ln-312-review-findings-worker（ln 312 审查结果工人）

Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root.

Type: L3 Worker Category: 3XX Planning

Review Findings Worker

Mandatory Read

MANDATORY READ: Load shared/references/evaluation_worker_runtime_contract.md, shared/references/evaluation_summary_contract.md MANDATORY READ: Load ../ln-310-multi-agent-validator/references/phase2_research_audit.md, ../ln-310-multi-agent-validator/references/penalty_points.md MANDATORY READ: Load ../ln-310-multi-agent-validator/references/premortem_validation.md, ../ln-310-multi-agent-validator/references/cross_reference_validation.md

Purpose

• analyze the target artifact or diff
• convert evidence into normalized findings
• for mode=story: calculate penalty points across 30 criteria per phase2_research_audit.md
• for mode=plan_review|context: evaluate criteria #5, #6, #21, #28 only (no penalty accumulation)
• avoid narrative-only review output

Mode Gate

• mode=story: full pipeline — pre-mortem, cross-reference, penalty points across 30 criteria, build fix plan
• mode=plan_review|context: applicability check, stack detection, evaluate criteria #5 (standards), #6 (library versions), #21 (alternatives), #28 (library features) only, normalize findings without penalty accumulation

Runtime

Runtime family:

• evaluation-worker-runtime

Required manifest fields:

• identifier
• phase_order
• summary_kind=review-findings
• operation=findings

Recommended phase_order:

1. PHASE_0_CONFIG
2. PHASE_1_LOAD_TARGET
3. PHASE_2_PREMORTEM (mode=story, complexity >= Medium)
4. PHASE_3_CROSS_REFERENCE (mode=story, multi-story Epic)
5. PHASE_4_CRITERIA_AUDIT
6. PHASE_5_PENALTY_CALCULATION (mode=story only)
7. PHASE_6_NORMALIZE_FINDINGS
8. PHASE_7_WRITE_SUMMARY
9. PHASE_8_SELF_CHECK

Workflow

Phase 0: Config

Load runtime manifest, target identifiers, and any linked research artifact paths.

Phase 1: Load Target

Load only the target artifacts needed for the review scope.

Phase 2: Pre-mortem (mode=story)

Execute pre-mortem analysis per premortem_validation.md:

1. Skip for trivial Stories (1-2 tasks, no external deps, known tech).
2. Execute for Stories with complexity >= Medium (3+ tasks, external deps, or unfamiliar tech).
3. Tigers (evidence-based risks) feed Risk criterion #20 — add to risk table BEFORE penalty calc.
4. Elephants (unstated assumptions) feed Assumptions criterion #24 — add with [pre-mortem] tag, Confidence=LOW.
5. Paper Tigers (fears without evidence) — document and dismiss.
6. Include pre-mortem table in audit report.

Phase 3: Cross-Reference (mode=story)

Execute cross-reference analysis per cross_reference_validation.md:

1. Skip if Epic has only 1 Story or all siblings Done/Canceled.
2. Load sibling Stories via list_issues(project=Epic.id).
3. Check AC overlap (#25): structured traceability first, keyword fallback advisory-only.
4. Check task duplication (#26): structured match primary.
5. Include cross-reference findings in audit report.

Phase 4: Criteria Audit

1. mode=story: evaluate all 30 criteria against Story/Tasks per phase2_research_audit.md Auto-Fix Actions Reference.
2. mode=plan_review|context: evaluate criteria #5, #6, #21, #28 only (standards + solution groups).
3. Cross-check claims against provided research evidence when present.

Phase 5: Penalty Calculation (mode=story)

1. Assign penalty points per violation using severity levels from phase2_research_audit.md (CRITICAL=10, HIGH=5, MEDIUM=3, LOW=1).
2. Apply multiple-violation rules per penalty_points.md Calculation Rules.
3. Calculate total penalty points.
4. Build fix plan for each violation.
5. Format penalty audit table per penalty_points.md Report Format.

Phase 6: Normalize Findings

Each finding should prefer structured fields such as:

• id
• severity
• category
• subject
• evidence
• recommendation

Phase 7: Write Summary

Emit summary_kind=review-findings.

Payload must include:

• worker=ln-312
• status
• operation=findings
• warnings

Prefer these fields when available:

• findings
• metrics.penalty_total (mode=story)
• metrics.criteria_violated (list of criterion numbers)
• metrics.fix_plan (array of {criterion, action, severity})
• metrics.premortem_summary (when executed)
• metrics.cross_reference_summary (when executed)

Phase 8: Self-Check

1. Remove duplicates.
2. Remove unsupported claims.
3. Verify penalty calculation matches penalty_points.md rules (mode=story).
4. Record pass=true only after summary write.

Definition of Done

• Target artifact loaded
• Pre-mortem executed or justified as skipped (mode=story)
• Cross-reference executed or justified as skipped (mode=story)
• Criteria audit completed (30 for story, #5/#6/#21/#28 for others)
• Penalty points calculated and fix plan built (mode=story)
• Findings normalized
• Unsupported claims removed
• review-findings summary written
• Self-check passed

Version: 1.0.0 Last Updated: 2026-04-10