@agentv/core 4.6.0 → 4.7.0

package/dist/index.js

@@ -1,5 +1,6 @@
 import {
   COMMON_TARGET_SETTINGS,
+  LLM_GRADER_CAPABLE_KINDS,
   TEST_MESSAGE_ROLES,
   buildDirectoryChain,
   buildSearchRoots,
@@ -21,9 +22,10 @@ import {
   normalizeLineEndings,
   readJsonFile,
   readTextFile,
+  resolveDelegatedTargetDefinition,
   resolveFileReference,
   resolveTargetDefinition
-} from "./chunk-AIQ5FO4G.js";
+} from "./chunk-75RFVESM.js";
 import {
   AgentvProvider
 } from "./chunk-PRNXHNLF.js";
@@ -855,8 +857,13 @@ 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) {
@@ -868,6 +875,7 @@ 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 } : {}
       });
@@ -937,7 +945,12 @@ 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",
@@ -963,6 +976,7 @@ 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 } : {}
@@ -1091,7 +1105,12 @@ 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",
@@ -1099,6 +1118,7 @@ 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;
@@ -1209,7 +1229,12 @@ 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",
@@ -1218,6 +1243,7 @@ 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 } : {}
       };
@@ -1280,7 +1306,12 @@ 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",
@@ -1288,6 +1319,7 @@ 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;
@@ -1301,13 +1333,19 @@ 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;
@@ -1321,13 +1359,19 @@ 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;
@@ -1359,13 +1403,19 @@ 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;
@@ -1411,13 +1461,19 @@ 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;
@@ -1431,7 +1487,12 @@ 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",
@@ -1439,6 +1500,7 @@ 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;
@@ -1450,13 +1512,19 @@ 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;
@@ -1470,13 +1538,19 @@ 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;
@@ -1488,13 +1562,19 @@ 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;
@@ -1508,13 +1588,19 @@ 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;
@@ -1526,13 +1612,19 @@ 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;
@@ -1545,7 +1637,12 @@ 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",
@@ -1553,18 +1650,25 @@ 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;
@@ -1576,13 +1680,19 @@ 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;
@@ -1618,7 +1728,12 @@ 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",
@@ -1626,6 +1741,7 @@ 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;
@@ -1695,7 +1811,12 @@ 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",
@@ -1703,12 +1824,18 @@ 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",
@@ -1719,6 +1846,7 @@ async function parseEvaluatorList(candidateEvaluators, searchRoots, evalId) {
       "weight",
       "config",
       "required",
+      "min_score",
       "negate",
       "max_steps",
       "maxSteps",
@@ -1748,6 +1876,7 @@ 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 } : {},
@@ -1879,10 +2008,23 @@ ${detailBlock}${ANSI_RESET4}`);
     console.warn(`${ANSI_YELLOW3}Warning: ${message}${ANSI_RESET4}`);
   }
 }
-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: ${rawRequired}' instead.`
+    );
+  }
+  return result;
 }
 function validateWeight(rawWeight, evaluatorName, evalId) {
   if (rawWeight === void 0) {
@@ -1925,16 +2067,30 @@ function parseRubricItems(rawRubrics, evaluatorName, evalId) {
     const id = asString(rawRubric.id) ?? `rubric-${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 min_score for rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': must be in (0, 1] (got ${ms})`
+        );
+      }
+      minScore = ms;
+      requiredMinScore = Math.round(ms * 10);
+    } else if (typeof rawRubric.required_min_score === "number") {
+      const rms = rawRubric.required_min_score;
+      if (!Number.isInteger(rms) || rms < 0 || rms > 10) {
         throw new Error(
-          `Invalid required_min_score for rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': must be an integer 0-10 (got ${minScore})`
+          `Invalid required_min_score for rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': must be an integer 0-10 (got ${rms})`
         );
       }
-      requiredMinScore = minScore;
+      requiredMinScore = rms;
+      minScore = rms / 10;
+      logWarning2(
+        `Rubric '${id}' in evaluator '${evaluatorName}' in '${evalId}': 'required_min_score: ${rms}' is deprecated. Use 'min_score: ${rms / 10}' (0-1 scale) instead.`
+      );
     }
     if (typeof rawRubric.required === "boolean") {
       required = rawRubric.required;
@@ -1954,6 +2110,7 @@ function parseRubricItems(rawRubrics, evaluatorName, evalId) {
         weight,
         ...expectedOutcome.length > 0 ? { outcome: expectedOutcome } : {},
         ...required !== void 0 ? { required } : {},
+        ...minScore !== void 0 ? { min_score: minScore } : {},
         ...requiredMinScore !== void 0 ? { required_min_score: requiredMinScore } : {},
         score_ranges: scoreRanges
       });
@@ -1970,6 +2127,7 @@ function parseRubricItems(rawRubrics, evaluatorName, evalId) {
         weight,
         // Default to required: true if not specified (backward compatibility)
         required: required ?? true,
+        ...minScore !== void 0 ? { min_score: minScore } : {},
         ...requiredMinScore !== void 0 ? { required_min_score: requiredMinScore } : {}
       });
     }
@@ -2098,12 +2256,22 @@ function parseInlineRubrics(rawRubrics) {
       id: asString(rubric.id) ?? `rubric-${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
       };
     }
@@ -2111,7 +2279,8 @@ 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) {
@@ -2511,6 +2680,9 @@ function resolveExpectedMessages(raw) {
 var ANSI_YELLOW5 = "\x1B[33m";
 var ANSI_RED2 = "\x1B[31m";
 var ANSI_RESET6 = "\x1B[0m";
+function matchesFilter(id, filter) {
+  return typeof filter === "string" ? micromatch.isMatch(id, filter) : filter.some((pattern) => micromatch.isMatch(id, pattern));
+}
 function detectFormat(filePath) {
   const ext = path6.extname(filePath).toLowerCase();
   if (ext === ".jsonl") return "jsonl";
@@ -2578,40 +2750,40 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
   const sidecar = await loadSidecarMetadata(absoluteTestPath, verbose);
   const rawFile = await readFile5(absoluteTestPath, "utf8");
   const rawCases = parseJsonlContent(rawFile, evalFilePath);
-  const fallbackEvalSet = path6.basename(absoluteTestPath, ".jsonl") || "eval";
-  const evalSetName = sidecar.name && sidecar.name.trim().length > 0 ? sidecar.name : fallbackEvalSet;
+  const fallbackSuiteName = path6.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: ${sidecar.description}`);
     }
   }
   const results = [];
   for (let lineIndex = 0; lineIndex < rawCases.length; lineIndex++) {
-    const evalcase = rawCases[lineIndex];
+    const testCaseConfig = rawCases[lineIndex];
     const lineNumber = lineIndex + 1;
-    const id = asString4(evalcase.id);
-    if (filterPattern && (!id || !micromatch.isMatch(id, filterPattern))) {
+    const id = asString4(testCaseConfig.id);
+    if (filterPattern && (!id || !matchesFilter(id, filterPattern))) {
       continue;
     }
-    const conversationId = asString4(evalcase.conversation_id);
-    let outcome = asString4(evalcase.criteria);
-    if (!outcome && evalcase.expected_outcome !== void 0) {
-      outcome = asString4(evalcase.expected_outcome);
+    const conversationId = asString4(testCaseConfig.conversation_id);
+    let outcome = asString4(testCaseConfig.criteria);
+    if (!outcome && testCaseConfig.expected_outcome !== void 0) {
+      outcome = asString4(testCaseConfig.expected_outcome);
