@zhixuan92/multi-model-agent-core 4.0.3 → 4.0.4

package/dist/review/reviewer-prompt-builder.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"reviewer-prompt-builder.d.ts","sourceRoot":"","sources":["../../src/review/reviewer-prompt-builder.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAE5D,MAAM,MAAM,kBAAkB,GAAG,UAAU,GAAG,cAAc,GAAG,OAAO,GAAG,QAAQ,GAAG,QAAQ,GAAG,aAAa,GAAG,OAAO,GAAG,SAAS,CAAC;AAEnI,qBAAa,qBAAqB;IAE9B,OAAO,CAAC,SAAS;IAKjB,OAAO,CAAC,gBAAgB;gBALhB,SAAS,EAAE;QACjB,IAAI,EAAE,cAAc,CAAC;QACrB,YAAY,EAAE,cAAc,CAAC;QAC7B,IAAI,EAAE,cAAc,CAAC;KACtB,EACO,gBAAgB,GAAE,OAAO,CAAC,MAAM,CAAC,kBAAkB,EAAE,cAAc,CAAC,CAAM;IAGpF,SAAS,CAAC,GAAG,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,GAAG;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE;IAOrG,cAAc,CAAC,GAAG,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,kBAAkB,CAAA;KAAE,GAAG;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE;IAQtI,SAAS,CAAC,GAAG,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,GAAG;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE;CAMtG"}
+{"version":3,"file":"reviewer-prompt-builder.d.ts","sourceRoot":"","sources":["../../src/review/reviewer-prompt-builder.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,qBAAqB,EAAE,MAAM,uBAAuB,CAAC;AAEnF,MAAM,MAAM,kBAAkB,GAAG,UAAU,GAAG,cAAc,GAAG,OAAO,GAAG,QAAQ,GAAG,QAAQ,GAAG,aAAa,GAAG,OAAO,GAAG,SAAS,CAAC;AAEnI,qBAAa,qBAAqB;IAE9B,OAAO,CAAC,SAAS;IAKjB,OAAO,CAAC,gBAAgB;gBALhB,SAAS,EAAE;QACjB,IAAI,EAAE,cAAc,CAAC;QACrB,YAAY,EAAE,cAAc,CAAC;QAC7B,IAAI,EAAE,cAAc,CAAC;KACtB,EACO,gBAAgB,GAAE,OAAO,CAAC,MAAM,CAAC,kBAAkB,EAAE,cAAc,CAAC,CAAM;IAGpF,SAAS,CAAC,GAAG,EAAE,qBAAqB,GAAG;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE;IAOnF,cAAc,CAAC,GAAG,EAAE,qBAAqB,GAAG;QAAE,KAAK,CAAC,EAAE,kBAAkB,CAAA;KAAE,GAAG;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE;IAQzH,SAAS,CAAC,GAAG,EAAE,qBAAqB,GAAG;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE;CAMpF"}

package/dist/review/reviewer-prompt-builder.js.map

@@ -1 +1 @@
-{"version":3,"file":"reviewer-prompt-builder.js","sourceRoot":"","sources":["../../src/review/reviewer-prompt-builder.ts"],"names":[],"mappings":"AAIA,MAAM,OAAO,qBAAqB;IAEtB;IAKA;IANV,YACU,SAIP,EACO,mBAAwE,EAAE;QAL1E,cAAS,GAAT,SAAS,CAIhB;QACO,qBAAgB,GAAhB,gBAAgB,CAA0D;IACjF,CAAC;IAEJ,SAAS,CAAC,GAA4C;QACpD,OAAO;YACL,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY;YAC9C,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC;SACrD,CAAC;IACJ,CAAC;IAED,cAAc,CAAC,GAAwE;QACrF,MAAM,QAAQ,GAAG,CAAC,GAAG,CAAC,KAAK,KAAK,SAAS,IAAI,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC;QAC9G,OAAO;YACL,YAAY,EAAE,QAAQ,CAAC,YAAY;YACnC,UAAU,EAAE,QAAQ,CAAC,eAAe,CAAC,GAAG,CAAC;SAC1C,CAAC;IACJ,CAAC;IAED,SAAS,CAAC,GAA4C;QACpD,OAAO;YACL,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY;YAC9C,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC;SACrD,CAAC;IACJ,CAAC;CACF"}
+{"version":3,"file":"reviewer-prompt-builder.js","sourceRoot":"","sources":["../../src/review/reviewer-prompt-builder.ts"],"names":[],"mappings":"AAIA,MAAM,OAAO,qBAAqB;IAEtB;IAKA;IANV,YACU,SAIP,EACO,mBAAwE,EAAE;QAL1E,cAAS,GAAT,SAAS,CAIhB;QACO,qBAAgB,GAAhB,gBAAgB,CAA0D;IACjF,CAAC;IAEJ,SAAS,CAAC,GAA0B;QAClC,OAAO;YACL,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY;YAC9C,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC;SACrD,CAAC;IACJ,CAAC;IAED,cAAc,CAAC,GAA2D;QACxE,MAAM,QAAQ,GAAG,CAAC,GAAG,CAAC,KAAK,KAAK,SAAS,IAAI,IAAI,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC;QAC9G,OAAO;YACL,YAAY,EAAE,QAAQ,CAAC,YAAY;YACnC,UAAU,EAAE,QAAQ,CAAC,eAAe,CAAC,GAAG,CAAC;SAC1C,CAAC;IACJ,CAAC;IAED,SAAS,CAAC,GAA0B;QAClC,OAAO;YACL,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,YAAY;YAC9C,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC;SACrD,CAAC;IACJ,CAAC;CACF"}

package/dist/review/templates/diff-review.d.ts

@@ -1,3 +1,11 @@
 import type { ReviewTemplate } from './shared.js';
+/**
+ * Diff reviewer — the final cross-check that looks at the cumulative
+ * diff in isolation and decides if it's acceptable as a whole.
+ *
+ * Tool sweep #6 fix: the template name was a misnomer pre-fix —
+ * the reviewer received only the worker's text summary, never the
+ * actual diff. Now it sees the diff (matches the name).
+ */
 export declare const diffTemplate: ReviewTemplate;
 //# sourceMappingURL=diff-review.d.ts.map

package/dist/review/templates/diff-review.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"diff-review.d.ts","sourceRoot":"","sources":["../../../src/review/templates/diff-review.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD,eAAO,MAAM,YAAY,EAAE,cAK1B,CAAC"}
+{"version":3,"file":"diff-review.d.ts","sourceRoot":"","sources":["../../../src/review/templates/diff-review.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD;;;;;;;GAOG;AACH,eAAO,MAAM,YAAY,EAAE,cAgC1B,CAAC"}

package/dist/review/templates/diff-review.js

@@ -1,7 +1,39 @@
+/**
+ * Diff reviewer — the final cross-check that looks at the cumulative
+ * diff in isolation and decides if it's acceptable as a whole.
+ *
+ * Tool sweep #6 fix: the template name was a misnomer pre-fix —
+ * the reviewer received only the worker's text summary, never the
+ * actual diff. Now it sees the diff (matches the name).
+ */
 export const diffTemplate = {
-    systemPrompt: `You are reviewing a diff. Reply with EXACTLY one of: APPROVE, CONCERNS: <reasons>, or REJECT: <reason>`,
+    systemPrompt: [
+        'You are reviewing the cumulative diff produced by all rework rounds. Decide whether the overall change is acceptable.',
+        '',
+        'Reply with EXACTLY one of these single-line verdicts:',
+        '- `APPROVE` — the diff is acceptable as-is.',
+        '- `CONCERNS: <one-line summary of concerns>` — the diff has flaws but is on the right track.',
+        '- `REJECT: <one-line reason>` — the diff is wrong enough that it should not be applied.',
+        '',
+        'Decision criteria:',
+        '- Is the diff scoped to what the brief asked for? Out-of-scope edits are a CONCERN at minimum.',
+        '- Are the changes internally consistent (e.g., a renamed symbol updated everywhere it appears in the diff)?',
+        '- Does the diff introduce obvious correctness or security issues that the spec/quality reviews missed?',
+        '',
+        'Do NOT re-litigate the brief itself — that is the spec reviewer\'s job. Focus on the diff as a whole.',
+    ].join('\n'),
     buildUserPrompt(ctx) {
-        return `Task: ${ctx.brief}\n\nWorker output:\n${ctx.workerOutput}`;
+        const parts = [];
+        parts.push(`# Task brief\n${ctx.brief}`);
+        parts.push(`# Worker's most recent summary\n${ctx.workerOutput || '(no summary)'}`);
+        if (ctx.diff && ctx.diff.length > 0) {
+            parts.push(`# Cumulative diff (what you are reviewing)\n\n\`\`\`diff\n${ctx.diff}\n\`\`\``);
+        }
+        else {
+            parts.push(`# Cumulative diff\n(no file changes detected — APPROVE only if the brief requested a no-op)`);
+        }
+        parts.push(`# Decide\nReply with one of APPROVE, CONCERNS: <text>, or REJECT: <text>.`);
+        return parts.join('\n\n');
     },
 };
 //# sourceMappingURL=diff-review.js.map

