@nathapp/nax 0.63.0 → 0.63.1

package/dist/nax.js

@@ -7598,6 +7598,12 @@ ${SEMANTIC_OUTPUT_SCHEMA}`;
 ` + `Output ONLY the JSON object from your review \u2014 no markdown fences, no explanation.
 ` + "The object must start with { and end with }.";
   }
+  static jsonRetryCondensed(maxFindings = 3) {
+    return `Your previous response was truncated and could not be parsed as valid JSON.
+Respond with a condensed summary: at most ${maxFindings} findings, highest severity first.
+Output ONLY a complete, valid JSON object. It must start with { and end with }.
+Schema: {"passed": boolean, "findings": [{"severity": string, "category": string, "file": string, "line": number, "issue": string, "suggestion": string}]}`;
+  }
 }
 function buildAttemptContextBlock(priorFailures) {
   if (!priorFailures || priorFailures.length === 0)
@@ -23835,7 +23841,7 @@ var init_schemas3 = __esm(() => {
   });
   SemanticReviewConfigSchema = exports_external.object({
     modelTier: ModelTierSchema.default("balanced"),
-    diffMode: exports_external.enum(["embedded", "ref"]).default("embedded"),
+    diffMode: exports_external.enum(["embedded", "ref"]).default("ref"),
     resetRefOnRerun: exports_external.boolean().default(false),
     rules: exports_external.array(exports_external.string()).default([]),
     timeoutMs: exports_external.number().int().positive().default(600000),
@@ -24270,7 +24276,7 @@ var init_schemas3 = __esm(() => {
       blockingThreshold: "error",
       semantic: {
         modelTier: "balanced",
-        diffMode: "embedded",
+        diffMode: "ref",
         resetRefOnRerun: false,
         rules: [],
         timeoutMs: 600000,
@@ -24469,6 +24475,7 @@ __export(exports_adapter, {
   closeAcpSession: () => closeAcpSession,
   _fallbackDeps: () => _fallbackDeps,
   _acpAdapterDeps: () => _acpAdapterDeps,
+  MAX_AGENT_OUTPUT_CHARS: () => MAX_AGENT_OUTPUT_CHARS,
   AcpAgentAdapter: () => AcpAgentAdapter
 });
 import { createHash } from "crypto";
@@ -30123,7 +30130,9 @@ async function collectStoryMetrics(ctx, storyStartTime) {
   const priorFailureCount = story.priorFailures?.length || 0;
   const attempts = priorFailureCount + Math.max(1, story.attempts || 1);
   const finalTier = escalationCount > 0 ? story.escalations[escalationCount - 1].toTier : routing.modelTier;
-  const firstPassSuccess = agentResult?.success === true && escalationCount === 0 && priorFailureCount === 0;
+  const autofixAttemptCount = ctx.autofixAttempt ?? 0;
+  const rectifyAttemptCount = ctx.rectifyAttempt ?? 0;
+  const firstPassSuccess = agentResult?.success === true && escalationCount === 0 && priorFailureCount === 0 && autofixAttemptCount === 0 && rectifyAttemptCount === 0;
   const agentUsed = routing.agent ?? ctx.agentManager?.getDefault() ?? resolveDefaultAgent(ctx.config);
   let modelUsed = routing.modelTier;
   try {
@@ -33060,8 +33069,27 @@ var init_event_bus = __esm(() => {
 });
 
 // src/pipeline/stages/autofix-adversarial.ts
-function splitAdversarialFindingsByScope(check2, testFilePatterns) {
-  if (check2.check !== "adversarial" || !check2.findings?.length) {
+function extractFilesFromLintOutput(output) {
+  if (!output.trim())
+    return [];
+  const files = new Set;
+  const PATH_RE = /^[ \t]*((?:\/[\w./-]+|\.\.?\/[\w./-]+|[\w][\w-]*(?:\/[\w./-]+)+))(?::\d+)?(?::\d+)?(?:\s|:|$)/gm;
+  let startIndex = 0;
+  while (startIndex <= output.length) {
+    PATH_RE.lastIndex = startIndex;
+    const m = PATH_RE.exec(output);
+    if (m === null)
+      break;
+    const candidate = m[1];
+    if (SOURCE_EXT_RE.test(candidate)) {
+      files.add(candidate);
+    }
+    startIndex = m.index + 1;
+  }
+  return Array.from(files);
+}
+function splitByStructuredFindings(check2, testFilePatterns) {
+  if (!check2.findings?.length) {
     return { testFindings: null, sourceFindings: null };
   }
   const testFs = check2.findings.filter((f) => isTestFile(f.file ?? "", testFilePatterns));
@@ -33073,12 +33101,36 @@ function splitAdversarialFindingsByScope(check2, testFilePatterns) {
   };
   return { testFindings: toCheck(testFs), sourceFindings: toCheck(sourceFs) };
 }
+function splitByOutputParsing(check2, testFilePatterns) {