       if (outcome) {
         logWarning4(
-          `Test '${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}: ${id ?? "unknown"}. Missing required fields: id, input, and at least one of criteria/expected_output/assert`
@@ -2648,18 +2820,23 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
       }
     }
     const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const caseExecution = isJsonObject(evalcase.execution) ? evalcase.execution : void 0;
+    const caseExecution = isJsonObject(testCaseConfig.execution) ? testCaseConfig.execution : void 0;
     const mergedExecution = caseExecution ?? globalExecution;
-    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    const testCaseEvaluatorKind = coerceEvaluator(testCaseConfig.evaluator, id) ?? globalEvaluator;
     let evaluators;
     try {
-      evaluators = await parseEvaluators(evalcase, mergedExecution, searchRoots, id ?? "unknown");
+      evaluators = await parseEvaluators(
+        testCaseConfig,
+        mergedExecution,
+        searchRoots,
+        id ?? "unknown"
+      );
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       logError2(`Skipping test '${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) {
@@ -2670,7 +2847,7 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
     const userFilePaths = collectResolvedInputFilePaths(inputMessages);
     const testCase = {
       id,
-      dataset: evalSetName,
+      suite: suiteName,
       conversation_id: conversationId,
       question,
       input: inputMessages,
@@ -2678,7 +2855,7 @@ async function loadTestsFromJsonl(evalFilePath, repoRoot, options) {
       reference_answer: referenceAnswer,
       file_paths: userFilePaths,
       criteria: outcome ?? "",
-      evaluator: evalCaseEvaluatorKind,
+      evaluator: testCaseEvaluatorKind,
       assertions: evaluators
     };
     results.push(testCase);
@@ -2861,6 +3038,9 @@ function buildChatPromptFromSegments(options) {
 var ANSI_YELLOW6 = "\x1B[33m";
 var ANSI_RED3 = "\x1B[31m";
 var ANSI_RESET7 = "\x1B[0m";
+function matchesFilter2(id, filter) {
+  return typeof filter === "string" ? micromatch2.isMatch(id, filter) : filter.some((pattern) => micromatch2.isMatch(id, pattern));
+}
 function resolveTests(suite) {
   if (suite.tests !== void 0) return suite.tests;
   if (suite.eval_cases !== void 0) {
@@ -2940,18 +3120,18 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
     throw new Error(`Invalid test file format: ${evalFilePath}`);
   }
   const suite = interpolated;
-  const evalSetNameFromSuite = asString5(suite.name)?.trim();
-  const fallbackEvalSet = path7.basename(absoluteTestPath).replace(/\.eval\.ya?ml$/i, "").replace(/\.ya?ml$/i, "") || "eval";
-  const evalSetName = evalSetNameFromSuite && evalSetNameFromSuite.length > 0 ? evalSetNameFromSuite : fallbackEvalSet;
-  const rawTestcases = resolveTests(suite);
+  const suiteNameFromFile = asString5(suite.name)?.trim();
+  const fallbackSuiteName = path7.basename(absoluteTestPath).replace(/\.eval\.ya?ml$/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 = path7.dirname(absoluteTestPath);
-  let expandedTestcases;
-  if (typeof rawTestcases === "string") {
-    const externalPath = path7.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 = path7.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: ${evalFilePath} - missing 'tests' field`);
   }
@@ -2966,32 +3146,33 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
   }
   const globalExecution = suiteAssertions !== void 0 ? { ...rawGlobalExecution ?? {}, assertions: suiteAssertions } : rawGlobalExecution;
   const results = [];
-  for (const rawEvalcase of expandedTestcases) {
-    if (!isJsonObject(rawEvalcase)) {
+  for (const rawTestCase of expandedTestCases) {
+    if (!isJsonObject(rawTestCase)) {
       logWarning5("Skipping invalid test entry (expected object)");
       continue;
     }
-    const evalcase = rawEvalcase;
-    const id = asString5(evalcase.id);
-    if (filterPattern && (!id || !micromatch2.isMatch(id, filterPattern))) {
+    const testCaseConfig = rawTestCase;
+    const id = asString5(testCaseConfig.id);
+    if (filterPattern && (!id || !matchesFilter2(id, filterPattern))) {
       continue;
     }
-    const conversationId = asString5(evalcase.conversation_id);
-    let outcome = asString5(evalcase.criteria);
-    if (!outcome && evalcase.expected_outcome !== void 0) {
-      outcome = asString5(evalcase.expected_outcome);
+    const conversationId = asString5(testCaseConfig.conversation_id);
+    let outcome = asString5(testCaseConfig.criteria);
+    if (!outcome && testCaseConfig.expected_outcome !== void 0) {
+      outcome = asString5(testCaseConfig.expected_outcome);
       if (outcome) {
         logWarning5(
-          `Test '${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: ${id ?? "unknown"}. Missing required fields: id, input, and at least one of criteria/expected_output/assertions`
@@ -3038,16 +3219,21 @@ async function loadTestsFromYaml(evalFilePath, repoRoot, options) {
       }
     }
     const question = inputTextParts.map((part) => part.trim()).filter((part) => part.length > 0).join(" ");
-    const evalCaseEvaluatorKind = coerceEvaluator(evalcase.evaluator, id) ?? globalEvaluator;
+    const testCaseEvaluatorKind = coerceEvaluator(testCaseConfig.evaluator, id) ?? globalEvaluator;
     let evaluators;
     try {
-      evaluators = await parseEvaluators(evalcase, globalExecution, searchRoots, id ?? "unknown");
+      evaluators = await parseEvaluators(
+        testCaseConfig,
+        globalExecution,
+        searchRoots,
+        id ?? "unknown"
+      );
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);
       logError3(`Skipping test '${id}': ${message}`);
       continue;
     }
-    const inlineRubrics = evalcase.rubrics;
+    const inlineRubrics = testCaseConfig.rubrics;
     if (inlineRubrics !== void 0 && Array.isArray(inlineRubrics)) {
       const rubricEvaluator = parseInlineRubrics(inlineRubrics);
       if (rubricEvaluator) {
@@ -3056,13 +3242,13 @@ 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,
@@ -3071,11 +3257,12 @@ 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);
   }
@@ -3619,7 +3806,7 @@ var AzureProvider = class {
     };
     this.retryConfig = config.retry;
     const azure = createAzure(buildAzureOptions(config));
-    this.model = azure.chat(config.deploymentName);
+    this.model = config.apiFormat === "responses" ? azure(config.deploymentName) : azure.chat(config.deploymentName);
   }
   id;
   kind = "azure";
@@ -3745,7 +3932,9 @@ 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) {
@@ -5216,15 +5405,16 @@ 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
@@ -5257,7 +5447,7 @@ var CliProvider = class {
               command: renderedCommand,
               stderr: result.stderr,
               exitCode: result.exitCode ?? 0,
-              cwd: this.config.cwd,
+              cwd: effectiveCwd,
               outputFile: outputFilePath
             }
           };
@@ -5275,7 +5465,7 @@ var CliProvider = class {
               command: renderedCommand,
               stderr: result.stderr,
               exitCode: result.exitCode ?? 0,
-              cwd: this.config.cwd,
+              cwd: effectiveCwd,
               outputFile: outputFilePath,
               error: errorMessage
             }
@@ -5290,7 +5480,7 @@ var CliProvider = class {
             command: renderedCommand,
             stderr: result.stderr,
             exitCode: result.exitCode ?? 0,
-            cwd: this.config.cwd,
+            cwd: effectiveCwd,
             outputFile: outputFilePath,
             recordId: evalCaseId
           }
@@ -7240,9 +7430,9 @@ var MockProvider = class {
 };
 