package/dist/review/templates/diff-review.js.map

@@ -1 +1 @@
-{"version":3,"file":"diff-review.js","sourceRoot":"","sources":["../../../src/review/templates/diff-review.ts"],"names":[],"mappings":"AAEA,MAAM,CAAC,MAAM,YAAY,GAAmB;IAC1C,YAAY,EAAE,wGAAwG;IACtH,eAAe,CAAC,GAAG;QACjB,OAAO,SAAS,GAAG,CAAC,KAAK,uBAAuB,GAAG,CAAC,YAAY,EAAE,CAAC;IACrE,CAAC;CACF,CAAC"}
+{"version":3,"file":"diff-review.js","sourceRoot":"","sources":["../../../src/review/templates/diff-review.ts"],"names":[],"mappings":"AAEA;;;;;;;GAOG;AACH,MAAM,CAAC,MAAM,YAAY,GAAmB;IAC1C,YAAY,EAAE;QACZ,uHAAuH;QACvH,EAAE;QACF,uDAAuD;QACvD,6CAA6C;QAC7C,8FAA8F;QAC9F,yFAAyF;QACzF,EAAE;QACF,oBAAoB;QACpB,gGAAgG;QAChG,6GAA6G;QAC7G,wGAAwG;QACxG,EAAE;QACF,uGAAuG;KACxG,CAAC,IAAI,CAAC,IAAI,CAAC;IAEZ,eAAe,CAAC,GAAG;QACjB,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,iBAAiB,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC;QACzC,KAAK,CAAC,IAAI,CAAC,mCAAmC,GAAG,CAAC,YAAY,IAAI,cAAc,EAAE,CAAC,CAAC;QAEpF,IAAI,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CAAC,6DAA6D,GAAG,CAAC,IAAI,UAAU,CAAC,CAAC;QAC9F,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC,6FAA6F,CAAC,CAAC;QAC5G,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,2EAA2E,CAAC,CAAC;QAExF,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;CACF,CAAC"}

package/dist/review/templates/finding-criteria.d.ts

@@ -0,0 +1,39 @@
+/**
+ * Shared finding-quality criteria — the single source of truth for what
+ * makes a "good finding" across the implementer + annotator stages.
+ *
+ * Tool sweep #12: the implementer prompts didn't share the annotator's
+ * rubric, so workers emitted weak narrative (miscalibrated severity,
+ * unsupported claims, speculative scope) and the annotator had to
+ * either downgrade everything or rubber-stamp. This module gives
+ * implementer prompts the SAME calibration the annotator uses, so the
+ * two stages converge:
+ *
+ *   - Implementer is told what counts as evidence, what severity means,
+ *     and what's out of scope BEFORE it writes findings.
+ *   - Annotator validates the worker emitted them properly (its rubric
+ *     in `annotator-shared.ts` references the same definitions).
+ *
+ * Result: fewer false positives, fewer missed criticals, less rework
+ * (the worker self-aligned with what the reviewer will check). No
+ * heuristic short-circuits — both stages run, they just spend less
+ * time correcting each other.
+ */
+/** Severity ladder for read-only finding emission AND annotator
+ *  validation. Same words, same meaning across stages. */
+export declare const SEVERITY_LADDER: string;
+/** Evidence-grounding rule. Required for every finding the worker
+ *  emits. The annotator rejects findings without quotable evidence. */
+export declare const EVIDENCE_GROUNDING: string;
+/** Scope discipline. Workers commonly hallucinate citations into files
+ *  they didn't open; this rule blocks the most common failure mode. */
+export declare const SCOPE_DISCIPLINE: string;
+/** What the annotator (which doubles as the read-only quality reviewer)
+ *  will check the worker's output against. Pre-fix the worker had no
+ *  visibility into this; now it can self-align. */
+export declare const ANNOTATOR_CHECK_AWARENESS_RO: string;
+/** Awareness block for delegate / execute-plan / retry implementers.
+ *  Tells the worker what spec_review and quality_review will judge,
+ *  so the worker self-aligns. */
+export declare const REVIEWER_AWARENESS_AP: string;
+//# sourceMappingURL=finding-criteria.d.ts.map

package/dist/review/templates/finding-criteria.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"finding-criteria.d.ts","sourceRoot":"","sources":["../../../src/review/templates/finding-criteria.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH;0DAC0D;AAC1D,eAAO,MAAM,eAAe,QAOhB,CAAC;AAEb;uEACuE;AACvE,eAAO,MAAM,kBAAkB,QAKnB,CAAC;AAEb;uEACuE;AACvE,eAAO,MAAM,gBAAgB,QAKjB,CAAC;AAEb;;mDAEmD;AACnD,eAAO,MAAM,4BAA4B,QAO7B,CAAC;AAEb;;iCAEiC;AACjC,eAAO,MAAM,qBAAqB,QAiBtB,CAAC"}

package/dist/review/templates/finding-criteria.js

@@ -0,0 +1,80 @@
+/**
+ * Shared finding-quality criteria — the single source of truth for what
+ * makes a "good finding" across the implementer + annotator stages.
+ *
+ * Tool sweep #12: the implementer prompts didn't share the annotator's
+ * rubric, so workers emitted weak narrative (miscalibrated severity,
+ * unsupported claims, speculative scope) and the annotator had to
+ * either downgrade everything or rubber-stamp. This module gives
+ * implementer prompts the SAME calibration the annotator uses, so the
+ * two stages converge:
+ *
+ *   - Implementer is told what counts as evidence, what severity means,
+ *     and what's out of scope BEFORE it writes findings.
+ *   - Annotator validates the worker emitted them properly (its rubric
+ *     in `annotator-shared.ts` references the same definitions).
+ *
+ * Result: fewer false positives, fewer missed criticals, less rework
+ * (the worker self-aligned with what the reviewer will check). No
+ * heuristic short-circuits — both stages run, they just spend less
+ * time correcting each other.
+ */
+/** Severity ladder for read-only finding emission AND annotator
+ *  validation. Same words, same meaning across stages. */
+export const SEVERITY_LADDER = [
+    'Severity (your judgment, calibrated to impact):',
+    '- critical: must fix first — RCE, data loss, auth bypass, build broken, data corruption.',
+    '- high:     real bug or security gap; blocks release.',
+    '- medium:   real issue; fix soon; not blocking.',
+    '- low:      minor or cosmetic issue; nice-to-fix; PASS in checklist contexts.',
+    'Calibrate to actual impact, not how alarming the wording sounds. Workers commonly inflate — resist the urge.',
+].join('\n');
+/** Evidence-grounding rule. Required for every finding the worker
+ *  emits. The annotator rejects findings without quotable evidence. */
+export const EVIDENCE_GROUNDING = [
+    'Evidence grounding (REQUIRED for every finding):',
+    '- Cite `file:line` (or `file:line-line` for a span) where the issue lives.',
+    '- Quote the exact code excerpt or command output that demonstrates the issue. Don\'t paraphrase — quote.',
+    '- If you cannot quote evidence directly, do NOT raise the finding. Note "investigation needed" in your summary instead.',
+].join('\n');
+/** Scope discipline. Workers commonly hallucinate citations into files
+ *  they didn't open; this rule blocks the most common failure mode. */
+export const SCOPE_DISCIPLINE = [
+    'Scope discipline:',
+    '- Only flag issues you verified by reading the file directly.',
+    '- If a claim depends on caller behavior or files outside the requested scope, mark "investigation needed" in your summary — do NOT speculate.',
+    '- Stay within the requested files. Don\'t audit dependencies you weren\'t asked about.',
+].join('\n');
+/** What the annotator (which doubles as the read-only quality reviewer)
+ *  will check the worker's output against. Pre-fix the worker had no
+ *  visibility into this; now it can self-align. */
+export const ANNOTATOR_CHECK_AWARENESS_RO = [
+    'After your output, an annotator validates each finding against this rubric:',
+    '- Is the severity calibrated to actual impact (or did you inflate)?',
+    '- Does the evidence directly support the claim, or is it paraphrased?',
+    '- Is the finding within scope, or is it speculation about untouched files?',
+    '- Is the finding on-brief (matching the focus / audit type / checklist item)?',
+    'Findings that fail any of these are downgraded or dropped. Self-check before emitting.',
+].join('\n');
+/** Awareness block for delegate / execute-plan / retry implementers.
+ *  Tells the worker what spec_review and quality_review will judge,
+ *  so the worker self-aligns. */
+export const REVIEWER_AWARENESS_AP = [
+    'After your edit, two reviewers see the cumulative diff (every change since task start):',
+    '',
+    '1. SPEC reviewer — "does the diff fulfill the brief?"',
+    '   - APPROVED requires: every brief item present in the diff, no missing pieces, no out-of-scope edits.',
+    '   - CHANGES_REQUIRED requires: a concrete concern tied to a specific diff line.',
+    '   - Empty diff = changes_required UNLESS the brief explicitly requested a no-op.',
+    '',
+    '2. QUALITY reviewer — "is the diff sound, safe, maintainable?"',
+    '   - Flags: correctness bugs, broken tests, races, security gaps, speculative dependencies.',
+    '   - Does NOT flag: stylistic preferences, comment bikeshedding, unrelated pre-existing code.',
+    '',
+    'Self-check before declaring done:',
+    '- Implemented EVERY item in the brief? (no gaps)',
+    '- No edits OUTSIDE the brief? (no scope creep)',
+    '- If your edit affects callers / tests not in your filePaths, note them in your summary — even if you can\'t update them.',
+    '- Are the changes minimal and verifiable, or could they be smaller?',
+].join('\n');
+//# sourceMappingURL=finding-criteria.js.map

