/audit - Multi-Agent Audit Command

Trigger parallel sub-agent swarms to audit, research, or check multiple items at once.

Usage

/audit [target]

Targets

Target	What It Does
skills	Audit all skills against official docs
rules	Audit all user-level rules
versions	Check all package versions in skills
[directory]	Audit all items in specified directory
[topic]	Research topic across multiple sources

Behavior

When invoked:

1. Identify scope - Count items to audit
2. Group logically - Batch related items (8-12 per agent optimal)
3. Launch parallel - All agents in ONE message for true parallelism
4. Compile findings - Create planning doc at planning/[TARGET]_AUDIT_[DATE].md

Agent Prompt Template

Each agent receives:

Audit [ITEMS] against official documentation.

Check for:
1. NEW features not documented
2. DEPRECATED patterns still recommended
3. BREAKING CHANGES not reflected
4. VERSION updates needed

Report: Accuracy %, gaps found, priority recommendations (HIGH/MEDIUM/LOW).

Examples

Audit all skills

/audit skills

→ Launches 68 parallel agents, creates planning/SKILL_UPDATES_[DATE].md

Audit rules by category

/audit rules

→ Launches 9 grouped agents (by domain), creates planning/RULES_UPDATES_[DATE].md

Research a topic

/audit "auth libraries for cloudflare workers"

→ Launches agents to research better-auth, clerk, lucia, oslo, etc.

Check versions

/audit versions

→ Parallel version checks, creates VERSIONS_REPORT.md

Output

Creates a tracking document that survives context compacts:

# [Target] Audit - [Date]

## Summary
- Total items: X
- Needing updates: Y
- Estimated effort: Z hours

## Priority Tiers
### TIER 1: URGENT
| Item | Issue | Est Hours | Status |

### TIER 2: HIGH
...

## Session Log
| Date | Work Done | Next |

Reference

See planning/multi-agent-research-protocol.md for full patterns.

Key Discovery

No practical agent limit - 68 agents ran successfully in parallel. Token usage is the constraint, not count.