 // src/evaluation/providers/pi-cli.ts
-import { spawn as spawn3 } from "node:child_process";
+import { execSync, spawn as spawn3 } from "node:child_process";
 import { randomUUID as randomUUID7 } from "node:crypto";
-import { createWriteStream as createWriteStream5 } from "node:fs";
+import { accessSync, createWriteStream as createWriteStream5, readFileSync as readFileSync2 } from "node:fs";
 import { mkdir as mkdir6, mkdtemp, rm, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import path19 from "node:path";
@@ -7300,6 +7490,75 @@ function subscribeToPiLogEntries(listener) {
   };
 }
 
+// src/evaluation/providers/pi-provider-aliases.ts
+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
 function extractPiTextContent(content) {
   if (typeof content === "string") {
@@ -7458,12 +7717,12 @@ 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");
@@ -7515,35 +7774,35 @@ ${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))) {
@@ -7834,6 +8093,24 @@ 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);
@@ -8018,17 +8295,43 @@ 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 = 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 = readFileSync2(cmdPath, "utf-8");
+    const match = content.match(/"?%_prog%"?\s+"([^"]+\.js)"/);
+    if (match) {
+      const dp0 = path19.dirname(path19.resolve(cmdPath));
+      const scriptPath = match[1].replace(/%dp0%[/\\]?/gi, `${dp0}${path19.sep}`);
+      try {
+        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 = spawn3(executable, allArgs, {
+    const child = spawn3(resolvedExe, allArgs, {
       cwd: options.cwd,
       env: options.env,
-      stdio: ["pipe", "pipe", "pipe"],
-      shell: false
+      stdio: ["pipe", "pipe", "pipe"]
     });
     let stdout = "";
     let stderr = "";
@@ -8083,13 +8386,43 @@ async function defaultPiRunner(options) {
 }
 