package/dist/review/templates/finding-criteria.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"finding-criteria.js","sourceRoot":"","sources":["../../../src/review/templates/finding-criteria.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH;0DAC0D;AAC1D,MAAM,CAAC,MAAM,eAAe,GAAG;IAC7B,iDAAiD;IACjD,0FAA0F;IAC1F,uDAAuD;IACvD,iDAAiD;IACjD,+EAA+E;IAC/E,8GAA8G;CAC/G,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb;uEACuE;AACvE,MAAM,CAAC,MAAM,kBAAkB,GAAG;IAChC,kDAAkD;IAClD,4EAA4E;IAC5E,0GAA0G;IAC1G,yHAAyH;CAC1H,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb;uEACuE;AACvE,MAAM,CAAC,MAAM,gBAAgB,GAAG;IAC9B,mBAAmB;IACnB,+DAA+D;IAC/D,+IAA+I;IAC/I,wFAAwF;CACzF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb;;mDAEmD;AACnD,MAAM,CAAC,MAAM,4BAA4B,GAAG;IAC1C,6EAA6E;IAC7E,qEAAqE;IACrE,uEAAuE;IACvE,4EAA4E;IAC5E,+EAA+E;IAC/E,wFAAwF;CACzF,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb;;iCAEiC;AACjC,MAAM,CAAC,MAAM,qBAAqB,GAAG;IACnC,yFAAyF;IACzF,EAAE;IACF,uDAAuD;IACvD,yGAAyG;IACzG,kFAAkF;IAClF,mFAAmF;IACnF,EAAE;IACF,gEAAgE;IAChE,6FAA6F;IAC7F,+FAA+F;IAC/F,EAAE;IACF,mCAAmC;IACnC,kDAAkD;IAClD,gDAAgD;IAChD,2HAA2H;IAC3H,qEAAqE;CACtE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC"}

package/dist/review/templates/quality-review-artifact.d.ts

@@ -1,3 +1,16 @@
 import type { ReviewTemplate } from './shared.js';
+/**
+ * Quality reviewer for artifact-producing routes (delegate, execute-plan,
+ * retry).
+ *
+ * Tool sweep #6: pre-fix this template only saw `Task: <brief>` +
+ * `Worker output: <text>`. Quality findings were rooted in the
+ * worker's prose claim — false positives ("the worker claimed X but
+ * didn't really") were common, and real regressions in the diff
+ * could be missed if the worker's summary glossed over them.
+ *
+ * Post-fix: reviewer sees the cumulative diff. Findings must be
+ * specific to diff lines.
+ */
 export declare const qualityAPTemplate: ReviewTemplate;
 //# sourceMappingURL=quality-review-artifact.d.ts.map

package/dist/review/templates/quality-review-artifact.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"quality-review-artifact.d.ts","sourceRoot":"","sources":["../../../src/review/templates/quality-review-artifact.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD,eAAO,MAAM,iBAAiB,EAAE,cAM/B,CAAC"}
+{"version":3,"file":"quality-review-artifact.d.ts","sourceRoot":"","sources":["../../../src/review/templates/quality-review-artifact.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD;;;;;;;;;;;;GAYG;AACH,eAAO,MAAM,iBAAiB,EAAE,cAsC/B,CAAC"}

package/dist/review/templates/quality-review-artifact.js

@@ -1,8 +1,46 @@
+/**
+ * Quality reviewer for artifact-producing routes (delegate, execute-plan,
+ * retry).
+ *
+ * Tool sweep #6: pre-fix this template only saw `Task: <brief>` +
+ * `Worker output: <text>`. Quality findings were rooted in the
+ * worker's prose claim — false positives ("the worker claimed X but
+ * didn't really") were common, and real regressions in the diff
+ * could be missed if the worker's summary glossed over them.
+ *
+ * Post-fix: reviewer sees the cumulative diff. Findings must be
+ * specific to diff lines.
+ */
 export const qualityAPTemplate = {
-    systemPrompt: `You are a code quality reviewer. Check whether the implementation is sound, safe, and maintainable.
-Return a JSON block with: {"verdict":"approved"|"concerns","concerns":["concern1",...]}`,
+    systemPrompt: [
+        'You are a code quality reviewer. Check whether the cumulative diff is sound, safe, and maintainable.',
+        '',
+        'Reply with a JSON block: {"verdict":"approved"|"concerns","concerns":["..."]}.',
+        '',
+        'Verdict rules:',
+        '- "approved": the diff has no quality concerns worth blocking on. The "concerns" list MUST be empty.',
+        '- "concerns": cite at least one specific concern, each tied to a specific diff line. Do not raise stylistic preferences not relevant to correctness, safety, or maintainability.',
+        '- An empty diff is "approved" (nothing to review).',
+        '',
+        'Examples of legitimate concerns: input not validated; resource not closed; race condition introduced; tests broken; new dependency introduced without justification.',
+        'Examples of out-of-scope concerns: bracket placement; comment punctuation; unrelated pre-existing code.',
+    ].join('\n'),
     buildUserPrompt(ctx) {
-        return `Task: ${ctx.brief}\n\nWorker output:\n${ctx.workerOutput}`;
+        const parts = [];
+        parts.push(`# Task brief\n${ctx.brief}`);
+        if (ctx.priorConcerns && ctx.priorConcerns.length > 0) {
+            parts.push(`# Prior reviewer concerns from earlier rounds in this chain\nVerify the rework has addressed each one:\n` +
+                ctx.priorConcerns.map((c, i) => `${i + 1}. ${c}`).join('\n'));
+        }
+        parts.push(`# Worker's most recent summary\n${ctx.workerOutput || '(no summary)'}`);
+        if (ctx.diff && ctx.diff.length > 0) {
+            parts.push(`# Cumulative diff (the truth of what changed)\n\n\`\`\`diff\n${ctx.diff}\n\`\`\``);
+        }
+        else {
+            parts.push(`# Cumulative diff\n(no file changes detected — verdict should be "approved")`);
+        }
+        parts.push(`# Decide\nIs the cumulative diff sound, safe, and maintainable? Reply with the JSON block specified in the system prompt.`);
+        return parts.join('\n\n');
     },
 };
 //# sourceMappingURL=quality-review-artifact.js.map

package/dist/review/templates/quality-review-artifact.js.map

@@ -1 +1 @@
-{"version":3,"file":"quality-review-artifact.js","sourceRoot":"","sources":["../../../src/review/templates/quality-review-artifact.ts"],"names":[],"mappings":"AAEA,MAAM,CAAC,MAAM,iBAAiB,GAAmB;IAC/C,YAAY,EAAE;wFACwE;IACtF,eAAe,CAAC,GAAG;QACjB,OAAO,SAAS,GAAG,CAAC,KAAK,uBAAuB,GAAG,CAAC,YAAY,EAAE,CAAC;IACrE,CAAC;CACF,CAAC"}
+{"version":3,"file":"quality-review-artifact.js","sourceRoot":"","sources":["../../../src/review/templates/quality-review-artifact.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAmB;IAC/C,YAAY,EAAE;QACZ,sGAAsG;QACtG,EAAE;QACF,gFAAgF;QAChF,EAAE;QACF,gBAAgB;QAChB,sGAAsG;QACtG,kLAAkL;QAClL,oDAAoD;QACpD,EAAE;QACF,sKAAsK;QACtK,yGAAyG;KAC1G,CAAC,IAAI,CAAC,IAAI,CAAC;IAEZ,eAAe,CAAC,GAAG;QACjB,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,iBAAiB,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC;QAEzC,IAAI,GAAG,CAAC,aAAa,IAAI,GAAG,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtD,KAAK,CAAC,IAAI,CACR,0GAA0G;gBAC1G,GAAG,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAC7D,CAAC;QACJ,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,mCAAmC,GAAG,CAAC,YAAY,IAAI,cAAc,EAAE,CAAC,CAAC;QAEpF,IAAI,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CAAC,gEAAgE,GAAG,CAAC,IAAI,UAAU,CAAC,CAAC;QACjG,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC,8EAA8E,CAAC,CAAC;QAC7F,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,2HAA2H,CAAC,CAAC;QAExI,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;CACF,CAAC"}