+  const files = extractFilesFromLintOutput(check2.output);
+  if (files.length === 0) {
+    if (check2.output.trim()) {
+      return { testFindings: null, sourceFindings: check2 };
+    }
+    return { testFindings: null, sourceFindings: null };
+  }
+  const hasTest = files.some((f) => isTestFile(f, testFilePatterns));
+  const hasSource = files.some((f) => !isTestFile(f, testFilePatterns));
+  return {
+    testFindings: hasTest ? check2 : null,
+    sourceFindings: hasSource ? check2 : null
+  };
+}
+function splitFindingsByScope(check2, testFilePatterns) {
+  if (check2.check === "adversarial") {
+    return splitByStructuredFindings(check2, testFilePatterns);
+  }
+  if (check2.check === "lint") {
+    return splitByOutputParsing(check2, testFilePatterns);
+  }
+  return { testFindings: null, sourceFindings: null };
+}
 async function runTestWriterRectification(ctx, testWriterChecks, story, agentManager, keepOpen = true) {
   const logger = getLogger();
   const twPrompt = RectifierPromptBuilder.testWriterRectification(testWriterChecks, story);
   const defaultAgent = agentManager.getDefault();
   if (!defaultAgent) {
-    logger.warn("autofix", "Test-writer rectification skipped \u2014 no default agent", { storyId: ctx.story.id });
+    logger.warn("autofix", "Test-writer rectification skipped -- no default agent", { storyId: ctx.story.id });
     return 0;
   }
   const modelTier = ctx.rootConfig.tdd?.sessionTiers?.testWriter ?? "balanced";
@@ -33104,17 +33156,19 @@ async function runTestWriterRectification(ctx, testWriterChecks, story, agentMan
     });
     return twResult.estimatedCost ?? 0;
   } catch {
-    logger.warn("autofix", "Test-writer rectification failed \u2014 proceeding with implementer", {
+    logger.warn("autofix", "Test-writer rectification failed -- proceeding with implementer", {
       storyId: ctx.story.id
     });
     return 0;
   }
 }
+var SOURCE_EXT_RE;
 var init_autofix_adversarial = __esm(() => {
   init_config();
   init_logger2();
   init_prompts();
   init_test_runners();
+  SOURCE_EXT_RE = /\.(ts|tsx|js|jsx|mjs|cjs|go|py|rs|rb|java|cs|cpp|c|h|swift|kt)$/;
 });
 
 // src/review/dialogue.ts
@@ -33520,6 +33574,17 @@ var init_agent_profiles = __esm(() => {
         toolSchemaDialect: "openai"
       }
     },
+    opencode: {
+      caps: {
+        maxContextTokens: 128000,
+        preferredPromptTokens: 12000,
+        supportsToolCalls: true,
+        supportsSystemPrompt: true,
+        supportsMarkdown: true,
+        systemPromptStyle: "markdown-sections",
+        toolSchemaDialect: "openai"
+      }
+    },
     local: {
       caps: {
         maxContextTokens: 32000,
@@ -36787,6 +36852,14 @@ var init_review_audit = __esm(() => {
   };
 });
 
+// src/review/truncation.ts
+function looksLikeTruncatedJson(raw) {
+  return raw.trimEnd().length >= MAX_AGENT_OUTPUT_CHARS - 100;
+}
+var init_truncation = __esm(() => {
+  init_adapter();
+});
+
 // src/review/adversarial.ts
 function validateAdversarialShape(parsed) {
   if (typeof parsed !== "object" || parsed === null)
@@ -36988,22 +37061,26 @@ async function runAdversarialReview(workdir, storyGitRef, story, adversarialConf
     return {
       check: "adversarial",
       success: true,
+      failOpen: true,
       command: "",
       exitCode: 0,
       output: `skipped: LLM call failed \u2014 ${String(err)}`,
       durationMs: Date.now() - startTime
     };
   }
-  if (!parseAdversarialResponse(rawResponse)) {
+  const isTruncated = looksLikeTruncatedJson(rawResponse);
+  if (isTruncated || !parseAdversarialResponse(rawResponse)) {
     retryAttempted = true;
+    const retryPrompt = isTruncated ? ReviewPromptBuilder.jsonRetryCondensed() : ReviewPromptBuilder.jsonRetry();
     logger?.info("adversarial", "JSON parse failed, retrying (1/1)", {
       storyId: story.id,
       rawHead: rawResponse.slice(0, 200),
-      responseLen: rawResponse.length
+      responseLen: rawResponse.length,
+      isTruncated
     });
     try {
       const retryResult = await agentManager.run({
-        runOptions: { prompt: ReviewPromptBuilder.jsonRetry(), ...runOpts, keepOpen: false }
+        runOptions: { prompt: retryPrompt, ...runOpts, keepOpen: false }
