multi-agent-codebase-auditlisted
Install: claude install-skill juliuswiener/nord-kit
# Multi-Agent Codebase Audit
## Overview
An architect-level audit that fans across **~12 specialist lanes in parallel**, **adversarially verifies high-severity findings** (3 skeptics per finding, majority refute → drop), then **synthesizes a severity-ranked report**. Read-only by default. Cleaning is a separate follow-up after human review of findings — auto-fix in an unfamiliar codebase is unsafe.
The audit surfaces *what will hurt*, not *what could be prettier*. Findings have category, severity, file/line, impact, recommendation, and effort. Linter-grade nits are out of scope — formatters and CI already catch those.
## When to use vs alternatives
| Need | Use |
|---|---|
| Quick single-pass critique, one critique file | `scrutinizing-projects` |
| Review a single PR or diff | `/code-review` or `/review` |
| Verify a specific fix works | `/verify` |
| Find a bug or trace a failure | `superpowers:systematic-debugging` |
| Security-only deep dive | `oh-my-claudecode:security-reviewer` directly |
| **Pre-release / pre-handover / due-diligence / quarterly health** | **this skill** |
## Running it
You MUST invoke the bundled Workflow script with **absolute paths** and **args as a JSON object** (not a string):
```
Workflow({
scriptPath: '/home/<user>/.claude/skills/multi-agent-codebase-audit/audit.workflow.js',
args: { path: '/absolute/path/to/repo', archetype: 'auto' }
})
```
**Common failure modes — avoid these:**
- ❌ `scriptPath: '~/.claude/...'` — tilde may not ex