Code Audit

Overview

Audit codebase sections against configurable standards using parallel specialist agents. Each section gets reviewed by the right type of agent against the relevant standards. Produces a structured report with file:line findings.

Announce at start: "I'm using the zenflow:audit skill to audit the codebase."

Modes:

• Full (default) — audit all configured sections
• Changed (/zenflow:audit changed) — only audit files modified in the current branch vs main

Changed Mode

When invoked with changed, only audit files that have been modified:

git diff --name-only main...HEAD

1. Get the list of changed files
2. For each changed file, find which audit section it belongs to (match against dir paths)
3. Only dispatch auditors for sections that have changed files, and only pass them the changed files
4. Files that don't match any configured section are reported as "uncovered" — suggest adding a section

This is fast enough for pre-merge checks and CI. Use full mode for periodic comprehensive audits.

Configuration

Audit config is read from the agents array in .claude/zen.local.md:

---
project:
  language: typescript
  framework: nextjs

agents:
  - domain: api-routes
    dir: packages/server/src/server/routes/
    glob: "*.ts"
    agent: Backend Architect
    rules:
      - error-handling
      - async-patterns
  - domain: frontend
    dir: packages/frontend/src/components/
    glob: "**/*.tsx"
    agent: Frontend Developer
    rules:
      - react-patterns
---

Config fields used by zenflow:audit

Field	Description
domain	Section label in the audit report
dir	Directory to audit
glob	File pattern (default: **/*)
agent	Agent type to dispatch
rules	Rule names — resolved to .claude/rules/{name}.md

Missing config

If no agents array exists in zen.local.md, run zenflow:init first:

"No zen config found. Run /zenflow:init to scan the codebase and generate config automatically."

The Process

Step 1: Load Config

1. Read .claude/zen.local.md
2. Parse the agents array from frontmatter
3. If missing, tell the user to run /zenflow:init and stop
4. For each agent entry, verify the dir exists and rules files exist
5. Report the audit plan: how many domains, how many files per domain, which agents will be dispatched

Step 2: Dispatch Auditors

Launch one subagent per section in parallel via the Agent tool:

Each auditor receives:

• The section name and file list
• The full text of each referenced standards file
• Instructions to check every file against every standard

Auditor instructions:

You are auditing {section.name} against these standards:

{standards text}

Files to audit:
{file list}

For each file, check compliance with every standard. Report:

## Findings

### {filename}:{line}
- **Standard:** {which standard is violated}
- **Severity:** Critical | Important | Minor
- **Issue:** {what's wrong}
- **Fix:** {how to fix it}

### Summary
- Files audited: N
- Findings: N (Critical: N, Important: N, Minor: N)
- Clean files: N

Agent type selection:

1. Config first — use the agent field from the matching agents entry
2. Discovery fallback — if no agent is set, glob .claude/agents/*.md and ~/.claude/agents/*.md, read each file's name and description, match by section domain and file types, fall back to general-purpose
3. Inform the user for any section using auto-selection:

   "[N] sections are using auto-selected agents. Run /zenflow:init to generate agent assignments, or specify agent in your zen.local.md config."

Step 3: Compile Report

After all auditors complete:

1. Merge findings from all sections into a single report
2. Sort by severity (Critical → Important → Minor)
3. Group by section
4. Present summary:

## Audit Report

**Sections audited:** 4
**Files audited:** 47
**Total findings:** 12 (Critical: 1, Important: 5, Minor: 6)

### Critical
1. [packages/server/src/server/routes/api.ts:45] — Missing error boundary...

### Important
...

### Minor
...

### Clean Sections
- Type Schemas (14 files) — no findings

Step 4: Action Plan

After presenting the report, ask the user:

Question: "How should we handle the findings?"

Options:

• Fix Critical only — address blockers immediately
• Fix Critical + Important — comprehensive fix pass
• Create plan — invoke zenflow:plan to create a structured fix plan
• Save report only — write findings to a file for later

If fixing: dispatch subagents to fix findings, then run zenflow:check-work.

Scheduled Audits

For recurring audits, suggest the user set up a cron job:

/loop 24h /zenflow:audit

STUCK Criteria

• If a standards file doesn't exist, skip that standard and warn
• If a section directory doesn't exist, skip and warn
• If an auditor fails, report partial results from completed auditors
• Never block on a single section failure

Related Skills

• zenflow:check-work — Runs after fixes are applied
• zenflow:plan — Creates a structured plan from audit findings
• zenflow:refactor — For larger structural improvements surfaced by audit