       });
       rawResponse = retryResult.output;
       llmCost += retryResult.estimatedCost ?? 0;
@@ -37058,6 +37135,7 @@ async function runAdversarialReview(workdir, storyGitRef, story, adversarialConf
     return {
       check: "adversarial",
       success: true,
+      failOpen: true,
       command: "",
       exitCode: 0,
       output: "adversarial review: could not parse LLM response (fail-open)",
@@ -37164,6 +37242,7 @@ var init_adversarial = __esm(() => {
   init_review_builder();
   init_diff_utils();
   init_review_audit();
+  init_truncation();
   _adversarialDeps = {
     writeReviewAudit
   };
@@ -37274,7 +37353,7 @@ async function runSemanticReview(workdir, storyGitRef, story, semanticConfig, ag
       durationMs: Date.now() - startTime
     };
   }
-  const diffMode = semanticConfig.diffMode ?? "embedded";
+  const diffMode = semanticConfig.diffMode ?? "ref";
   logger?.info("review", "Running semantic check", {
     storyId: story.id,
     modelTier: semanticConfig.modelTier,
@@ -37546,22 +37625,26 @@ ${formatFindings2(debateBlocking)}`,
     return {
       check: "semantic",
       success: true,
+      failOpen: true,
       command: "",
       exitCode: 0,
       output: `skipped: LLM call failed \u2014 ${String(err)}`,
       durationMs: Date.now() - startTime
     };
   }
-  if (!parseLLMResponse(rawResponse)) {
+  const isTruncated = looksLikeTruncatedJson(rawResponse);
+  if (isTruncated || !parseLLMResponse(rawResponse)) {
     retryAttempted = true;
+    const retryPrompt = isTruncated ? ReviewPromptBuilder.jsonRetryCondensed() : ReviewPromptBuilder.jsonRetry();
     logger?.info("semantic", "JSON parse failed, retrying (1/1)", {
       storyId: story.id,
       rawHead: rawResponse.slice(0, 200),
-      responseLen: rawResponse.length
+      responseLen: rawResponse.length,
+      isTruncated
     });
     try {
       const retryResult = await agentManager.run({
-        runOptions: { prompt: ReviewPromptBuilder.jsonRetry(), ...runOpts, keepOpen: false }
+        runOptions: { prompt: retryPrompt, ...runOpts, keepOpen: false }
       });
       rawResponse = retryResult.output;
       llmCost += retryResult.estimatedCost ?? 0;
@@ -37616,6 +37699,7 @@ ${formatFindings2(debateBlocking)}`,
     return {
       check: "semantic",
       success: true,
+      failOpen: true,
       command: "",
       exitCode: 0,
       output: "semantic review: could not parse LLM response (fail-open)",
@@ -37719,6 +37803,7 @@ var init_semantic = __esm(() => {
   init_test_runners();
   init_diff_utils();
   init_review_audit();
+  init_truncation();
   _semanticDeps = {
     createDebateSession: (opts) => new DebateSession(opts),
     writeReviewAudit
@@ -37873,7 +37958,7 @@ Stage and commit these files before running review.`
       };
       const semanticCfg = config2.semantic ?? {
         modelTier: "balanced",
-        diffMode: "embedded",
+        diffMode: "ref",
         resetRefOnRerun: false,
         rules: [],
         timeoutMs: 600000
@@ -38043,6 +38128,15 @@ function buildReviewSummary(checks3) {
   }
   return summary;
 }
+function formatFailureReason(check2) {
+  return check2.check === "semantic" || check2.check === "adversarial" ? `${check2.check} failed` : `${check2.check} failed (exit code ${check2.exitCode})`;
+}
+function buildFailureReason(checks3) {
+  const failedChecks = checks3.filter((check2) => !check2.success);
+  if (failedChecks.length === 0)
+    return;
+  return failedChecks.map(formatFailureReason).join(", ");
+}
 
 class ReviewOrchestrator {
   async review(reviewConfig, workdir, executionConfig, plugins, storyGitRef, scopePrefix, qualityCommands, storyId, story, agentManager, naxConfig, retrySkipChecks, featureName, resolverSession, priorFailures, featureContextMarkdown, contextBundles, projectDir, env2, naxIgnoreIndex) {
@@ -38116,8 +38210,7 @@ class ReviewOrchestrator {
       const allChecks = [...mechanicalResult.checks, ...llmCheckResults];
       const mechanicalPassed = mechanicalResult.success;
       const llmPassed = llmCheckResults.every((c) => c.success);
-      const firstFailure = allChecks.find((c) => !c.success);