 // src/evaluation/providers/pi-coding-agent.ts
-import { execSync } from "node:child_process";
+import { execSync as execSync2 } from "node:child_process";
 import { randomUUID as randomUUID8 } from "node:crypto";
-import { accessSync, createWriteStream as createWriteStream6 } from "node:fs";
+import { accessSync as accessSync2, createWriteStream as createWriteStream6, mkdirSync } from "node:fs";
 import { mkdir as mkdir7 } from "node:fs/promises";
-import path20 from "node:path";
+import path21 from "node:path";
 import { createInterface } from "node:readline";
-import { fileURLToPath as fileURLToPath3 } from "node:url";
+import { fileURLToPath as fileURLToPath3, pathToFileURL } from "node:url";
+
+// src/paths.ts
+import os2 from "node:os";
+import path20 from "node:path";
+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 path20.join(os2.homedir(), ".agentv");
+}
+function getWorkspacesRoot() {
+  return path20.join(getAgentvHome(), "workspaces");
+}
+function getSubagentsRoot() {
+  return path20.join(getAgentvHome(), "subagents");
+}
+function getTraceStateRoot() {
+  return path20.join(getAgentvHome(), "trace-state");
+}
+function getWorkspacePoolRoot() {
+  return path20.join(getAgentvHome(), "workspace-pool");
+}
+
+// src/evaluation/providers/pi-coding-agent.ts
 var piCodingAgentModule = null;
 var piAiModule = null;
 var loadingPromise = null;
@@ -8107,46 +8440,126 @@ async function promptInstall() {
     rl.close();
   }
 }
-function findAgentvRoot() {
-  const thisFile = fileURLToPath3(import.meta.url);
-  let dir = path20.dirname(thisFile);
-  for (let i = 0; i < 10; i++) {
+function findManagedSdkInstallRoot() {
+  return path21.join(getAgentvHome(), "deps", "pi-sdk");
+}
+function resolveGlobalNpmRoot() {
+  try {
+    const root = execSync2("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 path21.join(globalNpmRoot, ...moduleName.split("/"), "dist", "index.js");
+}
+function findAccessiblePath(paths) {
+  for (const candidate of paths) {
     try {
-      const pkg = path20.join(dir, "package.json");
-      accessSync(pkg);
-      return dir;
+      accessSync2(candidate);
+      return candidate;
     } catch {
-      const parent = path20.dirname(dir);
-      if (parent === dir) break;
-      dir = parent;
     }
   }
-  return path20.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}...`);
-      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([
+    path21.join(managedRoot, "node_modules", "@mariozechner", "pi-coding-agent", "dist", "index.js")
+  ]);
+  const piAiEntry = findAccessiblePath([
+    path21.join(managedRoot, "node_modules", "@mariozechner", "pi-ai", "dist", "index.js"),
+    path21.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(pathToFileURL(piCodingAgentEntry).href),
+      import(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),
+    path21.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(pathToFileURL(piCodingAgentEntry).href),
+      import(pathToFileURL(piAiEntry).href)
+    ]);
+    return true;
+  } catch {
+    return false;
+  }
+}
+function installSdkModules(installDir) {
+  console.error(`Installing @mariozechner/pi-coding-agent into ${installDir} via npm...`);
+  mkdirSync(installDir, { recursive: true });
+  execSync2("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) {
@@ -8174,7 +8587,9 @@ 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 {
@@ -8196,17 +8611,35 @@ 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({
@@ -8359,28 +8792,35 @@ ${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 path20.resolve(cwdOverride);
+      return path21.resolve(cwdOverride);
     }
     if (this.config.cwd) {
-      return path20.resolve(this.config.cwd);
+      return path21.resolve(this.config.cwd);
     }
     return process.cwd();
   }
@@ -8399,9 +8839,9 @@ ${fileList}`;
   }
   resolveLogDirectory() {
     if (this.config.logDir) {
-      return path20.resolve(this.config.logDir);
+      return path21.resolve(this.config.logDir);
     }
-    return path20.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
+    return path21.join(process.cwd(), ".agentv", "logs", "pi-coding-agent");
   }
   async createStreamLogger(request) {
     const logDir = this.resolveLogDirectory();
@@ -8415,7 +8855,7 @@ ${fileList}`;
       console.warn(`Skipping Pi stream logging (could not create ${logDir}): ${message}`);
       return void 0;
     }
-    const filePath = path20.join(logDir, buildLogFilename6(request, this.targetName));
+    const filePath = path21.join(logDir, buildLogFilename6(request, this.targetName));
     try {
       const logger = await PiStreamLogger2.create({
         filePath,
@@ -8640,7 +9080,7 @@ import path30 from "node:path";
 // src/evaluation/providers/vscode/utils/fs.ts
 import { constants as constants2 } from "node:fs";
 import { access as access2, mkdir as mkdir8, readdir as readdir2, rm as rm2, stat as stat2 } from "node:fs/promises";
-import path21 from "node:path";
+import path22 from "node:path";
 async function pathExists(target) {
   try {
     await access2(target, constants2.F_OK);
@@ -8656,7 +9096,7 @@ async function readDirEntries(target) {
   const entries = await readdir2(target, { withFileTypes: true });
   return entries.map((entry) => ({
     name: entry.name,
-    absolutePath: path21.join(target, entry.name),
+    absolutePath: path22.join(target, entry.name),
     isDirectory: entry.isDirectory()
   }));
 }
@@ -8671,9 +9111,9 @@ async function removeIfExists(target) {
 }
 
 // src/evaluation/providers/vscode/utils/path.ts
-import path22 from "node:path";
+import path23 from "node:path";
 function pathToFileUri2(filePath) {
-  const absolutePath = path22.isAbsolute(filePath) ? filePath : path22.resolve(filePath);
+  const absolutePath = path23.isAbsolute(filePath) ? filePath : path23.resolve(filePath);
   const normalizedPath = absolutePath.replace(/\\/g, "/");
   if (/^[a-zA-Z]:\//.test(normalizedPath)) {
     return `file:///${normalizedPath}`;
@@ -8682,7 +9122,7 @@ function pathToFileUri2(filePath) {
 }
 
 // src/evaluation/providers/vscode/dispatch/promptBuilder.ts
-import path23 from "node:path";
+import path24 from "node:path";
 
 // src/evaluation/providers/vscode/utils/template.ts
 function renderTemplate2(content, variables) {
@@ -8774,8 +9214,8 @@ function createBatchRequestPrompt(userQuery, responseFileTmp, responseFileFinal,
   });
 }
 function createBatchOrchestratorPrompt(requestFiles, responseFiles, templateContent) {
-  const requestLines = requestFiles.map((file, index) => `${index + 1}. messages/${path23.basename(file)}`).join("\n");
-  const responseList = responseFiles.map((file) => `"${path23.basename(file)}"`).join(", ");
+  const requestLines = requestFiles.map((file, index) => `${index + 1}. messages/${path24.basename(file)}`).join("\n");
+  const responseList = responseFiles.map((file) => `"${path24.basename(file)}"`).join(", ");
   return renderTemplate2(templateContent, {
     requestFiles: requestLines,
     responseList
@@ -8784,7 +9224,7 @@ function createBatchOrchestratorPrompt(requestFiles, responseFiles, templateCont
 
 // src/evaluation/providers/vscode/dispatch/responseWaiter.ts
 import { readFile as readFile9 } from "node:fs/promises";
-import path24 from "node:path";
+import path25 from "node:path";
 