package/dist/review/templates/shared.d.ts

@@ -1,9 +1,26 @@
+export interface ReviewTemplateContext {
+    workerOutput: string;
+    brief: string;
+    filesChanged?: string[];
+    /**
+     * Cumulative unified diff of every change made since task start,
+     * across all rework rounds. Empty/undefined when no changes were
+     * detected (or the route is read-only).
+     *
+     * Tool sweep #6: reviewer templates were operating blind on the
+     * worker's text claim alone. Passing the diff lets each reviewer
+     * be precise — verdicts are grounded in code evidence, not prose.
+     */
+    diff?: string;
+    /**
+     * Concrete concerns from previous reviewer rounds in this chain.
+     * Round 1 sees an empty list; round 2+ sees what round 1 flagged
+     * so it can verify "did the rework address X" rather than re-deriving.
+     */
+    priorConcerns?: string[];
+}
 export interface ReviewTemplate {
     systemPrompt: string;
-    buildUserPrompt(ctx: {
-        workerOutput: string;
-        brief: string;
-        filesChanged?: string[];
-    }): string;
+    buildUserPrompt(ctx: ReviewTemplateContext): string;
 }
 //# sourceMappingURL=shared.d.ts.map

package/dist/review/templates/shared.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"shared.d.ts","sourceRoot":"","sources":["../../../src/review/templates/shared.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,GAAG,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,EAAE,CAAA;KAAE,GAAG,MAAM,CAAC;CAChG"}
+{"version":3,"file":"shared.d.ts","sourceRoot":"","sources":["../../../src/review/templates/shared.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,qBAAqB;IACpC,YAAY,EAAE,MAAM,CAAC;IACrB,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;IACxB;;;;;;;;OAQG;IACH,IAAI,CAAC,EAAE,MAAM,CAAC;IACd;;;;OAIG;IACH,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,cAAc;IAC7B,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,CAAC,GAAG,EAAE,qBAAqB,GAAG,MAAM,CAAC;CACrD"}

package/dist/review/templates/spec-review.d.ts

@@ -1,3 +1,18 @@
 import type { ReviewTemplate } from './shared.js';
+/**
+ * Spec compliance reviewer.
+ *
+ * Tool sweep #6 rewrite: pre-fix this template gave the LLM only
+ * `Task: <brief>` + `Worker output: <text>`. The reviewer was
+ * reviewing the worker's CLAIM, not the worker's WORK — it had no
+ * way to verify whether the claim was true. Result: skeptical
+ * reviewers defaulted to "changes_required" and triggered endless
+ * rework spirals on already-correct edits.
+ *
+ * Post-fix: the reviewer also sees the cumulative diff (the truth)
+ * and any prior reviewer concerns it should verify are addressed.
+ * With evidence in hand, "changes_required" must point to a
+ * specific diff line — no more vague rejections.
+ */
 export declare const specTemplate: ReviewTemplate;
 //# sourceMappingURL=spec-review.d.ts.map

package/dist/review/templates/spec-review.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"spec-review.d.ts","sourceRoot":"","sources":["../../../src/review/templates/spec-review.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD,eAAO,MAAM,YAAY,EAAE,cAM1B,CAAC"}
+{"version":3,"file":"spec-review.d.ts","sourceRoot":"","sources":["../../../src/review/templates/spec-review.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAElD;;;;;;;;;;;;;;GAcG;AACH,eAAO,MAAM,YAAY,EAAE,cAsC1B,CAAC"}

package/dist/review/templates/spec-review.js

@@ -1,8 +1,48 @@
+/**
+ * Spec compliance reviewer.
+ *
+ * Tool sweep #6 rewrite: pre-fix this template gave the LLM only
+ * `Task: <brief>` + `Worker output: <text>`. The reviewer was
+ * reviewing the worker's CLAIM, not the worker's WORK — it had no
+ * way to verify whether the claim was true. Result: skeptical
+ * reviewers defaulted to "changes_required" and triggered endless
+ * rework spirals on already-correct edits.
+ *
+ * Post-fix: the reviewer also sees the cumulative diff (the truth)
+ * and any prior reviewer concerns it should verify are addressed.
+ * With evidence in hand, "changes_required" must point to a
+ * specific diff line — no more vague rejections.
+ */
 export const specTemplate = {
-    systemPrompt: `You are a spec compliance reviewer. Check whether the implementer satisfied the task exactly.
-Return a JSON block with: {"verdict":"approved"|"changes_required","concerns":["concern1",...]}`,
+    systemPrompt: [
+        'You are a spec compliance reviewer. Your job is to decide whether the cumulative diff fulfills the task brief — nothing else.',
+        '',
+        'Reply with a JSON block: {"verdict":"approved"|"changes_required","concerns":["..."]}.',
+        '',
+        'Verdict rules:',
+        '- "approved": the diff implements the brief, with no missing or wrong elements. The "concerns" list MUST be empty.',
+        '- "changes_required": cite at least one concrete concern, each tied to a specific diff line or a specific missing element from the brief. Do NOT use this verdict for stylistic preferences not in the brief.',
+        '- An empty diff (no files changed) is "changes_required" UNLESS the brief explicitly requested a no-op or "no change needed".',
+        '- A diff that fully satisfies the brief is "approved" even if you would have written it differently.',
+        '',
+        'You do not see future rework rounds. Decide on this evidence alone.',
+    ].join('\n'),
     buildUserPrompt(ctx) {
-        return `Task: ${ctx.brief}\n\nWorker output:\n${ctx.workerOutput}`;
+        const parts = [];
+        parts.push(`# Task brief\n${ctx.brief}`);
+        if (ctx.priorConcerns && ctx.priorConcerns.length > 0) {
+            parts.push(`# Prior reviewer concerns from earlier rounds in this chain\nVerify the rework has addressed each one:\n` +
+                ctx.priorConcerns.map((c, i) => `${i + 1}. ${c}`).join('\n'));
+        }
+        parts.push(`# Worker's most recent summary\n${ctx.workerOutput || '(no summary)'}`);
+        if (ctx.diff && ctx.diff.length > 0) {
+            parts.push(`# Cumulative diff (the truth of what changed)\n\n\`\`\`diff\n${ctx.diff}\n\`\`\``);
+        }
+        else {
+            parts.push(`# Cumulative diff\n(no file changes detected)`);
+        }
+        parts.push(`# Decide\nDoes the cumulative diff fulfill the task brief? Reply with the JSON block specified in the system prompt.`);
+        return parts.join('\n\n');
     },
 };
 //# sourceMappingURL=spec-review.js.map

package/dist/review/templates/spec-review.js.map

@@ -1 +1 @@
-{"version":3,"file":"spec-review.js","sourceRoot":"","sources":["../../../src/review/templates/spec-review.ts"],"names":[],"mappings":"AAEA,MAAM,CAAC,MAAM,YAAY,GAAmB;IAC1C,YAAY,EAAE;gGACgF;IAC9F,eAAe,CAAC,GAAG;QACjB,OAAO,SAAS,GAAG,CAAC,KAAK,uBAAuB,GAAG,CAAC,YAAY,EAAE,CAAC;IACrE,CAAC;CACF,CAAC"}
+{"version":3,"file":"spec-review.js","sourceRoot":"","sources":["../../../src/review/templates/spec-review.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;GAcG;AACH,MAAM,CAAC,MAAM,YAAY,GAAmB;IAC1C,YAAY,EAAE;QACZ,+HAA+H;QAC/H,EAAE;QACF,wFAAwF;QACxF,EAAE;QACF,gBAAgB;QAChB,oHAAoH;QACpH,+MAA+M;QAC/M,+HAA+H;QAC/H,sGAAsG;QACtG,EAAE;QACF,qEAAqE;KACtE,CAAC,IAAI,CAAC,IAAI,CAAC;IAEZ,eAAe,CAAC,GAAG;QACjB,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,iBAAiB,GAAG,CAAC,KAAK,EAAE,CAAC,CAAC;QAEzC,IAAI,GAAG,CAAC,aAAa,IAAI,GAAG,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtD,KAAK,CAAC,IAAI,CACR,0GAA0G;gBAC1G,GAAG,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAC7D,CAAC;QACJ,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,mCAAmC,GAAG,CAAC,YAAY,IAAI,cAAc,EAAE,CAAC,CAAC;QAEpF,IAAI,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CAAC,gEAAgE,GAAG,CAAC,IAAI,UAAU,CAAC,CAAC;QACjG,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC,+CAA+C,CAAC,CAAC;QAC9D,CAAC;QAED,KAAK,CAAC,IAAI,CAAC,sHAAsH,CAAC,CAAC;QAEnI,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;CACF,CAAC"}

