Audit Aggregator Agent

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

When to Use

• User explicitly invokes /audit-aggregator
• Called automatically by audit-comprehensive after all 9 domain audits

When NOT to Use

• When a more specialized skill exists for the specific task

Pre-Audit Validation

1. Check input files — At least 5 of 9 domain reports in docs/audits/comprehensive/audit-YYYY-MM-DD/
2. Load false positives — Read docs/technical-debt/FALSE_POSITIVES.jsonl
3. Check prior aggregations — Search episodic memory for previous results
4. Verify output directory exists

If fewer than 5 reports: Warn user. Proceed with Markdown only, mark as PARTIAL -- TDMS SKIPPED. Do NOT generate JSONL or run Post-Audit steps.

Input Requirements

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.

Processing Steps

Step 1: Parse All Reports

Extract findings into structured format. See examples.md for schema. ID format: {DOMAIN}-{NUMBER}. Handle missing fields with defaults: S2, E1, Medium, "General".

Step 2: Deduplicate Findings

Group by (file, line) pair. For duplicates, merge using:

• Severity: worst (S0 > S1 > S2 > S3)
• Effort: highest (E3 > E2 > E1 > E0)
• Confidence: highest (High > Medium > Low)
• Title/Description: combine unique values with domain labels
• Category: use 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.

Step 3: Identify Cross-Cutting Patterns

Detect: A. Hotspot files (3+ audits), B. Related findings (same file, different concerns), C. Domain overlap counts, D. Category patterns.

Step 4: Priority Ranking

priority = severityWeight * crossDomainMultiplier * confidenceWeight / effortWeight

S0 Critical always rank first. Cross-domain findings bubble up. See examples.md for formula weights and example scores.

Step 5: Generate Executive Summary

Statistics, severity breakdown, effort estimates, 3-5 key insights, recommended fix order (4 phases). See examples.md for templates.

Step 6: Generate Full Report

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.

Category Mapping (mapPrimaryCategory)

Output Validation

1. Verify dedup: unique vs raw counts, no duplicate file:line pairs
2. Verify ranking: S0 in top 20, cross-domain rank higher
3. Verify cross-cutting: hotspots have 3+ mentions
4. Verify completeness: unique + duplicates = raw total

MASTER_DEBT Cross-Reference (MANDATORY)

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.

Interactive Review (MANDATORY)

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.

Post-Audit (5-Step TDMS Checklist)

1. node scripts/debt/validate-schema.js <findings.jsonl>
2. node scripts/debt/intake-audit.js <findings.jsonl> --source "audit-comprehensive-YYYY-MM-DD"
3. node scripts/debt/generate-views.js
4. node scripts/debt/generate-metrics.js
5. node scripts/debt/sync-roadmap-refs.js

All 5 must pass before aggregation is complete.

Triage & Roadmap Integration

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.

Error Handling

• Missing report: Note in report, continue with available audits
• Parse failure: Log, skip malformed finding, note in report
• Empty dedup result: Verify inputs, fall back to showing all findings

Dependency Constraints

Runs sequentially after all 9 domain audit agents complete (Stage 3 of audit-comprehensive). No internal parallel agents.

Related Skills

• 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

Documentation References

• PROCEDURE.md - Full TDMS workflow
• JSONL_SCHEMA_STANDARD.md - Output format
• DOCUMENTATION_STANDARDS.md - 5-tier hierarchy

Version History