-      const failureReason = firstFailure ? firstFailure.check === "semantic" || firstFailure.check === "adversarial" ? `${firstFailure.check} failed` : `${firstFailure.check} failed (exit code ${firstFailure.exitCode})` : undefined;
+      const failureReason = buildFailureReason(allChecks);
       const reviewSummary = buildReviewSummary(llmCheckResults);
       builtIn = {
         success: mechanicalPassed && llmPassed,
@@ -38368,6 +38461,20 @@ var init_review = __esm(() => {
       ctx.reviewResult = result.builtIn;
       ctx.mechanicalFailedOnly = result.mechanicalFailedOnly;
       const reviewCost = (result.builtIn.checks ?? []).reduce((sum, c) => sum + (c.cost ?? 0), 0) || undefined;
+      const failOpenChecks = result.builtIn.success ? (result.builtIn.checks ?? []).filter((c) => c.failOpen).map((c) => c.check) : [];
+      if (failOpenChecks.length > 0 && (ctx.autofixAttempt ?? 0) > 0) {
+        logger.warn("review", "Fail-open on partial-progress retry \u2014 treating as failure (fail-closed on ambiguity)", {
+          storyId: ctx.story.id,
+          failOpenChecks,
+          autofixAttempt: ctx.autofixAttempt
+        });
+        ctx.reviewResult = {
+          ...result.builtIn,
+          success: false,
+          failureReason: `fail-open on retry: ${failOpenChecks.join(", ")}`
+        };
+        return { action: "continue", cost: reviewCost };
+      }
       if (!result.success) {
         const pluginFindings = result.builtIn.pluginReviewers?.flatMap((pr) => pr.findings ?? []) ?? [];
         const semanticFindings = (result.builtIn.checks ?? []).filter((c) => c.check === "semantic" && !c.success && c.findings?.length).flatMap((c) => c.findings ?? []);
@@ -38413,6 +38520,9 @@ async function recheckReview(ctx) {
   if (!reviewStage2.enabled(ctx))
     return true;
   await reviewStage2.execute(ctx);
+  const hasFailOpen = (ctx.reviewResult?.checks ?? []).some((c) => c.failOpen);
+  if (hasFailOpen)
+    return false;
   return ctx.reviewResult?.success === true;
 }
 function collectFailedChecks(ctx) {
@@ -38477,14 +38587,16 @@ async function runAgentRectification(ctx, lintFixCmd, formatFixCmd, effectiveWor
   let testWriterChecks = [];
   const stageTestFilePatterns = typeof ctx.rootConfig.execution?.smartTestRunner === "object" ? ctx.rootConfig.execution.smartTestRunner?.testFilePatterns : undefined;
   for (const check2 of failedChecks) {
-    if (check2.check === "adversarial" && check2.findings?.length) {
-      const { testFindings, sourceFindings } = splitAdversarialFindingsByScope(check2, stageTestFilePatterns);
-      if (testFindings)
-        testWriterChecks = [...testWriterChecks, testFindings];
-      if (sourceFindings) {
-        implementerChecks = implementerChecks.map((c) => c === check2 ? sourceFindings : c);
-      } else {
-        implementerChecks = implementerChecks.filter((c) => c !== check2);
+    if (check2.check === "adversarial" || check2.check === "lint") {
+      const { testFindings, sourceFindings } = splitFindingsByScope(check2, stageTestFilePatterns);
+      if (testFindings || sourceFindings) {
+        if (testFindings)
+          testWriterChecks = [...testWriterChecks, testFindings];
+        if (sourceFindings) {
+          implementerChecks = implementerChecks.map((c) => c === check2 ? sourceFindings : c);
+        } else {
+          implementerChecks = implementerChecks.filter((c) => c !== check2);
+        }
       }
     }
   }
@@ -38493,18 +38605,18 @@ async function runAgentRectification(ctx, lintFixCmd, formatFixCmd, effectiveWor
     if (ctx.routing.testStrategy === "no-test") {
       logger.warn("autofix", "Skipping test-writer rectification (no-test strategy)", {
         storyId: ctx.story.id,
-        skippedFindingCount: testWriterChecks.flatMap((c) => c.findings ?? []).length
+        checks: testWriterChecks.map((c) => c.check)
       });
     } else {
-      logger.info("autofix", "Routing test-file adversarial findings to test-writer session", {
+      logger.info("autofix", "Routing test-file findings to test-writer session", {
         storyId: ctx.story.id,
-        findingCount: testWriterChecks.flatMap((c) => c.findings ?? []).length