package/dist/tools/audit/tool-config.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-config.d.ts","sourceRoot":"","sources":["../../../src/tools/audit/tool-config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,6CAA6C,CAAC;AAClF,OAAO,EAAe,KAAK,KAAK,EAAE,MAAM,aAAa,CAAC;AAEtD,OAAO,EAAqB,KAAK,WAAW,EAAE,MAAM,qDAAqD,CAAC;AAE1G,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AAIvE,wBAAgB,aAAa,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI,CAYjE;AAID,MAAM,WAAW,cAAc;IAC7B,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAkCD,wBAAgB,cAAc,CAAC,KAAK,EAAE,KAAK,GAAG,cAAc,EAAE,CA0B7D;AAsDD,eAAO,MAAM,UAAU,EAAE,UAAU,CAAC,KAAK,EAAE,cAAc,EAAE,WAAW,CAyBrE,CAAC"}
+{"version":3,"file":"tool-config.d.ts","sourceRoot":"","sources":["../../../src/tools/audit/tool-config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,6CAA6C,CAAC;AAClF,OAAO,EAAe,KAAK,KAAK,EAAE,MAAM,aAAa,CAAC;AAEtD,OAAO,EAAqB,KAAK,WAAW,EAAE,MAAM,qDAAqD,CAAC;AAE1G,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AAUvE,wBAAgB,aAAa,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI,CAYjE;AAID,MAAM,WAAW,cAAc;IAC7B,aAAa,EAAE,MAAM,CAAC;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,EAAE,CAAC;IACpB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAkCD,wBAAgB,cAAc,CAAC,KAAK,EAAE,KAAK,GAAG,cAAc,EAAE,CA0B7D;AAgED,eAAO,MAAM,UAAU,EAAE,UAAU,CAAC,KAAK,EAAE,cAAc,EAAE,WAAW,CAyBrE,CAAC"}

package/dist/tools/audit/tool-config.js

@@ -3,6 +3,7 @@ import { qualityAuditTemplate } from '../../review/templates/quality-review-audi
 import { auditReportSchema } from '../../reporting/report-parser-slots/audit-report.js';
 import { auditHeadlineTemplate } from '../../reporting/headline-templates/audit.js';
 import { DEFAULT_TASK_TIMEOUT_MS } from '../../config/schema.js';
+import { SEVERITY_LADDER, EVIDENCE_GROUNDING, SCOPE_DISCIPLINE, ANNOTATOR_CHECK_AWARENESS_RO, } from '../../review/templates/finding-criteria.js';
 export function registerAudit(registry) {
     registry.register({
         routeName: 'audit',
@@ -73,7 +74,7 @@ export function auditBriefSlot(input) {
         }];
 }
 const FINDING_FORMAT_INSTRUCTIONS = [
-    'Produce a narrative audit report. Use this EXACT per-finding format so the deterministic extractor can recover findings if the structured reviewer pass fails:',
+    'Produce a narrative audit report. Use this EXACT per-finding format — both the structured reviewer and the deterministic fallback extract from this same format:',
     '',
     '## Finding 1: <one-line title>',
     '- Severity: critical | high | medium | low',
@@ -88,7 +89,22 @@ const FINDING_FORMAT_INSTRUCTIONS = [
     'Rules:',
     '- Each finding heading MUST start with "## Finding N: " (h2, "Finding ", number, colon, title) — number sequentially from 1.',
     '- Severity / Location / Issue / Suggestion bullets are on their own lines with the labels exactly as shown.',
-    '- Do NOT emit JSON. The structured reviewer extracts from this format; if that pass fails the deterministic fallback extracts from this same format — both produce identical structured output, so the format is the single source of truth.',
+    '- If you found no issues, say "No findings." in plain prose and emit zero `## Finding N:` blocks.',
+    '',
+    // Tool sweep #12: share the annotator's rubric with the implementer
+    // so the worker self-aligns with what the reviewer will check.
+    // Result: fewer downgraded findings, fewer missed criticals.
+    SEVERITY_LADDER,
+    '',
+    EVIDENCE_GROUNDING,
+    '',
+    SCOPE_DISCIPLINE,
+    '',
+    ANNOTATOR_CHECK_AWARENESS_RO,
+].join('\n');
+const DELTA_AUDIT_INSTRUCTIONS = [
+    'A prior audit report is provided in the context above. Verify which prior findings have been fixed.',
+    'In your output: **omit** fixed findings; **include** still-present findings (mark "unfixed from prior audit"); **include** new findings; end with a **Fixed** summary line.',
 ].join('\n');
 function buildFilePathsPrompt(filePaths) {
     if (filePaths.length === 0)
@@ -107,12 +123,13 @@ function buildPrompt(brief) {
         if (fileSection)
             parts.push(fileSection);
     }
+    // Tool sweep #11: emit FINDING_FORMAT_INSTRUCTIONS unconditionally
+    // (pre-fix the DELTA branch dropped them, leaving the worker without
+    // a format spec → annotator could not parse delta-mode output).
     if (brief.hasContextBlocks) {
-        parts.push('A prior audit report is provided as context above.', 'First, verify which prior findings have been fixed. Then perform a full audit as normal — do not skip areas or reduce thoroughness.', 'In your output:', '- **Omit** prior findings that have been fixed — do not re-report them.', '- **Include** prior findings that are still present (mark as "unfixed from prior audit").', '- **Include** any new findings not in the prior report.', '- End with a **Fixed** summary listing which prior findings were resolved.');
-    }
-    else {
-        parts.push(FINDING_FORMAT_INSTRUCTIONS);
+        parts.push(DELTA_AUDIT_INSTRUCTIONS);
     }
+    parts.push(FINDING_FORMAT_INSTRUCTIONS);
     return parts.join('\n\n');
 }
 export const toolConfig = {

