Audit Suite

---
description: "Run a preset audit workflow (pre-release, audio-deep, security-deep, etc.) — executes a sequence of audits"
argument-hint: "<preset-name>"
---

# Audit Suite — Preset Workflows

Run a predefined sequence of audits. This command **actually executes** each audit in the preset by launching a Task agent per audit.

See `.claude/commands/_audit-common.md` for project layout, methodology, and context management rules.

## Parameters (from $ARGUMENTS)

- `--preset <name>`: Which preset to run. Required. Options below.

## Available Presets

### `pre-release` — Run before tagging a release

1. /audit-regression — Verify all known fixes still in place
2. /audit-incremental — Check recent changes for new bugs
3. /audit-security --depth shallow — Quick security sanity check

**When**: Before every release tag.

### `comprehensive` — Full platform audit

1. /audit-engine — Deep audio engine dive
2. /audit-backend — Deep backend dive
3. /audit-frontend — Deep frontend dive
4. /audit-integration — Cross-layer flow tracing
5. /audit-security — OWASP Top 10
6. /audit-concurrency — Race conditions and state integrity
7. /audit-deprecation — Deprecated APIs and patterns
8. /audit-tech-debt — Accumulated debt (markers, dead code, duplication, complexity)

**When**: Monthly or after major architecture changes.

### `tech-debt-deep` — Surface accumulated debt

1. /audit-tech-debt — All 10 debt dimensions
2. /audit-incremental --commits 30 — New debt introduced this sprint

**When**: After a milestone closes, before opening the next.

### `post-sprint` — End-of-sprint verification

1. /audit-incremental --commits 30 — All changes this sprint
2. /audit-regression — Verify fixes from this sprint

**When**: End of every sprint.

### `security-deep` — Deep security review

1. /audit-security — Full OWASP Top 10
2. /audit-frontend --focus auth,security — Frontend auth/security
3. /audit-integration --flows 1,5 — Playback + WebSocket auth flows

**When**: After auth changes, before security review.

### `audio-deep` — Audio pipeline integrity

1. /audit-engine — Full engine audit
2. /audit-integration --flows 1,3 — Playback + Enhancement flows
3. /audit-concurrency --focus 1,2 — Player + Processing pipeline

**When**: After DSP changes, crossfade modifications, or parallel processing updates.

## Execution

### Step 1: Print the Execution Plan

Print a table showing which audits will run:

Audit Suite:

Step	Audit	Focus	Type
1	audit-regression	All known fixes	Lightweight
2	audit-incremental	Last 10 commits	Lightweight
3	audit-security --depth shallow	Quick check	Deep (subagents)

### Step 2: Launch Audits in Parallel

Each audit is independent — they read different files and write separate reports. Launch them ALL simultaneously using the Agent tool.

**Parallelization strategy by preset**:

- **`comprehensive`** (7 audits): Launch ALL 7 as background agents in a single message. Each writes to its own report file — no conflicts.
- **`pre-release`** (3 audits): Launch all 3 in parallel.
- **`post-sprint`** (2 audits): Launch both in parallel.
- **`security-deep`** (3 audits): Launch all 3 in parallel.
- **`audio-deep`** (3 audits): Launch all 3 in parallel.

**Agent configuration for each audit**:
- `subagent_type`: `general-purpose`
- `run_in_background`: `true`

**Agent prompt template** (adapt for each audit):

You are running the audit as part of the suite.

CRITICAL: You MUST perform a FRESH audit by reading and analyzing the CURRENT source code. Do NOT reuse, summarize, or reference any existing report files in docs/audits/. Existing reports may be outdated — the code has changed since they were written. Delete any previous report at the target path before starting.

Read the audit command file at .claude/commands/audit-<name>.md and follow ALL of its instructions exactly:

• Phase 1: Setup (create /tmp dirs, fetch dedup baseline)
• Phase 2: Launch dimension agents (as described in the file)
• Phase 3: Merge results into the final report
• Phase 4: Cleanup

<any --focus or --depth overrides from the preset>

Write the final report to: docs/audits/AUDIT__.md

IMPORTANT: Follow the context management rules from .claude/commands/_audit-common.md.

**Launch example for `comprehensive`**:
Send a SINGLE message with 7 Agent tool calls, all with `run_in_background: true`. Do NOT wait for one to finish before launching the next.

### Step 3: Wait for All Agents to Complete

You will be notified as each background agent completes. As each finishes, note:
- Whether it succeeded or failed
- The report path (if written)
- The finding count (read the executive summary section)

Wait until ALL agents have completed before proceeding to Step 4.

### Step 4: Print Final Summary

After ALL audits have completed, print:

Audit Suite Results:

#	Audit	Status	Report	Findings
1	audit-regression	DONE	docs/audits/AUDIT_REGRESSION_.md	3 PASS, 1 FAIL
2	audit-incremental	DONE	docs/audits/AUDIT_INCREMENTAL_.md	5 findings
3	audit-security	DONE	docs/audits/AUDIT_SECURITY_.md	2 CRITICAL, 4 HIGH

Next Steps

For each report with NEW findings, run /audit-publish in a new conversation:

• /audit-publish docs/audits/AUDIT_REGRESSION_<TODAY>.md
• /audit-publish docs/audits/AUDIT_INCREMENTAL_<TODAY>.md
• /audit-publish docs/audits/AUDIT_SECURITY_<TODAY>.md

## Important Notes

- Each audit runs as an isolated background agent — it will NOT exhaust this conversation's context.
- Deep audits (engine, backend, frontend, etc.) internally launch