+        checks: testWriterChecks.map((c) => c.check)
       });
       autofixCostAccum += await _autofixDeps.runTestWriterRectification(ctx, testWriterChecks, ctx.story, agentManager);
     }
   }
   if (implementerChecks.length === 0) {
-    logger.info("autofix", "All adversarial findings routed to test-writer \u2014 skipping implementer loop", {
+    logger.info("autofix", "All findings routed to test-writer \u2014 skipping implementer loop", {
       storyId: ctx.story.id
     });
     return { succeeded: false, cost: autofixCostAccum };
@@ -38845,13 +38957,11 @@ var init_autofix = __esm(() => {
       if (ctx.routing.testStrategy === "no-test") {
         const failedChecks = (reviewResult.checks ?? []).filter((c) => !c.success);
         if (failedChecks.length > 0 && failedChecks.every((c) => {
-          if (c.check !== "adversarial")
-            return false;
-          const { testFindings, sourceFindings } = splitAdversarialFindingsByScope(c, testFilePatterns);
+          const { testFindings, sourceFindings } = splitFindingsByScope(c, testFilePatterns);
           return testFindings !== null && sourceFindings === null;
         })) {
           const skippedFindingCount = failedChecks.flatMap((c) => c.findings ?? []).length;
-          logger.warn("autofix", "Adversarial review found test-file issues \u2014 skipped (no-test strategy)", {
+          logger.warn("autofix", "Review found test-file issues only \u2014 skipped (no-test strategy)", {
             storyId: ctx.story.id,
             skippedFindingCount
           });
@@ -40100,6 +40210,40 @@ var init_verification = __esm(() => {
   init_runners();
 });
 
+// src/verification/failure-records.ts
+function truncateUnmappedFailureOutput(output) {
+  const tailLines = output.split(`
+`).slice(-UNMAPPED_FAILURE_OUTPUT_MAX_LINES).join(`
+`);
+  if (tailLines.length <= UNMAPPED_FAILURE_OUTPUT_MAX_CHARS) {
+    return tailLines;
+  }
+  return `... (truncated)
+${tailLines.slice(-UNMAPPED_FAILURE_OUTPUT_MAX_CHARS)}`;
+}
+function buildFailureRecords(testSummary, rawOutput) {
+  if (testSummary.failures.length > 0) {
+    return testSummary.failures.map((failure) => ({
+      test: failure.testName,
+      file: failure.file,
+      message: failure.error,
+      output: failure.stackTrace.length > 0 ? failure.stackTrace.join(`
+`) : undefined
+    }));
+  }
+  if (testSummary.failed === 0) {
+    return [];
+  }
+  return [
+    {
+      test: `Unmapped test failures (${testSummary.failed} detected)`,
+      message: "Structured test failure parsing returned no failure records. Diagnose the regression from the raw test output.",
+      output: rawOutput?.trim() ? truncateUnmappedFailureOutput(rawOutput.trim()) : undefined
+    }
+  ];
+}
+var UNMAPPED_FAILURE_OUTPUT_MAX_LINES = 200, UNMAPPED_FAILURE_OUTPUT_MAX_CHARS = 8000;
+
 // src/tdd/cleanup.ts
 async function getPgid(pid) {
   try {
@@ -40275,7 +40419,7 @@ async function runFullSuiteGate(story, config2, workdir, agentManager, implement
         failures: filteredFailures,
         failed: wasFiltered ? filteredFailures.length : testSummary.failed
       };
-      return await runRectificationLoop(story, config2, workdir, agentManager, implementerTier, lite, logger, filteredSummary, rectificationConfig, effectiveTestCmd, fullSuiteTimeout, featureName, projectDir);
+      return await runRectificationLoop(story, config2, workdir, agentManager, implementerTier, lite, logger, filteredSummary, rectificationConfig, effectiveTestCmd, fullSuiteTimeout, fullSuiteResult.output, featureName, projectDir);
     }
     if (testSummary.passed > 0) {
       logger.info("tdd", "Full suite gate passed (non-zero exit, 0 failures, tests detected)", {
@@ -40303,7 +40447,7 @@ async function runFullSuiteGate(story, config2, workdir, agentManager, implement
   });
   return { passed: false, cost: 0 };
 }
-async function runRectificationLoop(story, config2, workdir, agentManager, implementerTier, lite, logger, testSummary, rectificationConfig, testCmd, fullSuiteTimeout, featureName, projectDir) {