package/dist/tools/audit/tool-config.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-config.js","sourceRoot":"","sources":["../../../src/tools/audit/tool-config.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAc,MAAM,aAAa,CAAC;AACtD,OAAO,EAAE,oBAAoB,EAAE,MAAM,gDAAgD,CAAC;AACtF,OAAO,EAAE,iBAAiB,EAAoB,MAAM,qDAAqD,CAAC;AAC1G,OAAO,EAAE,qBAAqB,EAAE,MAAM,6CAA6C,CAAC;AAGpF,OAAO,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AAEjE,MAAM,UAAU,aAAa,CAAC,QAA6B;IACzD,QAAQ,CAAC,QAAQ,CAAC;QAChB,SAAS,EAAE,OAAO;QAClB,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;QAClB,OAAO,EAAE,MAAM;QACf,MAAM,EAAE,WAAW;QACnB,YAAY,EAAE,WAAW;QACzB,gBAAgB,EAAE,SAAS;QAC3B,oBAAoB,EAAE,KAAK;QAC3B,iBAAiB,EAAE,eAAe;KACnC,CAAC,CAAC;AACL,CAAC;AAcD,MAAM,qBAAqB,GAA2B;IACpD,QAAQ,EAAE,+KAA+K;IACzL,WAAW,EAAE,yKAAyK;IACtL,WAAW,EAAE,wKAAwK;IACrL,KAAK,EAAE,kIAAkI;IACzI,OAAO,EAAE,uIAAuI;CACjJ,CAAC;AAEF,MAAM,kBAAkB,GAAG,+PAA+P,CAAC;AAE3R,SAAS,oBAAoB,CAAC,SAA6B;IACzD,IAAI,SAAS,KAAK,SAAS;QAAE,OAAO,+CAA+C,CAAC;IACpF,IAAI,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC;QAAE,OAAO,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1D,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,oBAAoB,CAAC,SAA6B,EAAE,gBAAyB;IACpF,IAAI,IAAY,CAAC;IACjB,IAAI,SAAS,KAAK,SAAS,EAAE,CAAC;QAC5B,IAAI,GAAG,qBAAqB,CAAC,OAAO,CAAC;IACvC,CAAC;SAAM,IAAI,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;QACpC,IAAI,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,qBAAqB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAChE,CAAC;SAAM,CAAC;QACN,IAAI,GAAG,qBAAqB,CAAC,SAAS,CAAC,IAAI,qBAAqB,CAAC,OAAO,CAAC;IAC3E,CAAC;IACD,OAAO,gBAAgB,CAAC,CAAC,CAAC,IAAI,GAAG,kBAAkB,CAAC,CAAC,CAAC,IAAI,CAAC;AAC7D,CAAC;AAED,SAAS,UAAU,CAAC,KAAyB;IAC3C,OAAO,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;AACxD,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,KAAY;IACzC,MAAM,gBAAgB,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,eAAe,CAAC,IAAI,KAAK,CAAC,eAAe,CAAC,MAAM,GAAG,CAAC,CAAC;IAClG,MAAM,aAAa,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAC5D,MAAM,IAAI,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,EAAE,gBAAgB,CAAC,CAAC;IACrE,MAAM,UAAU,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAE5E,0DAA0D;IAC1D,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,UAAU,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QAC1D,OAAO,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;YAC3B,aAAa;YACb,IAAI;YACJ,SAAS,EAAE,CAAC,EAAE,CAAC;YACf,gBAAgB;YAChB,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,WAAW,EAAE,EAAE;SAChB,CAAC,CAAC,CAAC;IACN,CAAC;IAED,OAAO,CAAC;YACN,aAAa;YACb,IAAI;YACJ,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,SAAS,EAAE,UAAU;YACrB,gBAAgB;YAChB,eAAe,EAAE,KAAK,CAAC,eAAe;SACvC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,2BAA2B,GAAG;IAClC,gKAAgK;IAChK,EAAE;IACF,gCAAgC;IAChC,4CAA4C;IAC5C,yCAAyC;IACzC,oCAAoC;IACpC,2CAA2C;IAC3C,EAAE;IACF,gCAAgC;IAChC,iBAAiB;IACjB,OAAO;IACP,EAAE;IACF,QAAQ;IACR,8HAA8H;IAC9H,6GAA6G;IAC7G,8OAA8O;CAC/O,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,SAAS,oBAAoB,CAAC,SAAmB;IAC/C,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACtC,OAAO,kCAAkC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;AACrF,CAAC;AAED,SAAS,WAAW,CAAC,KAAqB;IACxC,MAAM,KAAK,GAAa,CAAC,aAAa,KAAK,CAAC,aAAa,UAAU,CAAC,CAAC;IAErE,IAAI,KAAK,CAAC,WAAW,EAAE,CAAC;QACtB,KAAK,CAAC,IAAI,CAAC,kCAAkC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;IACpE,CAAC;SAAM,CAAC;QACN,IAAI,KAAK,CAAC,QAAQ;YAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;QACjE,MAAM,WAAW,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAC1D,IAAI,WAAW;YAAE,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IAC3C,CAAC;IAED,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CACR,oDAAoD,EACpD,qIAAqI,EACrI,iBAAiB,EACjB,yEAAyE,EACzE,2FAA2F,EAC3F,yDAAyD,EACzD,4EAA4E,CAC7E,CAAC;IACJ,CAAC;SAAM,CAAC;QACN,KAAK,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAC1C,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC5B,CAAC;AAED,MAAM,CAAC,MAAM,UAAU,GAAmD;IACxE,IAAI,EAAE,OAAO;IACb,QAAQ,EAAE,WAAW;IACrB,SAAS,EAAE,SAAS;IACpB,SAAS,EAAE,cAAc;IACzB,aAAa,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC,CAAC;QAC9B,MAAM,EAAE,WAAW,CAAC,KAAK,CAAC;QAC1B,SAAS,EAAE,SAAS;QACpB,YAAY,EAAE,cAAc;QAC5B,kBAAkB,EAAE,KAAK;QACzB,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,KAAK,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,KAAK,IAAI,MAAM;QAC3C,SAAS,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,SAAS,IAAI,uBAAuB;QACpE,UAAU,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,UAAU,IAAI,EAAE;QACjD,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,aAAa,IAAI,UAAU;QAC/D,GAAG,EAAE,GAAG,CAAC,cAAc,EAAE,GAAG,IAAI,GAAG,CAAC,GAAG;QACvC,eAAe,EAAE,KAAK,CAAC,eAAe;QACtC,SAAS,EAAE,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;QACnE,SAAS,EAAE,GAAG,CAAC,SAAS;KACZ,CAAA;IACd,YAAY,EAAE,iBAAiB;IAC/B,gBAAgB,EAAE,qBAAqB;IACvC,eAAe,EAAE;QACf,SAAS,EAAE,oBAAoB;KAChC;CACF,CAAC"}
+{"version":3,"file":"tool-config.js","sourceRoot":"","sources":["../../../src/tools/audit/tool-config.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAc,MAAM,aAAa,CAAC;AACtD,OAAO,EAAE,oBAAoB,EAAE,MAAM,gDAAgD,CAAC;AACtF,OAAO,EAAE,iBAAiB,EAAoB,MAAM,qDAAqD,CAAC;AAC1G,OAAO,EAAE,qBAAqB,EAAE,MAAM,6CAA6C,CAAC;AAGpF,OAAO,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AACjE,OAAO,EACL,eAAe,EACf,kBAAkB,EAClB,gBAAgB,EAChB,4BAA4B,GAC7B,MAAM,4CAA4C,CAAC;AAEpD,MAAM,UAAU,aAAa,CAAC,QAA6B;IACzD,QAAQ,CAAC,QAAQ,CAAC;QAChB,SAAS,EAAE,OAAO;QAClB,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;QAClB,OAAO,EAAE,MAAM;QACf,MAAM,EAAE,WAAW;QACnB,YAAY,EAAE,WAAW;QACzB,gBAAgB,EAAE,SAAS;QAC3B,oBAAoB,EAAE,KAAK;QAC3B,iBAAiB,EAAE,eAAe;KACnC,CAAC,CAAC;AACL,CAAC;AAcD,MAAM,qBAAqB,GAA2B;IACpD,QAAQ,EAAE,+KAA+K;IACzL,WAAW,EAAE,yKAAyK;IACtL,WAAW,EAAE,wKAAwK;IACrL,KAAK,EAAE,kIAAkI;IACzI,OAAO,EAAE,uIAAuI;CACjJ,CAAC;AAEF,MAAM,kBAAkB,GAAG,+PAA+P,CAAC;AAE3R,SAAS,oBAAoB,CAAC,SAA6B;IACzD,IAAI,SAAS,KAAK,SAAS;QAAE,OAAO,+CAA+C,CAAC;IACpF,IAAI,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC;QAAE,OAAO,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC1D,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,oBAAoB,CAAC,SAA6B,EAAE,gBAAyB;IACpF,IAAI,IAAY,CAAC;IACjB,IAAI,SAAS,KAAK,SAAS,EAAE,CAAC;QAC5B,IAAI,GAAG,qBAAqB,CAAC,OAAO,CAAC;IACvC,CAAC;SAAM,IAAI,KAAK,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC;QACpC,IAAI,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,qBAAqB,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAChE,CAAC;SAAM,CAAC;QACN,IAAI,GAAG,qBAAqB,CAAC,SAAS,CAAC,IAAI,qBAAqB,CAAC,OAAO,CAAC;IAC3E,CAAC;IACD,OAAO,gBAAgB,CAAC,CAAC,CAAC,IAAI,GAAG,kBAAkB,CAAC,CAAC,CAAC,IAAI,CAAC;AAC7D,CAAC;AAED,SAAS,UAAU,CAAC,KAAyB;IAC3C,OAAO,KAAK,KAAK,SAAS,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;AACxD,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,KAAY;IACzC,MAAM,gBAAgB,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,eAAe,CAAC,IAAI,KAAK,CAAC,eAAe,CAAC,MAAM,GAAG,CAAC,CAAC;IAClG,MAAM,aAAa,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;IAC5D,MAAM,IAAI,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,EAAE,gBAAgB,CAAC,CAAC;IACrE,MAAM,UAAU,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAE5E,0DAA0D;IAC1D,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,UAAU,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QAC1D,OAAO,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;YAC3B,aAAa;YACb,IAAI;YACJ,SAAS,EAAE,CAAC,EAAE,CAAC;YACf,gBAAgB;YAChB,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,WAAW,EAAE,EAAE;SAChB,CAAC,CAAC,CAAC;IACN,CAAC;IAED,OAAO,CAAC;YACN,aAAa;YACb,IAAI;YACJ,QAAQ,EAAE,KAAK,CAAC,QAAQ;YACxB,SAAS,EAAE,UAAU;YACrB,gBAAgB;YAChB,eAAe,EAAE,KAAK,CAAC,eAAe;SACvC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,2BAA2B,GAAG;IAClC,kKAAkK;IAClK,EAAE;IACF,gCAAgC;IAChC,4CAA4C;IAC5C,yCAAyC;IACzC,oCAAoC;IACpC,2CAA2C;IAC3C,EAAE;IACF,gCAAgC;IAChC,iBAAiB;IACjB,OAAO;IACP,EAAE;IACF,QAAQ;IACR,8HAA8H;IAC9H,6GAA6G;IAC7G,mGAAmG;IACnG,EAAE;IACF,oEAAoE;IACpE,+DAA+D;IAC/D,6DAA6D;IAC7D,eAAe;IACf,EAAE;IACF,kBAAkB;IAClB,EAAE;IACF,gBAAgB;IAChB,EAAE;IACF,4BAA4B;CAC7B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,wBAAwB,GAAG;IAC/B,qGAAqG;IACrG,6KAA6K;CAC9K,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,SAAS,oBAAoB,CAAC,SAAmB;IAC/C,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACtC,OAAO,kCAAkC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;AACrF,CAAC;AAED,SAAS,WAAW,CAAC,KAAqB;IACxC,MAAM,KAAK,GAAa,CAAC,aAAa,KAAK,CAAC,aAAa,UAAU,CAAC,CAAC;IAErE,IAAI,KAAK,CAAC,WAAW,EAAE,CAAC;QACtB,KAAK,CAAC,IAAI,CAAC,kCAAkC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAC;IACpE,CAAC;SAAM,CAAC;QACN,IAAI,KAAK,CAAC,QAAQ;YAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;QACjE,MAAM,WAAW,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAC1D,IAAI,WAAW;YAAE,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IAC3C,CAAC;IAED,mEAAmE;IACnE,qEAAqE;IACrE,gEAAgE;IAChE,IAAI,KAAK,CAAC,gBAAgB,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,wBAAwB,CAAC,CAAC;IACvC,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAExC,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AAC5B,CAAC;AAED,MAAM,CAAC,MAAM,UAAU,GAAmD;IACxE,IAAI,EAAE,OAAO;IACb,QAAQ,EAAE,WAAW;IACrB,SAAS,EAAE,SAAS;IACpB,SAAS,EAAE,cAAc;IACzB,aAAa,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC,CAAC;QAC9B,MAAM,EAAE,WAAW,CAAC,KAAK,CAAC;QAC1B,SAAS,EAAE,SAAS;QACpB,YAAY,EAAE,cAAc;QAC5B,kBAAkB,EAAE,KAAK;QACzB,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,KAAK,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,KAAK,IAAI,MAAM;QAC3C,SAAS,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,SAAS,IAAI,uBAAuB;QACpE,UAAU,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,UAAU,IAAI,EAAE;QACjD,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,aAAa,IAAI,UAAU;QAC/D,GAAG,EAAE,GAAG,CAAC,cAAc,EAAE,GAAG,IAAI,GAAG,CAAC,GAAG;QACvC,eAAe,EAAE,KAAK,CAAC,eAAe;QACtC,SAAS,EAAE,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;QACnE,SAAS,EAAE,GAAG,CAAC,SAAS;KACZ,CAAA;IACd,YAAY,EAAE,iBAAiB;IAC/B,gBAAgB,EAAE,qBAAqB;IACvC,eAAe,EAAE;QACf,SAAS,EAAE,oBAAoB;KAChC;CACF,CAAC"}

