue;
+    }
+    if (!entry.type) continue;
+    if (entry.timestamp) {
+      if (!startTimestamp) startTimestamp = entry.timestamp;
+      endTimestamp = entry.timestamp;
+    }
+    const payload = entry.payload ?? {};
+    switch (entry.type) {
+      case "session_meta": {
+        sessionId = String(payload.id ?? "");
+        cwd = payload.cwd ? String(payload.cwd) : void 0;
+        version = payload.cli_version ? String(payload.cli_version) : void 0;
+        if (payload.model && !model) {
+          model = String(payload.model);
+        }
+        break;
+      }
+      case "turn_context": {
+        if (payload.model && !model) {
+          model = String(payload.model);
+        }
+        if (payload.cwd && !cwd) {
+          cwd = String(payload.cwd);
+        }
+        break;
+      }
+      case "response_item": {
+        const itemType = String(payload.type ?? "");
+        const role = String(payload.role ?? "");
+        switch (itemType) {
+          case "message": {
+            if (role === "developer") break;
+            const content = extractResponseItemContent(payload.content);
+            if (role === "user" && content) {
+              messages.push({ role: "user", content });
+            } else if (role === "assistant" && content) {
+              messages.push({ role: "assistant", content });
+            }
+            break;
+          }
+          case "function_call": {
+            const toolName = String(payload.name ?? "");
+            const callId = String(payload.call_id ?? "");
+            let input;
+            if (typeof payload.arguments === "string") {
+              try {
+                input = JSON.parse(payload.arguments);
+              } catch {
+                input = payload.arguments;
+              }
+            } else {
+              input = payload.arguments;
+            }
+            const toolCall = { tool: toolName, input, id: callId };
+            const msgIdx = messages.length;
+            messages.push({
+              role: "assistant",
+              toolCalls: [toolCall]
+            });
+            if (callId) {
+              pendingCalls.set(callId, { msgIdx, toolIdx: 0 });
+            }
+            break;
+          }
+          case "custom_tool_call": {
+            const toolName = String(payload.name ?? "");
+            const callId = String(payload.call_id ?? "");
+            let input;
+            if (typeof payload.arguments === "string") {
+              try {
+                input = JSON.parse(payload.arguments);
+              } catch {
+                input = payload.arguments;
+              }
+            } else {
+              input = payload.arguments;
+            }
+            const toolCall = { tool: toolName, input, id: callId };
+            const msgIdx = messages.length;
+            messages.push({
+              role: "assistant",
+              toolCalls: [toolCall]
+            });
+            if (callId) {
+              pendingCalls.set(callId, { msgIdx, toolIdx: 0 });
+            }
+            break;
+          }
+          case "function_call_output":
+          case "custom_tool_call_output": {
+            const callId = String(payload.call_id ?? "");
+            const pending = pendingCalls.get(callId);
+            if (pending) {
+              const existingMsg = messages[pending.msgIdx];
+              const existingCalls = [...existingMsg.toolCalls ?? []];
+              existingCalls[pending.toolIdx] = {
+                ...existingCalls[pending.toolIdx],
+                output: payload.output
+              };
+              messages[pending.msgIdx] = { ...existingMsg, toolCalls: existingCalls };
+              pendingCalls.delete(callId);
+            }
+            break;
+          }
+          // Skip reasoning blocks (thinking tokens)
+          case "reasoning":
+            break;
+        }
+        break;
+      }
+    }
+  }
+  let durationMs;
+  if (startTimestamp && endTimestamp) {
+    durationMs = new Date(endTimestamp).getTime() - new Date(startTimestamp).getTime();
+  }
+  const source = {
+    provider: "codex",
+    sessionId,
+    cwd,
+    startedAt: startTimestamp,
+    model,
+    version
+  };
+  return {
+    messages,
+    source,
+    // Codex rollout files don't include token counts (only rate limit info)
+    tokenUsage: void 0,
+    durationMs,
+    costUsd: null
+  };
+}
+function extractResponseItemContent(content) {
+  if (typeof content === "string") return content;
+  if (!Array.isArray(content)) return void 0;
+  const parts = [];
