| name | audit |
| description | Run a comprehensive codebase audit across architecture, quality, security, and performance dimensions. Dispatches specialized agents. |
| argument-hint | [all|12factor|patterns|rust|security|smells|efficiency] |
Run a codebase audit on the Gorilla Coach project. Scope: $ARGUMENTS (default: all).
Dispatch Table
Map the requested scope to agent(s):
| Argument | Agent(s) | Scope passed to agent |
|---|
all (or empty) | audit-architecture, audit-quality, audit-security-perf | all dimensions |
12factor | audit-architecture | 12-Factor only |
patterns | audit-architecture | GoF patterns only |
rust | audit-quality | Rust best practices only |
smells | audit-quality | Code smells and dead code only |
security | audit-security-perf | Security only |
efficiency | audit-security-perf | Efficiency only |
Execution
-
Preparation: Read CLAUDE.md to pass project context to agents.
-
Dispatch: Launch the appropriate agent(s) based on the argument. CRITICAL: When multiple agents are needed (e.g., all), you MUST launch ALL of them in parallel by including multiple Agent tool calls in a single message. Do NOT run them sequentially. Pass each agent a prompt that includes:
- The specific dimension(s) to audit
- The workspace root path:
/home/mo/data/Documents/git/gorilla_coach
- Instruction to read CLAUDE.md first
-
Aggregation: Wait for all parallel agents to complete, then produce a consolidated report:
- Deduplicate findings that appear in multiple agents' results
- Sort by severity: CRITICAL > HIGH > MEDIUM > LOW
- Group by dimension
Output Format
# Codebase Audit Report
**Scope:** [dimensions audited]
**Date:** [today]
## Summary
| Severity | Count |
|----------|-------|
| CRITICAL | N |
| HIGH | N |
| MEDIUM | N |
| LOW | N |
## Findings
### CRITICAL
[findings sorted by dimension]
### HIGH
[findings sorted by dimension]
### MEDIUM
[findings sorted by dimension]
### LOW
[findings sorted by dimension]
## Positive Observations
[well-implemented patterns worth preserving]
Rules
- Read-only — no file modifications, no fixes
- Every finding must have a concrete
file:line reference
- Respect CLAUDE.md conventions — don't flag things the project does intentionally
- Precision over recall — fewer accurate findings beat many speculative ones
- Acknowledge what's done well, not just problems