 // src/evaluation/providers/vscode/utils/time.ts
 function sleep2(ms) {
@@ -8843,7 +9283,7 @@ 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) => path24.basename(file)).join(", ");
+    const fileList = responseFilesFinal.map((file) => path25.basename(file)).join(", ");
     console.error(`waiting for ${responseFilesFinal.length} batch response(s): ${fileList}`);
   }
   const deadline = Date.now() + timeoutMs;
@@ -8852,7 +9292,7 @@ async function waitForBatchResponses(responseFilesFinal, pollInterval = 1e3, sil
     while (pending.size > 0) {
       if (Date.now() >= deadline) {
         if (!silent) {
-          const remaining = [...pending].map((f) => path24.basename(f)).join(", ");
+          const remaining = [...pending].map((f) => path25.basename(f)).join(", ");
           console.error(
             `error: timed out after ${Math.round(timeoutMs / 1e3)}s waiting for batch responses. Still pending: ${remaining}`
           );
@@ -8908,36 +9348,6 @@ import { promisify as promisify2 } from "node:util";
 
 // src/evaluation/providers/vscode/dispatch/constants.ts
 import path26 from "node:path";
-
-// src/paths.ts
-import os2 from "node:os";
-import path25 from "node:path";
-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 path25.join(os2.homedir(), ".agentv");
-}
-function getWorkspacesRoot() {
-  return path25.join(getAgentvHome(), "workspaces");
-}
-function getSubagentsRoot() {
-  return path25.join(getAgentvHome(), "subagents");
-}
-function getTraceStateRoot() {
-  return path25.join(getAgentvHome(), "trace-state");
-}
-function getWorkspacePoolRoot() {
-  return path25.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") {
@@ -10194,9 +10604,10 @@ function resolveAndCreateProvider(definition, env = process.env) {
 }
 
