Menu
Aggregate and deduplicate findings from multiple audit reports
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Run a single-session AI optimization audit on the codebase
Run a single-session code review audit on the codebase
Run a single-session engineering productivity audit on the codebase
Run a comprehensive enhancement audit across the entire project - code, product, UX, content, workflows, infrastructure, external services, and meta-tooling.
Run a single-session performance audit on the codebase
Run a comprehensive multi-stage automation audit with parallel agents
| name | audit-aggregator |
| description | Aggregate and deduplicate findings from multiple audit reports |
| supports_parallel | false |
| fallback_available | false |
| estimated_time_parallel | 10 min |
| estimated_time_sequential | 10 min |
Purpose: Merge findings from 9 domain-specific audit reports into a single comprehensive, priority-ranked report with cross-domain insights and deduplication.
Input: Multiple audit report files (markdown format) Output:
COMPREHENSIVE_AUDIT_REPORT.md + comprehensive-findings.jsonl
/audit-aggregatoraudit-comprehensive after all 9 domain auditsdocs/audits/comprehensive/audit-YYYY-MM-DD/docs/technical-debt/FALSE_POSITIVES.jsonlIf fewer than 5 reports: Warn user. Proceed with Markdown only, mark as PARTIAL -- TDMS SKIPPED. Do NOT generate JSONL or run Post-Audit steps.
Expected Files (9 reports in docs/audits/comprehensive/audit-YYYY-MM-DD/):
audit-code-report.md, audit-security-report.md,
audit-performance-report.md, audit-documentation-report.md,
audit-refactoring-report.md, audit-process-report.md,
audit-engineering-productivity-report.md, audit-enhancements-report.md,
audit-ai-optimization-report.md
Required Finding Format:
### [ID] [Title] (Severity: S0-S3, Effort: E0-E3, Confidence: High/Medium/Low)
with File, Category, Description, Evidence, Recommendation fields.
Extract findings into structured format. See examples.md for
schema. ID format: {DOMAIN}-{NUMBER}. Handle missing fields with defaults: S2,
E1, Medium, "General".
Group by (file, line) pair. For duplicates, merge using:
mapPrimaryCategory — if 2+ domains, use highest-severity
domain's category (priority: security > performance > code-quality > others)See examples.md for deduplication code and merge examples.
Detect: A. Hotspot files (3+ audits), B. Related findings (same file, different concerns), C. Domain overlap counts, D. Category patterns.
priority = severityWeight * crossDomainMultiplier * confidenceWeight / effortWeight
S0 Critical always rank first. Cross-domain findings bubble up. See examples.md for formula weights and example scores.
Statistics, severity breakdown, effort estimates, 3-5 key insights, recommended fix order (4 phases). See examples.md for templates.
Output: docs/audits/comprehensive/COMPREHENSIVE_AUDIT_REPORT.md
Structure: Executive Summary, Priority-Ranked Top 20, Cross-Domain Insights (hotspots, overlaps, categories), Full Findings by severity, Appendix.
Also produce JSONL: comprehensive-findings.jsonl with all deduplicated
findings. See examples.md for schema.
mapPrimaryCategory)| Domain | TDMS Category |
|---|---|
code | code-quality |
security | security |
performance | performance |
refactoring | refactoring |
documentation | documentation |
process | process |
engineering-productivity | engineering-productivity |
enhancements | enhancements |
ai-optimization | ai-optimization |
Before presenting findings for review, cross-reference against
docs/technical-debt/MASTER_DEBT.jsonl by file path, title similarity, and root
cause. Classify as: Already Tracked (skip) | New (review) | Possibly
Related (flag). Write DEDUP_VS_MASTER_DEBT.md. Only pass New and Possibly
Related to Interactive Review.
Present only new/possibly-related findings in batches of 10, sorted by severity then confidence. For each:
[N/TOTAL] FINGERPRINT
Title | Severity | Effort | Confidence | Category | Domains
Files | Description | Suggested Fix
Action? [keep / drop / adjust-severity / adjust-effort / edit]
After each batch show running tally. Post-review: write drops to
FALSE_POSITIVES.jsonl, apply adjustments to JSONL.
node scripts/debt/validate-schema.js <findings.jsonl>node scripts/debt/intake-audit.js <findings.jsonl> --source "audit-comprehensive-YYYY-MM-DD"node scripts/debt/generate-views.jsnode scripts/debt/generate-metrics.jsnode scripts/debt/sync-roadmap-refs.jsAll 5 must pass before aggregation is complete.
After TDMS intake: prioritize items in 3+ domain audits, S0/S1, or blocking
dependencies. Auto-assign tracks by category + file patterns (see
examples.md for track assignment rules). Update ROADMAP.md,
validate with node scripts/debt/sync-roadmap-refs.js --check-only.
Review cadence: Full triage after comprehensive audit, category triage after single audit, weekly check unplaced-items.md, pre-sprint review S0/S1.
Runs sequentially after all 9 domain audit agents complete (Stage 3 of
audit-comprehensive). No internal parallel agents.
audit-comprehensive (parent orchestrator)/audit-code, /audit-security, /audit-performance, /audit-refactoring/audit-documentation, /audit-process, /audit-engineering-productivity/audit-enhancements, /audit-ai-optimization/create-audit| Version | Date | Description |
|---|---|---|
| 1.7 | 2026-02-25 | Ported to framework: removed project-specific doc refs |
| 1.6 | 2026-02-24 | Trim to <500 lines: extract code examples to examples.md |
| 1.5 | 2026-02-22 | Add mandatory MASTER_DEBT cross-reference step |
| 1.4 | 2026-02-17 | Add mandatory Interactive Review phase before TDMS intake |
| 1.3 | 2026-02-16 | AUDIT_STANDARDS compliance: Pre-Audit, JSONL, Post-Audit |
| 1.2 | 2026-02-14 | 9-domain coverage |
| 1.1 | 2026-02-03 | Added Triage & Roadmap Integration section |
| 1.0 | 2026-01-28 | Initial skill creation |