+  for (const block of content) {
+    if (typeof block === "object" && block !== null) {
+      const b = block;
+      if (typeof b.text === "string") {
+        parts.push(b.text);
+      }
+    }
+  }
+  return parts.length > 0 ? parts.join("") : void 0;
+}
+
+// src/import/codex-session-discovery.ts
 init_cjs_shims();
 var import_promises36 = require("fs/promises");
 var import_node_os8 = require("os");
 var import_node_path53 = __toESM(require("path"), 1);
-var DEFAULT_PROJECTS_DIR = () => import_node_path53.default.join((0, import_node_os8.homedir)(), ".claude", "projects");
+var DEFAULT_SESSIONS_DIR = () => import_node_path53.default.join((0, import_node_os8.homedir)(), ".codex", "sessions");
+async function discoverCodexSessions(opts) {
+  const sessionsDir = opts?.sessionsDir ?? DEFAULT_SESSIONS_DIR();
+  const limit = opts?.latest ? 1 : opts?.limit ?? 10;
+  const sessions = [];
+  let yearDirs;
+  try {
+    yearDirs = await (0, import_promises36.readdir)(sessionsDir);
+  } catch {
+    return [];
+  }
+  for (const year of yearDirs) {
+    const yearPath = import_node_path53.default.join(sessionsDir, year);
+    let monthDirs;
+    try {
+      monthDirs = await (0, import_promises36.readdir)(yearPath);
+    } catch {
+      continue;
+    }
+    for (const month of monthDirs) {
+      const monthPath = import_node_path53.default.join(yearPath, month);
+      let dayDirs;
+      try {
+        dayDirs = await (0, import_promises36.readdir)(monthPath);
+      } catch {
+        continue;
+      }
+      for (const day of dayDirs) {
+        if (opts?.date) {
+          const dirDate = `${year}-${month}-${day}`;
+          if (dirDate !== opts.date) continue;
+        }
+        const dayPath = import_node_path53.default.join(monthPath, day);
+        let files;
+        try {
+          files = await (0, import_promises36.readdir)(dayPath);
+        } catch {
+          continue;
+        }
+        for (const file of files) {
+          if (!file.startsWith("rollout-") || !file.endsWith(".jsonl")) continue;
+          const filePath = import_node_path53.default.join(dayPath, file);
+          const nameWithoutExt = file.replace(/\.jsonl$/, "");
+          const parts = nameWithoutExt.split("-");
+          const sessionId = parts.length >= 6 ? parts.slice(-5).join("-") : nameWithoutExt;
+          let updatedAt;
+          try {
+            const fileStat = await (0, import_promises36.stat)(filePath);
+            updatedAt = fileStat.mtime;
+          } catch {
+            updatedAt = /* @__PURE__ */ new Date(0);
+          }
+          sessions.push({ sessionId, filePath, filename: file, updatedAt });
+        }
+      }
+    }
+  }
+  sessions.sort((a, b) => b.updatedAt.getTime() - a.updatedAt.getTime());
+  return sessions.slice(0, limit);
+}
+
+// src/import/session-discovery.ts
+init_cjs_shims();
+var import_promises37 = require("fs/promises");
+var import_node_os9 = require("os");
+var import_node_path54 = __toESM(require("path"), 1);
+var DEFAULT_PROJECTS_DIR = () => import_node_path54.default.join((0, import_node_os9.homedir)(), ".claude", "projects");
 function encodeProjectPath(projectPath) {
   return projectPath.replace(/\//g, "-");
 }
@@ -22071,7 +22956,7 @@ async function discoverClaudeSessions(opts) {
   const limit = opts?.latest ? 1 : opts?.limit ?? 10;
   let projectDirs;
   try {
-    projectDirs = await (0, import_promises36.readdir)(projectsDir);
+    projectDirs = await (0, import_promises37.readdir)(projectsDir);
   } catch {
     return [];
   }
@@ -22081,10 +22966,10 @@ async function discoverClaudeSessions(opts) {
   }
   const sessions = [];
   for (const projectDir of projectDirs) {
-    const dirPath = import_node_path53.default.join(projectsDir, projectDir);
+    const dirPath = import_node_path54.default.join(projectsDir, projectDir);
     let entries;
     try {
-      entries = await (0, import_promises36.readdir)(dirPath);
+      entries = await (0, import_promises37.readdir)(dirPath);
     } catch {
       continue;
     }
@@ -22092,10 +22977,10 @@ async function discoverClaudeSessions(opts) {
       if (!entry.endsWith(".jsonl")) continue;
       const sessionId = entry.replace(/\.jsonl$/, "");
       if (opts?.sessionId && sessionId !== opts.sessionId) continue;
-      const filePath = import_node_path53.default.join(dirPath, entry);
+      const filePath = import_node_path54.default.join(dirPath, entry);
       let updatedAt;
       try {
-        const fileStat = await (0, import_promises36.stat)(filePath);
+        const fileStat = await (0, import_promises37.stat)(filePath);