+async function runRectificationLoop(story, config2, workdir, agentManager, implementerTier, lite, logger, testSummary, rectificationConfig, testCmd, fullSuiteTimeout, testOutput, featureName, projectDir) {
   const rectificationState = {
     attempt: 0,
     initialFailures: testSummary.failed,
@@ -40324,6 +40468,7 @@ async function runRectificationLoop(story, config2, workdir, agentManager, imple
     isolationPassed: true
   };
   let gateCostAccum = 0;
+  let currentTestOutput = testOutput;
   const fixed = await runSharedRectificationLoop({
     stage: "tdd",
     storyId: story.id,
@@ -40343,13 +40488,7 @@ async function runRectificationLoop(story, config2, workdir, agentManager, imple
     }),
     canContinue: (state) => state.isolationPassed && _rectificationGateDeps.shouldRetryRectification(state, rectificationConfig),
     buildPrompt: async () => {
-      const failureRecords = testSummary.failures.map((f) => ({
-        test: f.testName,
-        file: f.file,
-        message: f.error,
-        output: f.stackTrace.length > 0 ? f.stackTrace.join(`
-`) : undefined
-      }));
+      const failureRecords = buildFailureRecords(testSummary, currentTestOutput);
       return RectifierPromptBuilder.for("tdd-suite-failure").story(story).priorFailures(failureRecords).testCommand(testCmd).conventions().task().build();
     },
     runAttempt: async (attempt, rectificationPrompt) => {
@@ -40420,6 +40559,7 @@ async function runRectificationLoop(story, config2, workdir, agentManager, imple
       }
       if (retryFullSuite.output) {
         const newTestSummary = _rectificationGateDeps.parseTestOutput(retryFullSuite.output);
+        currentTestOutput = retryFullSuite.output;
         state.currentFailures = newTestSummary.failed;
         testSummary.failures = newTestSummary.failures;
         testSummary.failed = newTestSummary.failed;
@@ -42213,6 +42353,7 @@ async function _defaultRunDebate(storyId, stageConfig, prompt, config2, agentMan
   return { output, totalCostUsd };
 }
 async function runRectificationLoop2(opts) {
+  const loopStartMs = Date.now();
   const {
     config: config2,
     workdir,
@@ -42231,6 +42372,7 @@ async function runRectificationLoop2(opts) {
   const agentManager = opts.agentManager ?? _rectificationDeps.createManager(config2);
   const rectificationConfig = config2.execution.rectification;
   const testSummary = parseTestOutput(testOutput);
+  let currentTestOutput = testOutput;
   const label = promptPrefix ? "regression rectification" : "rectification";
   const rectificationState = {
     attempt: 0,
@@ -42290,13 +42432,7 @@ ${debateResult.output}`;
           });
         }
       }
-      const failureRecords = testSummary.failures.map((f) => ({
-        test: f.testName,
-        file: f.file,
-        message: f.error,
-        output: f.stackTrace.length > 0 ? f.stackTrace.join(`
-`) : undefined
-      }));
+      const failureRecords = buildFailureRecords(testSummary, currentTestOutput);
       let rectificationPrompt = await RectifierPromptBuilder.for("verify-failure").story(story).priorFailures(failureRecords).testCommand(testCommand).conventions().task().build();
       if (diagnosisPrefix) {
         rectificationPrompt = `${diagnosisPrefix}
@@ -42379,6 +42515,7 @@ ${rectificationPrompt}`;
       }
       if (retryVerification.output) {
         const newTestSummary = parseTestOutput(retryVerification.output);
+        currentTestOutput = retryVerification.output;
         state.currentFailures = newTestSummary.failed;
         state.lastExitCode = retryVerification.status === "SUCCESS" ? 0 : 1;
         testSummary.failures = newTestSummary.failures;
@@ -42504,7 +42641,7 @@ ${escalationPrompt}`;
     }
     throw error48;
   });
-  return { succeeded, cost: costAccum };
+  return { succeeded, cost: costAccum, durationMs: Date.now() - loopStartMs };
 }
 function runRectificationLoopFromCtx(ctx, opts) {
   return runRectificationLoop2({
@@ -44853,7 +44990,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@nathapp/nax",
-    version: "0.63.0",
+    version: "0.63.1",
     description: "AI Coding Agent Orchestrator \u2014 loops until done",
     type: "module",
     bin: {
@@ -44934,8 +45071,8 @@ var init_version = __esm(() => {
   NAX_VERSION = package_default.version;