 // src/evaluation/evaluators/scoring.ts
-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)) {
@@ -10385,13 +10796,13 @@ async function execFileWithStdinNode(argv, stdinPayload, options) {
 async function execShellWithStdin(command, stdinPayload, options = {}) {
   const { mkdir: mkdir16, readFile: readFile15, rm: rm6, writeFile: writeFile9 } = await import("node:fs/promises");
   const { tmpdir: tmpdir3 } = await import("node:os");
-  const path49 = await import("node:path");
+  const path50 = await import("node:path");
   const { randomUUID: randomUUID10 } = await import("node:crypto");
-  const dir = path49.join(tmpdir3(), `agentv-exec-${randomUUID10()}`);
+  const dir = path50.join(tmpdir3(), `agentv-exec-${randomUUID10()}`);
   await mkdir16(dir, { recursive: true });
-  const stdinPath = path49.join(dir, "stdin.txt");
-  const stdoutPath = path49.join(dir, "stdout.txt");
-  const stderrPath = path49.join(dir, "stderr.txt");
+  const stdinPath = path50.join(dir, "stdin.txt");
+  const stdoutPath = path50.join(dir, "stdout.txt");
+  const stderrPath = path50.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("node:child_process");
@@ -11589,7 +12000,7 @@ ${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}`);
@@ -11643,54 +12054,106 @@ ${outputSchema}`;
   async runWithRetry(options) {
     const { context, 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 generateText2({
-            model,
-            system: systemPrompt,
-            messages: [
-              {
-                role: "user",
-                content: [
-                  { type: "text", text: userPrompt },
-                  ...toAiSdkImageParts(images)
-                ]
-              }
-            ],
-            ...modelOptions
-          }) : await generateText2({
-            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,
+          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,
+          graderProvider,
           systemPrompt,
-          evalCaseId: context.evalCase.id,
-          attempt: context.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, 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 generateText2({
+        model,
+        system: systemPrompt,
+        messages: [
+          {
+            role: "user",
+            content: [
+              { type: "text", text: userPrompt },
+              ...toAiSdkImageParts(images)
+            ]
+          }
+        ],
+        ...modelOptions
+      }) : await generateText2({
+        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: context.evalCase.id,
+      attempt: context.attempt,
+      maxOutputTokens: this.maxOutputTokens,
+      temperature: this.temperature
+    });
+    return {
+      text: extractLastAssistantContent(response.output),
+      providerResponse: response,
+      tokenUsage: response.tokenUsage
+    };
   }
 };
 function buildOutputSchema() {
@@ -11710,6 +12173,29 @@ 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:
@@ -11809,19 +12295,21 @@ 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({
@@ -11898,11 +12386,11 @@ function createFilesystemTools(workspacePath) {
       execute: async (input) => {
         try {
           const resolved = resolveSandboxed(workspacePath, input.path);
-          const stat10 = await fs2.stat(resolved);
-          if (stat10.isDirectory()) {
+          const stat11 = await fs2.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 fs2.open(resolved, "r");
           try {
             await fd.read(buffer, 0, buffer.length, 0);
@@ -11910,8 +12398,8 @@ 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) };
         }
@@ -11962,8 +12450,8 @@ async function searchDirectory(dirPath, workspacePath, regex, matches) {
       const ext = path35.extname(entry.name).toLowerCase();
       if (BINARY_EXTENSIONS.has(ext)) continue;
       try {
-        const stat10 = await fs2.stat(fullPath);
-        if (stat10.size > MAX_FILE_SIZE) continue;
+        const stat11 = await fs2.stat(fullPath);
+        if (stat11.size > MAX_FILE_SIZE) continue;
         const content = await fs2.readFile(fullPath, "utf-8");
         const lines = content.split("\n");
         for (let i = 0; i < lines.length; i++) {
@@ -12604,115 +13092,115 @@ var FieldAccuracyEvaluator = class {
    * Evaluate a single field against the expected value.
    */
   evaluateField(fieldConfig, candidateData, expectedData) {
-    const { path: path49, match, required = true, weight = 1 } = fieldConfig;
-    const candidateValue = resolvePath(candidateData, path49);
-    const expectedValue = resolvePath(expectedData, path49);
+    const { path: path50, match, required = true, weight = 1 } = fieldConfig;
+    const candidateValue = resolvePath(candidateData, path50);
+    const expectedValue = resolvePath(expectedData, path50);
     if (expectedValue === void 0) {
       return {
-        path: path49,
+        path: path50,
         score: 1,
         // No expected value means no comparison needed
         weight,
         hit: true,
-        message: `${path49}: no expected value`
+        message: `${path50}: no expected value`
       };
     }
     if (candidateValue === void 0) {
       if (required) {
         return {
-          path: path49,
+          path: path50,
           score: 0,
           weight,
           hit: false,
-          message: `${path49} (required, missing)`
+          message: `${path50} (required, missing)`
         };
       }
       return {
-        path: path49,
+        path: path50,
         score: 1,
         // Don't penalize missing optional fields
         weight: 0,
         // Zero weight means it won't affect the score
         hit: true,
-        message: `${path49}: optional field missing`
+        message: `${path50}: optional field missing`
       };
     }
     switch (match) {
       case "exact":
-        return this.compareExact(path49, candidateValue, expectedValue, weight);
+        return this.compareExact(path50, candidateValue, expectedValue, weight);
       case "numeric_tolerance":
         return this.compareNumericTolerance(
-          path49,
+          path50,
           candidateValue,
           expectedValue,
           fieldConfig,
           weight
         );
       case "date":
-        return this.compareDate(path49, candidateValue, expectedValue, fieldConfig, weight);
+        return this.compareDate(path50, candidateValue, expectedValue, fieldConfig, weight);
       default:
         return {
-          path: path49,
+          path: path50,
           score: 0,
           weight,
           hit: false,
-          message: `${path49}: unknown match type "${match}"`
+          message: `${path50}: unknown match type "${match}"`
         };
     }
   }
   /**
    * Exact equality comparison.
    */
-  compareExact(path49, candidateValue, expectedValue, weight) {
+  compareExact(path50, candidateValue, expectedValue, weight) {
     if (deepEqual(candidateValue, expectedValue)) {
       return {
-        path: path49,
+        path: path50,
         score: 1,
         weight,
         hit: true,
-        message: path49
+        message: path50
       };
     }
     if (typeof candidateValue !== typeof expectedValue) {
       return {
-        path: path49,
+        path: path50,
         score: 0,
         weight,
         hit: false,
-        message: `${path49} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
+        message: `${path50} (type mismatch: got ${typeof candidateValue}, expected ${typeof expectedValue})`
       };
     }
     return {
-      path: path49,
+      path: path50,
       score: 0,
       weight,
       hit: false,
-      message: `${path49} (value mismatch)`
+      message: `${path50} (value mismatch)`
     };
   }
   /**
    * Numeric comparison with absolute or relative tolerance.
    */
-  compareNumericTolerance(path49, candidateValue, expectedValue, fieldConfig, weight) {
+  compareNumericTolerance(path50, candidateValue, expectedValue, fieldConfig, weight) {
     const { tolerance = 0, relative = false } = fieldConfig;
     const candidateNum = toNumber(candidateValue);
     const expectedNum = toNumber(expectedValue);
     if (candidateNum === null || expectedNum === null) {
       return {
-        path: path49,
+        path: path50,
         score: 0,
         weight,
         hit: false,
-        message: `${path49} (non-numeric value)`
+        message: `${path50} (non-numeric value)`
       };
     }
     if (!Number.isFinite(candidateNum) || !Number.isFinite(expectedNum)) {
       return {
-        path: path49,
+        path: path50,
         score: 0,
         weight,
         hit: false,
-        message: `${path49} (invalid numeric value)`
+        message: `${path50} (invalid numeric value)`
       };
     }
     const diff = Math.abs(candidateNum - expectedNum);
@@ -12725,61 +13213,61 @@ var FieldAccuracyEvaluator = class {
     }
     if (withinTolerance) {
       return {
-        path: path49,
+        path: path50,
         score: 1,
         weight,
         hit: true,
-        message: `${path49} (within tolerance: diff=${diff.toFixed(2)})`
+        message: `${path50} (within tolerance: diff=${diff.toFixed(2)})`
       };
     }
     return {
-      path: path49,
+      path: path50,
       score: 0,
       weight,
       hit: false,
-      message: `${path49} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
+      message: `${path50} (outside tolerance: diff=${diff.toFixed(2)}, tolerance=${tolerance})`
     };
   }
   /**
    * Date comparison with format normalization.
    */
-  compareDate(path49, candidateValue, expectedValue, fieldConfig, weight) {
+  compareDate(path50, candidateValue, expectedValue, fieldConfig, weight) {
     const formats = fieldConfig.formats ?? DEFAULT_DATE_FORMATS;
     const candidateDate = parseDate(String(candidateValue), formats);
     const expectedDate = parseDate(String(expectedValue), formats);
     if (candidateDate === null) {
       return {
-        path: path49,
+        path: path50,
         score: 0,
         weight,
         hit: false,
-        message: `${path49} (unparseable candidate date)`
+        message: `${path50} (unparseable candidate date)`
       };
     }
     if (expectedDate === null) {
       return {
-        path: path49,
+        path: path50,
         score: 0,
         weight,
         hit: false,
-        message: `${path49} (unparseable expected date)`
+        message: `${path50} (unparseable expected date)`
       };
     }
     if (candidateDate.getFullYear() === expectedDate.getFullYear() && candidateDate.getMonth() === expectedDate.getMonth() && candidateDate.getDate() === expectedDate.getDate()) {
       return {
-        path: path49,
+        path: path50,
         score: 1,
         weight,
         hit: true,
-        message: path49
+        message: path50
       };
     }
     return {
-      path: path49,
+      path: path50,
       score: 0,
       weight,
       hit: false,
-      message: `${path49} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
+      message: `${path50} (date mismatch: got ${formatDateISO(candidateDate)}, expected ${formatDateISO(expectedDate)})`
     };
   }
   /**
@@ -12812,11 +13300,11 @@ var FieldAccuracyEvaluator = class {
     };
   }
 };
-function resolvePath(obj, path49) {
-  if (!path49 || !obj) {
+function resolvePath(obj, path50) {
+  if (!path50 || !obj) {
     return void 0;
   }
-  const parts = path49.split(/\.|\[|\]/).filter((p) => p.length > 0);
+  const parts = path50.split(/\.|\[|\]/).filter((p) => p.length > 0);
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0) {
@@ -13308,8 +13796,8 @@ var TokenUsageEvaluator = class {
 };
 