package/dist/tools/debug/tool-config.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-config.d.ts","sourceRoot":"","sources":["../../../src/tools/debug/tool-config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,6CAA6C,CAAC;AAElF,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAEzC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AAEvE,OAAO,EAAkB,KAAK,cAAc,EAAE,MAAM,4CAA4C,CAAC;AAIjG,wBAAgB,aAAa,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI,CAYjE;AA0BD,eAAO,MAAM,UAAU,EAAE,UAAU,CAAC,KAAK,EAAE,cAAc,EAAE,OAAO,CAkCjE,CAAC"}
+{"version":3,"file":"tool-config.d.ts","sourceRoot":"","sources":["../../../src/tools/debug/tool-config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,mBAAmB,EAAE,MAAM,6CAA6C,CAAC;AAElF,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AAEzC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AAEvE,OAAO,EAAkB,KAAK,cAAc,EAAE,MAAM,4CAA4C,CAAC;AAUjG,wBAAgB,aAAa,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI,CAYjE;AAoCD,eAAO,MAAM,UAAU,EAAE,UAAU,CAAC,KAAK,EAAE,cAAc,EAAE,OAAO,CAoCjE,CAAC"}

package/dist/tools/debug/tool-config.js

@@ -3,6 +3,7 @@ import { qualityDebugTemplate } from '../../review/reviewer-engine.js';
 import { debugBriefSlot } from '../../intake/brief-compiler-slots/debug.js';
 import { debugHeadlineTemplate } from '../../reporting/headline-templates/debug.js';
 import { DEFAULT_TASK_TIMEOUT_MS } from '../../config/schema.js';
+import { SEVERITY_LADDER, EVIDENCE_GROUNDING, SCOPE_DISCIPLINE, ANNOTATOR_CHECK_AWARENESS_RO, } from '../../review/templates/finding-criteria.js';
 export function registerDebug(registry) {
     registry.register({
         routeName: 'debug',
@@ -17,13 +18,13 @@ export function registerDebug(registry) {
     });
 }
 const FINDING_FORMAT_INSTRUCTIONS = [
-    'Use hypothesis-driven debugging. Use this EXACT per-finding format so the deterministic extractor can recover findings if the structured reviewer pass fails:',
+    'Use hypothesis-driven debugging. Use this EXACT per-finding format — both the structured reviewer and the deterministic fallback extract from this same format:',
     '',
     '## Finding 1: <one-line title>',
     '- Severity: critical | high | medium | low',
     '- Hypothesis: the candidate cause',
     '- Evidence: trace, log, or code path with file:line',
-    '- Fix: proposed change',
+    '- Fix: proposed change (PROPOSE only — do NOT apply the fix)',
     '',
     '## Finding 2: <one-line title>',
     '- Severity: ...',
@@ -32,7 +33,17 @@ const FINDING_FORMAT_INSTRUCTIONS = [
     'Rules:',
     '- Each finding heading MUST start with "## Finding N: " (h2, "Finding ", number, colon, title) — number sequentially from 1.',
     '- Severity / Hypothesis / Evidence / Fix bullets are on their own lines with the labels exactly as shown.',
-    '- Do NOT emit JSON. Both the structured reviewer and the deterministic fallback extract from this same format — the format is the single source of truth.',
+    '- This is a read-only diagnostic — do NOT edit any file. Propose fixes; the caller applies them.',
+    '- Limit yourself to 3-5 most-likely hypotheses. Do not enumerate implausible ones to pad the list.',
+    '',
+    // Tool sweep #12: shared rubric so worker self-aligns with the annotator.
+    SEVERITY_LADDER,
+    '',
+    EVIDENCE_GROUNDING,
+    '',
+    SCOPE_DISCIPLINE,
+    '',
+    ANNOTATOR_CHECK_AWARENESS_RO,
 ].join('\n');
 function buildFilePathsPrompt(filePaths) {
     if (!filePaths || filePaths.length === 0)
@@ -60,7 +71,7 @@ export const toolConfig = {
             agentType: 'complex',
             reviewPolicy: 'quality_only',
             briefQualityPolicy: 'off',
-            done: 'Identify the root cause with evidence (file, line, mechanism). Propose a fix. Verify the fix resolves the problem.',
+            done: 'Identify the root cause with evidence (file, line, mechanism) and PROPOSE a fix. Do NOT apply the fix — debug is a read-only diagnostic; the caller decides whether to apply.',
             tools: ctx.config.defaults?.tools ?? 'full',
             timeoutMs: ctx.config.defaults?.timeoutMs ?? DEFAULT_TASK_TIMEOUT_MS,
             maxCostUSD: ctx.config.defaults?.maxCostUSD ?? 10,
@@ -68,6 +79,8 @@ export const toolConfig = {
             cwd: ctx.projectContext?.cwd ?? ctx.cwd,
             contextBlockIds: brief.contextBlockIds,
             autoCommit: false,
+            filePaths: brief.filePaths && brief.filePaths.length > 0 ? brief.filePaths : undefined,
+            mainModel: ctx.mainModel ?? undefined,
         };
     },
     reportSchema: { parse: (_text) => { throw new Error('no structured report emitted by this executor'); } },