   NAX_COMMIT = (() => {
     try {
-      if (/^[0-9a-f]{6,10}$/.test("e6619c9f"))
-        return "e6619c9f";
+      if (/^[0-9a-f]{6,10}$/.test("579e7fb0"))
+        return "579e7fb0";
     } catch {}
     try {
       const result = Bun.spawnSync(["git", "rev-parse", "--short", "HEAD"], {
@@ -45976,7 +46113,10 @@ async function runDeferredRegression(options) {
       failedTestFiles: [],
       passedTests: 0,
       rectificationAttempts: 0,
-      affectedStories: []
+      affectedStories: [],
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   if (regressionMode !== "deferred") {
@@ -45987,7 +46127,10 @@ async function runDeferredRegression(options) {
       failedTestFiles: [],
       passedTests: 0,
       rectificationAttempts: 0,
-      affectedStories: []
+      affectedStories: [],
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   const testCommand = config2.quality.commands.test ?? "bun test";
@@ -46017,7 +46160,10 @@ async function runDeferredRegression(options) {
       failedTestFiles: [],
       passedTests: 0,
       rectificationAttempts: 0,
-      affectedStories: []
+      affectedStories: [],
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   logger?.info("regression", "Running deferred full-suite regression gate", {
@@ -46033,7 +46179,10 @@ async function runDeferredRegression(options) {
       failedTestFiles: [],
       passedTests: fullSuiteResult.passCount ?? 0,
       rectificationAttempts: 0,
-      affectedStories: []
+      affectedStories: [],
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   if (fullSuiteResult.status === "TIMEOUT" && acceptOnTimeout) {
@@ -46044,7 +46193,10 @@ async function runDeferredRegression(options) {
       failedTestFiles: [],
       passedTests: 0,
       rectificationAttempts: 0,
-      affectedStories: []
+      affectedStories: [],
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   if (!fullSuiteResult.output) {
@@ -46055,7 +46207,10 @@ async function runDeferredRegression(options) {
       failedTestFiles: [],
       passedTests: fullSuiteResult.passCount ?? 0,
       rectificationAttempts: 0,
-      affectedStories: []
+      affectedStories: [],
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   const testSummary = _regressionDeps.parseTestOutput(fullSuiteResult.output);
@@ -46067,7 +46222,10 @@ async function runDeferredRegression(options) {
       failedTestFiles: [],
       passedTests: 0,
       rectificationAttempts: 0,
-      affectedStories: []
+      affectedStories: [],
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   const affectedStories = new Set;
@@ -46108,18 +46266,24 @@ async function runDeferredRegression(options) {
       failedTestFiles: Array.from(testFilesInFailures),
       passedTests: testSummary.passed,
       rectificationAttempts: 0,
-      affectedStories: Array.from(affectedStories)
+      affectedStories: Array.from(affectedStories),
+      storyCosts: {},
+      storyDurations: {},
+      storyOutcomes: {}
     };
   }
   let rectificationAttempts = 0;
   let storiesRectified = 0;
   let currentTestOutput = fullSuiteResult.output;
   const affectedStoriesList = Array.from(affectedStoriesObjs.values());
+  const storyCostAccum = {};
+  const storyDurationAccum = {};
+  const storyOutcomeAccum = {};
   for (const story of affectedStoriesList) {
     for (let attempt = 0;attempt < maxRectificationAttempts; attempt++) {
       rectificationAttempts++;
       logger?.info("regression", `Rectifying story ${story.id} (attempt ${attempt + 1}/${maxRectificationAttempts})`);
-      const fixed = await _regressionDeps.runRectificationLoop({
+      const rectResult = await _regressionDeps.runRectificationLoop({
         config: config2,
         workdir,
         story,
@@ -46132,7 +46296,12 @@ Your story ${story.id} broke tests in the full suite. Fix these regressions.`,
         agentManager,
         featureName: prd.feature
       });
-      if (fixed) {
+      storyCostAccum[story.id] = (storyCostAccum[story.id] ?? 0) + rectResult.cost;
+      storyDurationAccum[story.id] = (storyDurationAccum[story.id] ?? 0) + rectResult.durationMs;
+      if (!storyOutcomeAccum[story.id]) {