         updatedAt = fileStat.mtime;
       } catch {
         updatedAt = /* @__PURE__ */ new Date(0);
@@ -22112,13 +22997,91 @@ async function discoverClaudeSessions(opts) {
   return sessions.slice(0, limit);
 }
 
+// src/import/transcript-provider.ts
+init_cjs_shims();
+
 // src/import/types.ts
 init_cjs_shims();
-var import_promises37 = require("fs/promises");
+var import_promises38 = require("fs/promises");
+function toTranscriptJsonLine(entry) {
+  const firstUserMessage = entry.messages.find((m) => m.role === "user");
+  const input = typeof firstUserMessage?.content === "string" ? firstUserMessage.content : "";
+  return {
+    input,
+    output: entry.messages,
+    token_usage: entry.tokenUsage ? {
+      input: entry.tokenUsage.input,
+      output: entry.tokenUsage.output,
+      cached: entry.tokenUsage.cached
+    } : void 0,
+    duration_ms: entry.durationMs,
+    cost_usd: entry.costUsd,
+    source: {
+      provider: entry.source.provider,
+      session_id: entry.source.sessionId,
+      model: entry.source.model,
+      timestamp: entry.source.startedAt,
+      git_branch: entry.source.gitBranch,
+      cwd: entry.source.cwd ?? entry.source.projectPath,
+      version: entry.source.version
+    }
+  };
+}
+async function readTranscriptJsonl(filePath) {
+  const text = await (0, import_promises38.readFile)(filePath, "utf8");
+  return text.split("\n").filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
+}
 async function readTranscriptFile(filePath) {
-  return (0, import_promises37.readFile)(filePath, "utf8");
+  return (0, import_promises38.readFile)(filePath, "utf8");
 }
 
+// src/import/transcript-provider.ts
+var TranscriptProvider = class _TranscriptProvider {
+  id;
+  kind = "transcript";
+  targetName;
+  lines;
+  cursor = 0;
+  constructor(targetName, lines) {
+    this.targetName = targetName;
+    this.id = `transcript:${targetName}`;
+    this.lines = lines;
+  }
+  /**
+   * Create a TranscriptProvider from a JSONL file path.
+   */
+  static async fromFile(filePath) {
+    const lines = await readTranscriptJsonl(filePath);
+    if (lines.length === 0) {
+      throw new Error(`Transcript file is empty: ${filePath}`);
+    }
+    const providerName = lines[0].source.provider ?? "transcript";
+    return new _TranscriptProvider(providerName, lines);
+  }
+  get lineCount() {
+    return this.lines.length;
+  }
+  async invoke(_request) {
+    if (this.cursor >= this.lines.length) {
+      throw new Error(
+        `Transcript exhausted: ${this.lines.length} line(s) available but ${this.cursor + 1} invocations attempted. Each transcript line maps to one test case.`
+      );
+    }
+    const line = this.lines[this.cursor++];
+    return {
+      output: line.output,
+      tokenUsage: line.token_usage ? {
+        input: line.token_usage.input,
+        output: line.token_usage.output,
+        cached: line.token_usage.cached
+      } : void 0,
+      durationMs: line.duration_ms,
+      costUsd: line.cost_usd ?? void 0,
+      startTime: line.source.timestamp
+    };
+  }
+};
+
 // src/index.ts
 function createAgentKernel() {
   return { status: "stub" };
@@ -22133,6 +23096,7 @@ function createAgentKernel() {
   DEFAULT_EVALUATOR_TEMPLATE,
   DEFAULT_EVAL_PATTERNS,
   DEFAULT_EXPLORATION_TOOLS,
+  DEFAULT_THRESHOLD,
   DeterministicAssertionEvaluator,
   EvaluatorRegistry,
   ExecutionMetricsEvaluator,
@@ -22154,6 +23118,7 @@ function createAgentKernel() {
   TemplateNotFoundError,
   TokenUsageEvaluator,
   ToolTrajectoryEvaluator,
+  TranscriptProvider,
   WorkspaceCreationError,
   WorkspacePoolManager,
   addProject,
@@ -22190,6 +23155,7 @@ function createAgentKernel() {
   detectFormat,
   discoverAssertions,
   discoverClaudeSessions,
+  discoverCodexSessions,
   discoverCopilotSessions,
   discoverGraders,
   discoverJudges,
@@ -22250,6 +23216,8 @@ function createAgentKernel() {
   normalizeLineEndings,
   parseAgentSkillsEvals,
   parseClaudeSession,
+  parseCodexSession,
+  parseCopilotEvents,
   parseJsonFromText,
   parseJsonSafe,
   readJsonFile,
@@ -22257,8 +23225,10 @@ function createAgentKernel() {
   readTestSuiteMetadata,
   readTextFile,
   readTranscriptFile,
+  readTranscriptJsonl,
   removeProject,
   resolveAndCreateProvider,
+  resolveDelegatedTargetDefinition,
   resolveFileReference,
   resolveTargetDefinition,
   resolveWorkspaceTemplate,
@@ -22288,6 +23258,7 @@ function createAgentKernel() {
   substituteVariables,
   toCamelCaseDeep,
   toSnakeCaseDeep,
+  toTranscriptJsonLine,
   tokensPerTool,
   touchProject,
   transpileEvalYaml,