Deep Code Review — Multi-Agent Analysis

You are orchestrating a comprehensive code review using 5 specialized reviewer agents running in parallel. This produces a detailed review document with severity-tagged findings.

Scope

The review scope is: $ARGUMENTS (default: all — review the entire src/ directory)

Instructions

1. Prepare the review

• •Generate a real timestamp by running date -u +%Y-%m-%dT%H-%M-%S via the Bash tool. Use the actual output — do NOT hardcode or guess a timestamp.
• •Determine files to review:
  • •If scope is all or empty: all .ts files under src/
  • •If scope is a directory: all .ts files under that directory
  • •If scope is a glob: matching files
• •Read the reference checklists:
  • •.claude/skills/deep-review/references/review-checklist.md
  • •.claude/skills/deep-review/references/security-checklist.md
• •Check if docs/reviews/accepted-deviations.md exists. If it does, read it — these are items that have been reviewed and accepted in prior reviews. Agents MUST skip these items.
• •Check if a coding standard exists at .claude/docs/ (e.g., a TypeScript coding standard) and load it if present

2. Launch 5 parallel reviewer agents

Use the Task tool to launch these 5 agents in parallel. Each agent should receive:

• •The list of files to review
• •The relevant checklist/standard sections
• •If the accepted deviations file exists, its full contents — agents MUST NOT re-flag any accepted deviation
• •Instructions to output findings in the structured format below

Read the agent definitions from .claude/agents/ for each reviewer's specific instructions:

Agent 1: Security Reviewer

• •Agent definition: .claude/agents/security-reviewer.md
• •Focus: OWASP Top 10, input validation (Rules 7.2, 7.4), secrets, injection, crypto

Agent 2: Standards Reviewer

• •Agent definition: .claude/agents/standards-reviewer.md
• •Focus: All shall-level rules from the coding standard

Agent 3: YAGNI Reviewer

• •Agent definition: .claude/agents/yagni-reviewer.md
• •Focus: Over-engineering, dead code, unnecessary abstractions, premature optimization

Agent 4: Architecture Reviewer

• •Agent definition: .claude/agents/architecture-reviewer.md
• •Focus: Modularity (Rule 10.3), dependency direction, SOLID principles, circular deps

Agent 5: Goal Reviewer

• •Agent definition: .claude/agents/goal-reviewer.md
• •Focus: Does the code achieve its stated goals? Edge cases? Completeness?

3. Consolidate findings

After all 5 agents complete, merge their findings into a single review document.

Each finding should be tagged with a severity:

• •CRITICAL — Must fix. Security vulnerability, data loss risk, or standard violation that could cause runtime failure
• •HIGH — Should fix. Significant standard violation or architectural concern
• •MEDIUM — Consider fixing. Minor standard deviation or improvement opportunity
• •LOW — Nice to have. Style suggestion or minor optimization

De-duplicate findings that appear in multiple reviewers' output (keep the highest severity).

4. Write the review document

Create the directory docs/reviews/ if it doesn't exist, then write the consolidated review to docs/reviews/review-${TIMESTAMP}.md.

Use this template — note the distinction between open findings - [ ] and the two resolution states [x] Fixed vs [~] Accepted:

# Deep Code Review — ${TIMESTAMP}

**Scope**: ${SCOPE}
**Files reviewed**: N
**Reviewers**: Security, Standards, YAGNI, Architecture, Goal

## Executive Summary

<2-4 sentences summarizing overall code quality and top concerns>

**Findings**: N total (X critical, Y high, Z medium, W low)

## Critical Findings

- [ ] C-1: `file.ts:42` — **[Security]** Description of the finding

## High Findings

- [ ] H-1: `file.ts:88` — **[Architecture]** Description of the finding

## Medium Findings

- [ ] M-1: `file.ts:22` — **[YAGNI]** Description of the finding

## Low Findings

- [ ] L-1: `file.ts:5` — **[Goal]** Description of the finding

## Per-Reviewer Details

### Security Review

<Security reviewer's detailed findings>

### Standards Compliance Review

<Standards reviewer's detailed findings>

### YAGNI Review

<YAGNI reviewer's detailed findings>

### Architecture Review

<Architecture reviewer's detailed findings>

### Goal Alignment Review

<Goal reviewer's detailed findings>

## Recommendations

1. Prioritized list of actions
2. ...

---

_Generated by /deep-review on ${TIMESTAMP}_

Resolution marking (for when findings are addressed later — NOT during the review itself):

• •- [x] H-1: ... **Fixed:** <brief description of the code change>
• •- [~] H-1: ... **Accepted:** <rationale for why this is not being fixed>

When a finding is accepted, it should also be added to docs/reviews/accepted-deviations.md so future reviews skip it.

5. Report to user

• •Show the file path of the generated review
• •Display the executive summary and finding counts by severity
• •If there are critical findings, emphasize they must be addressed before deployment