@dtlabs/ai-toolkit 0.17.0 → 0.17.1

package/CHANGELOG.md

@@ -1,4 +1,12 @@
 ## 0.1.4 (2026-07-21)
+## [0.17.1](https://github.com/dtlab-tech/ai-toolkit/compare/v0.17.0...v0.17.1) (2026-10-08)
+
+
+### Bug Fixes
+
+* **task-executor:** persist the reviewer's raw response on its receipt ([487b267](https://github.com/dtlab-tech/ai-toolkit/commit/487b2676467b72d4f6649c784b9fc6311d7819ac))
+* **task-executor:** persist the reviewer's raw response on its receipt ([1ca8491](https://github.com/dtlab-tech/ai-toolkit/commit/1ca8491e81ea35cc27a5f5b9f2b970fe85b1aeef))
+
 ## [0.17.0](https://github.com/dtlab-tech/ai-toolkit/compare/v0.16.9...v0.17.0) (2026-10-08)
 
 

package/lib/task-executor/index.js

@@ -715,7 +715,7 @@ function _extractReviewVerdict(spawnResult) {
     : null;
 
   if (text === null) {
-    return { passed: null, criticalFindings: [] };
+    return { passed: null, criticalFindings: [], responseText: null };
   }
 
   const verdictMatch = text.match(/Verdict:\s*(PASS|FAIL)/i);
@@ -754,7 +754,16 @@ function _extractReviewVerdict(spawnResult) {
     }
   }
 
-  return { passed: passed, criticalFindings: criticalFindings };
+  // Persisted verbatim on the review receipt (persistVerificationReviewOutcome)
+  // so a FAIL this function cannot fully explain — any future template
+  // variation neither the CRITICAL section nor the Build/Tests fold-in
+  // above recognizes — is still diagnosable from the receipt itself,
+  // instead of requiring the diff to be reconstructed and re-judged by hand
+  // after the fact (confirmed necessary against a real run: this exact
+  // "reviewPassed:false, criticalFindings:[]" shape recurred on a THIRD,
+  // previously-clean task, meaning the Build/Tests fold-in does not cover
+  // every real cause of an unexplained FAIL).
+  return { passed: passed, criticalFindings: criticalFindings, responseText: text };
 }
 
 /**
@@ -824,7 +833,8 @@ function _extractReviewVerdict(spawnResult) {
  *   runId: string, taskId: string, attemptNumber: number,
  *   agentId: string, nativeAgentName: string,
  *   ledgerOperationId: string, spawnResult: object,
- *   reviewPassed: (boolean|null), criticalFindings: string[]
+ *   reviewPassed: (boolean|null), criticalFindings: string[],
+ *   responseText: (string|null)
  * }>}
  * @throws {Error} NO_LOCK_HELD, AGENT_NOT_VERIFIED, DEFINITION_HASH_MISMATCH,
  *   DISPATCH_VALIDATION_ERROR, or any error surfaced by spawnClaudeAgent
@@ -918,6 +928,7 @@ async function runReview(args) {
     spawnResult: spawnResult,
     reviewPassed: verdict.passed,
     criticalFindings: verdict.criticalFindings,
+    responseText: verdict.responseText,
   };
 }
 

package/lib/task-executor/store.js

@@ -777,6 +777,14 @@ function persistVerificationReviewOutcome(executionRoot, runId, taskId, attemptN
       attemptNumber: attemptNumber,
       reviewPassed: review.reviewPassed === undefined ? null : review.reviewPassed,
       criticalFindings: review.criticalFindings,
+      // Verbatim review-solution reply, persisted so a FAIL neither the
+      // CRITICAL section nor the Build/Tests fold-in can explain
+      // (_extractReviewVerdict's own criticalFindings: [] either way) is
+      // still diagnosable from this receipt — never requires reconstructing
+      // and re-judging the diff by hand after the fact. null (never '') when
+      // the reply had no readable envelope at all — see
+      // _extractReviewVerdict's own null-compatibility.
+      responseText: review.responseText === undefined ? null : review.responseText,
       authoritative: verificationPassed === true,
     });
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dtlabs/ai-toolkit",
-  "version": "0.17.0",
+  "version": "0.17.1",
   "description": "AI Toolkit — reusable agents and procedures for developers to build AI-powered applications.",
   "files": [
     "docs/",