+        storyOutcomeAccum[story.id] = rectResult.succeeded;
+      }
+      if (rectResult.succeeded) {
         storiesRectified++;
         logger?.info("regression", `Story ${story.id} rectified successfully`);
         logger?.info("regression", "Re-running full suite after story rectification", {
@@ -46155,7 +46324,10 @@ Your story ${story.id} broke tests in the full suite. Fix these regressions.`,
             failedTestFiles: Array.from(testFilesInFailures),
             passedTests: midResult.passCount ?? 0,
             rectificationAttempts,
-            affectedStories: Array.from(affectedStories)
+            affectedStories: Array.from(affectedStories),
+            storyCosts: storyCostAccum,
+            storyDurations: storyDurationAccum,
+            storyOutcomes: storyOutcomeAccum
           };
         }
         logger?.warn("regression", "Full suite still failing after story rectification \u2014 continuing", {
@@ -46185,7 +46357,10 @@ Your story ${story.id} broke tests in the full suite. Fix these regressions.`,
     failedTestFiles: Array.from(testFilesInFailures),
     passedTests: retryResult.passCount ?? 0,
     rectificationAttempts,
-    affectedStories: Array.from(affectedStories)
+    affectedStories: Array.from(affectedStories),
+    storyCosts: storyCostAccum,
+    storyDurations: storyDurationAccum,
+    storyOutcomes: storyOutcomeAccum
   };
 }
 var _regressionDeps;
@@ -46286,6 +46461,50 @@ async function handleRunCompletion(options) {
           }, workdir);
         }
       }
+      const regressionStoryCosts = regressionResult.storyCosts ?? {};
+      const regressionStoryDurations = regressionResult.storyDurations ?? {};
+      const regressionStoryOutcomes = regressionResult.storyOutcomes ?? {};
+      if (Object.keys(regressionStoryCosts).length > 0) {
+        const existingIndex = new Map(allStoryMetrics.map((m, i) => [m.storyId, i]));
+        const rectCompletedAt = new Date().toISOString();
+        const defaultAgent = options.agentManager?.getDefault() ?? resolveDefaultAgent(config2);
+        for (const [storyId, storyCost] of Object.entries(regressionStoryCosts)) {
+          const storyDuration = regressionStoryDurations[storyId] ?? 0;
+          const storySuccess = regressionStoryOutcomes[storyId] ?? regressionResult.success;
+          const existingIdx = existingIndex.get(storyId);
+          if (existingIdx === undefined) {
+            const regrStory = prd.userStories.find((s) => s.id === storyId);
+            allStoryMetrics.push({
+              storyId,
+              complexity: regrStory?.routing?.complexity ?? "medium",
+              modelTier: "balanced",
+              modelUsed: defaultAgent,
+              attempts: 1,
+              finalTier: "balanced",
+              success: storySuccess,
+              cost: storyCost,
+              durationMs: storyDuration,
+              firstPassSuccess: false,
+              startedAt: rectCompletedAt,
+              completedAt: rectCompletedAt,
+              source: "rectification",
+              rectificationCost: storyCost,
+              fullSuiteGatePassed: false,
+              runtimeCrashes: 0
+            });
+          } else {
+            const existing = allStoryMetrics[existingIdx];
+            allStoryMetrics[existingIdx] = {
+              ...existing,
+              cost: existing.cost + storyCost,
+              durationMs: existing.durationMs + storyDuration,
+              rectificationCost: (existing.rectificationCost ?? 0) + storyCost,
+              firstPassSuccess: false,
+              success: existing.success && storySuccess
+            };
+          }
+        }
+      }
     }
   }
   const durationMs = Date.now() - startTime;
@@ -46392,6 +46611,7 @@ async function handleRunCompletion(options) {
 }
 var _runCompletionDeps;
 var init_run_completion = __esm(() => {
+  init_agents();
   init_runner4();
   init_logger2();
   init_metrics();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nathapp/nax",
-  "version": "0.63.0",
+  "version": "0.63.1",
   "description": "AI Coding Agent Orchestrator — loops until done",
   "type": "module",
   "bin": {