 // src/evaluation/evaluators/tool-trajectory.ts
-function getNestedValue(obj, path49) {
-  const parts = path49.split(".");
+function getNestedValue(obj, path50) {
+  const parts = path50.split(".");
   let current = obj;
   for (const part of parts) {
     if (current === null || current === void 0 || typeof current !== "object") {
@@ -15081,7 +15569,7 @@ 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") {
@@ -15094,7 +15582,17 @@ 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 });
     }
@@ -15391,7 +15889,7 @@ async function executeWorkspaceScript(config, context, 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) {
@@ -15483,7 +15981,7 @@ 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 ${evalFilePath}`);
     }
     return [];
   }
@@ -15509,20 +16007,10 @@ async function runEvaluation(options) {
     if (resolvedTargetsByName.has(name)) {
       return resolvedTargetsByName.get(name);
     }
-    let definition = targetDefinitions.get(name);
+    const definition = resolveDelegatedTargetDefinition(name, targetDefinitions, envLookup);
     if (!definition) {
       return void 0;
     }
-    for (let depth = 0; depth < 5; depth++) {
-      const useTarget = definition.use_target;
-      if (typeof useTarget !== "string" || useTarget.trim().length === 0) break;
-      const envMatch = useTarget.trim().match(/^\$\{\{\s*([A-Z0-9_]+)\s*\}\}$/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;
@@ -15545,6 +16033,9 @@ 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);
@@ -15875,7 +16366,7 @@ 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: [],
@@ -15912,7 +16403,7 @@ 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: [],
@@ -16224,7 +16715,7 @@ async function runBatchEvaluation(options) {
         targetResolver,
         availableTargets,
         verbose,
-        threshold: batchThreshold
+        threshold: evalCase.threshold ?? batchThreshold
       });
       if (providerError) {
         result = {
@@ -16686,8 +17177,9 @@ 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 ? {
@@ -16701,7 +17193,7 @@ 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,
@@ -16902,7 +17394,8 @@ async function evaluateCandidate(options) {
     targetResolver,
     availableTargets,
     fileChanges,
-    workspacePath
+    workspacePath,
+    threshold: evalThreshold
   });
   const completedAt = nowFn();
   let agentRequest;
@@ -16933,7 +17426,7 @@ 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,
@@ -16976,7 +17469,8 @@ async function runEvaluatorsForCase(options) {
     targetResolver,
     availableTargets,
     fileChanges,
-    workspacePath
+    workspacePath,
+    threshold
   } = options;
   if (evalCase.assertions && evalCase.assertions.length > 0) {
     return runEvaluatorList({
@@ -17002,7 +17496,8 @@ async function runEvaluatorsForCase(options) {
       targetResolver,
       availableTargets,
       fileChanges,
-      workspacePath
+      workspacePath,
+      threshold
     });
   }
   const evaluatorKind = evalCase.evaluator ?? "llm-grader";
@@ -17104,7 +17599,8 @@ 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,
@@ -17139,7 +17635,8 @@ 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",
@@ -17173,9 +17670,10 @@ 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");
@@ -17186,17 +17684,23 @@ 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" ? micromatch3.isMatch(id, filter) : filter.some((pattern) => micromatch3.isMatch(id, pattern));
+}
 function filterEvalCases(evalCases, filter) {
   if (!filter) {
     return evalCases;
   }
-  return evalCases.filter((evalCase) => micromatch3.isMatch(evalCase.id, filter));
+  return evalCases.filter((evalCase) => matchesFilter3(evalCase.id, filter));
 }
 function buildEvaluatorRegistry(overrides, resolveGraderProvider) {
   const llmGrader = overrides?.["llm-grader"] ?? new LlmGraderEvaluator({
@@ -17283,7 +17787,7 @@ 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,
@@ -17555,6 +18059,7 @@ async function evaluate(config) {
     verbose: config.verbose,
     maxConcurrency: config.workers ?? 3,
     filter: config.filter,
+    threshold: config.threshold,
     evalCases,
     onResult: async (result) => {
       collectedResults.push(result);
@@ -17565,19 +18070,19 @@ 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++;
     }
   }
@@ -17608,7 +18113,7 @@ async function discoverDefaultTarget(repoRoot) {
   return null;
 }
 async function loadEnvHierarchy(repoRoot, startPath) {
-  const { readFileSync: readFileSync3 } = await import("node:fs");
+  const { readFileSync: readFileSync4 } = await import("node:fs");
   const chain = buildDirectoryChain(startPath, repoRoot);
   const envFiles = [];
   for (const dir of chain) {
@@ -17617,7 +18122,7 @@ 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;
@@ -17690,7 +18195,7 @@ var CONFIG_FILE_NAMES = [
 ];
 async function loadTsConfig(projectRoot) {
   const { existsSync: existsSync7 } = await import("node:fs");
-  const { pathToFileURL } = await import("node:url");
+  const { pathToFileURL: pathToFileURL2 } = await import("node:url");
   const { join: join2 } = await import("node:path");
   for (const fileName of CONFIG_FILE_NAMES) {
     const filePath = join2(projectRoot, fileName);
@@ -17698,7 +18203,7 @@ 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);
@@ -17832,7 +18337,7 @@ function shouldSkipCacheForTemperature(targetConfig) {
 }
 
 // src/projects.ts
-import { existsSync as existsSync6, mkdirSync, readFileSync as readFileSync2, readdirSync as readdirSync3, statSync as statSync2, writeFileSync } from "node:fs";
+import { existsSync as existsSync6, mkdirSync as mkdirSync2, readFileSync as readFileSync3, readdirSync as readdirSync3, statSync as statSync2, writeFileSync } from "node:fs";
 import path47 from "node:path";
 import { parse as parseYaml3, stringify as stringifyYaml } from "yaml";
 function getProjectsRegistryPath() {
@@ -17844,7 +18349,7 @@ function loadProjectRegistry() {
     return { projects: [] };
   }
   try {
-    const raw = readFileSync2(registryPath, "utf-8");
+    const raw = readFileSync3(registryPath, "utf-8");
     const parsed = parseYaml3(raw);
     if (!parsed || !Array.isArray(parsed.projects)) {
       return { projects: [] };
@@ -17858,7 +18363,7 @@ function saveProjectRegistry(registry) {
   const registryPath = getProjectsRegistryPath();
   const dir = path47.dirname(registryPath);
   if (!existsSync6(dir)) {
-    mkdirSync(dir, { recursive: true });
+    mkdirSync2(dir, { recursive: true });
   }
   writeFileSync(registryPath, stringifyYaml(registry), "utf-8");
 }
@@ -18124,7 +18629,7 @@ 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];
@@ -18333,7 +18838,7 @@ 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 */
@@ -18682,11 +19187,241 @@ function extractToolResultContent(content) {
   return parts.length > 0 ? parts.join("") : void 0;
 }
 