package/dist/tools/debug/tool-config.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-config.js","sourceRoot":"","sources":["../../../src/tools/debug/tool-config.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAE1C,OAAO,EAAE,oBAAoB,EAAE,MAAM,iCAAiC,CAAC;AAGvE,OAAO,EAAE,cAAc,EAAuB,MAAM,4CAA4C,CAAC;AACjG,OAAO,EAAE,qBAAqB,EAAE,MAAM,6CAA6C,CAAC;AACpF,OAAO,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AAEjE,MAAM,UAAU,aAAa,CAAC,QAA6B;IACzD,QAAQ,CAAC,QAAQ,CAAC;QAChB,SAAS,EAAE,OAAO;QAClB,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;QAClB,OAAO,EAAE,MAAM;QACf,MAAM,EAAE,WAAW;QACnB,YAAY,EAAE,WAAW;QACzB,gBAAgB,EAAE,SAAS;QAC3B,oBAAoB,EAAE,KAAK;QAC3B,iBAAiB,EAAE,eAAe;KACnC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,2BAA2B,GAAG;IAClC,+JAA+J;IAC/J,EAAE;IACF,gCAAgC;IAChC,4CAA4C;IAC5C,mCAAmC;IACnC,qDAAqD;IACrD,wBAAwB;IACxB,EAAE;IACF,gCAAgC;IAChC,iBAAiB;IACjB,OAAO;IACP,EAAE;IACF,QAAQ;IACR,8HAA8H;IAC9H,2GAA2G;IAC3G,2JAA2J;CAC5J,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,SAAS,oBAAoB,CAAC,SAAoB;IAChD,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACpD,OAAO,kCAAkC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;AACrF,CAAC;AAED,MAAM,CAAC,MAAM,UAAU,GAA+C;IACpE,IAAI,EAAE,OAAO;IACb,QAAQ,EAAE,WAAW;IACrB,SAAS,EAAE,SAAS;IACpB,SAAS,EAAE,cAAc;IACzB,aAAa,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE;QAC5B,MAAM,KAAK,GAAa,CAAC,0BAA0B,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QACpE,IAAI,KAAK,CAAC,OAAO;YAAE,KAAK,CAAC,IAAI,CAAC,YAAY,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3D,IAAI,KAAK,CAAC,UAAU;YAAE,KAAK,CAAC,IAAI,CAAC,uBAAuB,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC;QAC5E,MAAM,WAAW,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAC1D,IAAI,WAAW;YAAE,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QACzC,KAAK,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;QACxC,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAElC,OAAO;YACL,MAAM;YACN,SAAS,EAAE,SAAS;YACpB,YAAY,EAAE,cAAc;YAC5B,kBAAkB,EAAE,KAAK;YACzB,IAAI,EAAE,oHAAoH;YAC1H,KAAK,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,KAAK,IAAI,MAAM;YAC3C,SAAS,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,SAAS,IAAI,uBAAuB;YACpE,UAAU,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,UAAU,IAAI,EAAE;YACjD,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,aAAa,IAAI,UAAU;YAC/D,GAAG,EAAE,GAAG,CAAC,cAAc,EAAE,GAAG,IAAI,GAAG,CAAC,GAAG;YACvC,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,UAAU,EAAE,KAAK;SAClB,CAAC;IACJ,CAAC;IACD,YAAY,EAAE,EAAE,KAAK,EAAE,CAAC,KAAK,EAAE,EAAE,GAAG,MAAM,IAAI,KAAK,CAAC,+CAA+C,CAAC,CAAC,CAAC,CAAC,EAAE;IACzG,gBAAgB,EAAE,qBAAqB;IACvC,eAAe,EAAE;QACf,SAAS,EAAE,oBAAoB;KAChC;CACF,CAAC"}
+{"version":3,"file":"tool-config.js","sourceRoot":"","sources":["../../../src/tools/debug/tool-config.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAE1C,OAAO,EAAE,oBAAoB,EAAE,MAAM,iCAAiC,CAAC;AAGvE,OAAO,EAAE,cAAc,EAAuB,MAAM,4CAA4C,CAAC;AACjG,OAAO,EAAE,qBAAqB,EAAE,MAAM,6CAA6C,CAAC;AACpF,OAAO,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AACjE,OAAO,EACL,eAAe,EACf,kBAAkB,EAClB,gBAAgB,EAChB,4BAA4B,GAC7B,MAAM,4CAA4C,CAAC;AAEpD,MAAM,UAAU,aAAa,CAAC,QAA6B;IACzD,QAAQ,CAAC,QAAQ,CAAC;QAChB,SAAS,EAAE,OAAO;QAClB,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;QAClB,OAAO,EAAE,MAAM;QACf,MAAM,EAAE,WAAW;QACnB,YAAY,EAAE,WAAW;QACzB,gBAAgB,EAAE,SAAS;QAC3B,oBAAoB,EAAE,KAAK;QAC3B,iBAAiB,EAAE,eAAe;KACnC,CAAC,CAAC;AACL,CAAC;AAED,MAAM,2BAA2B,GAAG;IAClC,iKAAiK;IACjK,EAAE;IACF,gCAAgC;IAChC,4CAA4C;IAC5C,mCAAmC;IACnC,qDAAqD;IACrD,8DAA8D;IAC9D,EAAE;IACF,gCAAgC;IAChC,iBAAiB;IACjB,OAAO;IACP,EAAE;IACF,QAAQ;IACR,8HAA8H;IAC9H,2GAA2G;IAC3G,kGAAkG;IAClG,oGAAoG;IACpG,EAAE;IACF,0EAA0E;IAC1E,eAAe;IACf,EAAE;IACF,kBAAkB;IAClB,EAAE;IACF,gBAAgB;IAChB,EAAE;IACF,4BAA4B;CAC7B,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,SAAS,oBAAoB,CAAC,SAAoB;IAChD,IAAI,CAAC,SAAS,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACpD,OAAO,kCAAkC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;AACrF,CAAC;AAED,MAAM,CAAC,MAAM,UAAU,GAA+C;IACpE,IAAI,EAAE,OAAO;IACb,QAAQ,EAAE,WAAW;IACrB,SAAS,EAAE,SAAS;IACpB,SAAS,EAAE,cAAc;IACzB,aAAa,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE;QAC5B,MAAM,KAAK,GAAa,CAAC,0BAA0B,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QACpE,IAAI,KAAK,CAAC,OAAO;YAAE,KAAK,CAAC,IAAI,CAAC,YAAY,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3D,IAAI,KAAK,CAAC,UAAU;YAAE,KAAK,CAAC,IAAI,CAAC,uBAAuB,KAAK,CAAC,UAAU,EAAE,CAAC,CAAC;QAC5E,MAAM,WAAW,GAAG,oBAAoB,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;QAC1D,IAAI,WAAW;YAAE,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QACzC,KAAK,CAAC,IAAI,CAAC,2BAA2B,CAAC,CAAC;QACxC,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAElC,OAAO;YACL,MAAM;YACN,SAAS,EAAE,SAAS;YACpB,YAAY,EAAE,cAAc;YAC5B,kBAAkB,EAAE,KAAK;YACzB,IAAI,EAAE,+KAA+K;YACrL,KAAK,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,KAAK,IAAI,MAAM;YAC3C,SAAS,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,SAAS,IAAI,uBAAuB;YACpE,UAAU,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,UAAU,IAAI,EAAE;YACjD,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,aAAa,IAAI,UAAU;YAC/D,GAAG,EAAE,GAAG,CAAC,cAAc,EAAE,GAAG,IAAI,GAAG,CAAC,GAAG;YACvC,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,UAAU,EAAE,KAAK;YACjB,SAAS,EAAE,KAAK,CAAC,SAAS,IAAI,KAAK,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS;YACtF,SAAS,EAAE,GAAG,CAAC,SAAS,IAAI,SAAS;SACtC,CAAC;IACJ,CAAC;IACD,YAAY,EAAE,EAAE,KAAK,EAAE,CAAC,KAAK,EAAE,EAAE,GAAG,MAAM,IAAI,KAAK,CAAC,+CAA+C,CAAC,CAAC,CAAC,CAAC,EAAE;IACzG,gBAAgB,EAAE,qBAAqB;IACvC,eAAe,EAAE;QACf,SAAS,EAAE,oBAAoB;KAChC;CACF,CAAC"}

package/dist/tools/execute-plan/tool-config.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-config.d.ts","sourceRoot":"","sources":["../../../src/tools/execute-plan/tool-config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,mBAAmB,EAAE,MAAM,6CAA6C,CAAC;AAMlF,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AACvE,OAAO,EAA4B,KAAK,oBAAoB,EAAE,MAAM,mDAAmD,CAAC;AAKxH,eAAO,MAAM,sBAAsB;;;;;;;;;;;;kBAOxB,CAAC;AAEZ,MAAM,MAAM,oBAAoB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAE1E,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI,CAYvE;AAsCD,eAAO,MAAM,UAAU,EAAE,UAAU,CAAC,oBAAoB,EAAE,oBAAoB,CA4B7E,CAAC"}
+{"version":3,"file":"tool-config.d.ts","sourceRoot":"","sources":["../../../src/tools/execute-plan/tool-config.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,mBAAmB,EAAE,MAAM,6CAA6C,CAAC;AAMlF,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,sCAAsC,CAAC;AACvE,OAAO,EAA4B,KAAK,oBAAoB,EAAE,MAAM,mDAAmD,CAAC;AAMxH,eAAO,MAAM,sBAAsB;;;;;;;;;;;;kBAOxB,CAAC;AAEZ,MAAM,MAAM,oBAAoB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,sBAAsB,CAAC,CAAC;AAE1E,wBAAgB,mBAAmB,CAAC,QAAQ,EAAE,mBAAmB,GAAG,IAAI,CAYvE;AA0CD,eAAO,MAAM,UAAU,EAAE,UAAU,CAAC,oBAAoB,EAAE,oBAAoB,CA4B7E,CAAC"}