-// src/import/session-discovery.ts
+// src/import/codex-parser.ts
+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
 import { readdir as readdir8, stat as stat9 } from "node:fs/promises";
 import { homedir as homedir3 } from "node:os";
 import path48 from "node:path";
-var DEFAULT_PROJECTS_DIR = () => path48.join(homedir3(), ".claude", "projects");
+var DEFAULT_SESSIONS_DIR = () => path48.join(homedir3(), ".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 readdir8(sessionsDir);
+  } catch {
+    return [];
+  }
+  for (const year of yearDirs) {
+    const yearPath = path48.join(sessionsDir, year);
+    let monthDirs;
+    try {
+      monthDirs = await readdir8(yearPath);
+    } catch {
+      continue;
+    }
+    for (const month of monthDirs) {
+      const monthPath = path48.join(yearPath, month);
+      let dayDirs;
+      try {
+        dayDirs = await readdir8(monthPath);
+      } catch {
+        continue;
+      }
+      for (const day of dayDirs) {
+        if (opts?.date) {
+          const dirDate = `${year}-${month}-${day}`;
+          if (dirDate !== opts.date) continue;
+        }
+        const dayPath = path48.join(monthPath, day);
+        let files;
+        try {
+          files = await readdir8(dayPath);
+        } catch {
+          continue;
+        }
+        for (const file of files) {
+          if (!file.startsWith("rollout-") || !file.endsWith(".jsonl")) continue;
+          const filePath = path48.join(dayPath, file);
+          const nameWithoutExt = file.replace(/\.jsonl$/, "");
+          const parts = nameWithoutExt.split("-");
+          const sessionId = parts.length >= 6 ? parts.slice(-5).join("-") : nameWithoutExt;
+          let updatedAt;
+          try {
+            const fileStat = await stat9(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
+import { readdir as readdir9, stat as stat10 } from "node:fs/promises";
+import { homedir as homedir4 } from "node:os";
+import path49 from "node:path";
+var DEFAULT_PROJECTS_DIR = () => path49.join(homedir4(), ".claude", "projects");
 function encodeProjectPath(projectPath) {
   return projectPath.replace(/\//g, "-");
 }
@@ -18695,7 +19430,7 @@ async function discoverClaudeSessions(opts) {
   const limit = opts?.latest ? 1 : opts?.limit ?? 10;
   let projectDirs;
   try {
-    projectDirs = await readdir8(projectsDir);
+    projectDirs = await readdir9(projectsDir);
   } catch {
     return [];
   }
@@ -18705,10 +19440,10 @@ async function discoverClaudeSessions(opts) {
   }
   const sessions = [];
   for (const projectDir of projectDirs) {
-    const dirPath = path48.join(projectsDir, projectDir);
+    const dirPath = path49.join(projectsDir, projectDir);
     let entries;
     try {
-      entries = await readdir8(dirPath);
+      entries = await readdir9(dirPath);
     } catch {
       continue;
     }
@@ -18716,10 +19451,10 @@ async function discoverClaudeSessions(opts) {
       if (!entry.endsWith(".jsonl")) continue;
       const sessionId = entry.replace(/\.jsonl$/, "");
       if (opts?.sessionId && sessionId !== opts.sessionId) continue;
-      const filePath = path48.join(dirPath, entry);
+      const filePath = path49.join(dirPath, entry);
       let updatedAt;
       try {
-        const fileStat = await stat9(filePath);
+        const fileStat = await stat10(filePath);
         updatedAt = fileStat.mtime;
       } catch {
         updatedAt = /* @__PURE__ */ new Date(0);
@@ -18738,10 +19473,85 @@ async function discoverClaudeSessions(opts) {
 
 // src/import/types.ts
 import { readFile as readFile14 } from "node: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 readFile14(filePath, "utf8");
+  return text.split("\n").filter((line) => line.trim().length > 0).map((line) => JSON.parse(line));
+}
 async function readTranscriptFile(filePath) {
   return readFile14(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" };
@@ -18755,6 +19565,7 @@ export {
   DEFAULT_EVALUATOR_TEMPLATE,
   DEFAULT_EVAL_PATTERNS,
   DEFAULT_EXPLORATION_TOOLS,
+  DEFAULT_THRESHOLD,
   DeterministicAssertionEvaluator,
   EvaluatorRegistry,
   ExecutionMetricsEvaluator,
@@ -18776,6 +19587,7 @@ export {
   TemplateNotFoundError,
   TokenUsageEvaluator,
   ToolTrajectoryEvaluator,
+  TranscriptProvider,
   WorkspaceCreationError,
   WorkspacePoolManager,
   addProject,
@@ -18812,6 +19624,7 @@ export {
   detectFormat,
   discoverAssertions,
   discoverClaudeSessions,
+  discoverCodexSessions,
   discoverCopilotSessions,
   discoverGraders,
   discoverGraders as discoverJudges,
@@ -18872,6 +19685,8 @@ export {
   normalizeLineEndings,
   parseAgentSkillsEvals,
   parseClaudeSession,
+  parseCodexSession,
+  parseCopilotEvents,
   parseJsonFromText,
   parseJsonSafe,
   readJsonFile,
@@ -18879,8 +19694,10 @@ export {
   readTestSuiteMetadata,
   readTextFile,
   readTranscriptFile,
+  readTranscriptJsonl,
   removeProject,
   resolveAndCreateProvider,
+  resolveDelegatedTargetDefinition,
   resolveFileReference,
   resolveTargetDefinition,
   resolveWorkspaceTemplate,
@@ -18910,6 +19727,7 @@ export {
   substituteVariables,
   toCamelCaseDeep,
   toSnakeCaseDeep,
+  toTranscriptJsonLine,
   tokensPerTool,
   touchProject,
   transpileEvalYaml,