team-review

name	team-review
description	Parallel multi-agent code review from multiple angles. Trigger: team review, parallel review, multi-reviewer, thorough review

/team-review — Parallel Subagent Code Review

Spawn parallel review subagents that examine the codebase simultaneously from different angles. Each reviewer runs in its own context via the Task tool.

This is a read-only operation. No reviewer edits files. The lead synthesizes all findings.

---

When to Use (vs. /review)

Use /review when...	Use /team-review when...
Reviewing a small change (< 5 files)	Reviewing a large change (10+ files)
Quick feedback is needed	Thorough, multi-dimensional analysis is needed
Token budget is a concern	Quality and depth matter more than cost
Changes are in a single module	Changes span multiple modules/layers

---

Steps

1. Determine Scope

• /team-review → all uncommitted changes
• /team-review [file, dir, or commit range] → specific scope
• /team-review --staged → staged changes only

2. Analyze the Change Set

Read all changed files. Categorize the nature of changes to determine which reviewers to spawn.

3. Spawn Review Subagents

Read the agent definition files from .claude/agents/ and spawn parallel Task tool calls. All reviewers are read-only — they produce reports, never edit files.

Standard Team (most changes) — spawn ALL in the same response:

Task(subagent_type="security-code-auditor",
     prompt="<content of .claude/agents/security-reviewer.md>\n\n
     Review these files for security issues: {changed_files}\n
     Focus: OWASP top 10, input validation, secrets, auth\n
     Return a structured report with severity ratings.")

Task(subagent_type="senior-code-reviewer",
     prompt="<content of .claude/agents/code-reviewer.md>\n\n
     Review these files: {changed_files}\n
     Focus: correctness, readability, DRY/KISS/SOLID, anti-patterns\n
     Return a structured report with severity ratings.")

Task(subagent_type="senior-code-reviewer",
     prompt="<content of .claude/agents/test-reviewer.md>\n\n
     Review these test files: {changed_test_files}\n
     And their corresponding source files: {changed_source_files}\n
     Focus: TDD compliance, coverage gaps, test quality, test smells\n
     Return a structured report with severity ratings.")

Task(subagent_type="senior-code-reviewer",
     prompt="<content of .claude/agents/architecture-reviewer.md>\n\n
     Review these files: {changed_files}\n
     Focus: module boundaries, coupling, SOLID at system level\n
     Return a structured report with severity ratings.")

Extended Team (for large/critical changes, add):

Task(subagent_type="senior-code-reviewer",
     prompt="<content of .claude/agents/performance-reviewer.md>\n\n...")

Task(subagent_type="senior-code-reviewer",
     prompt="<content of .claude/agents/doc-reviewer.md>\n\n...")

4. Collect Results

All Task calls return their reports. Collect all outputs.

5. Synthesize

Merge all reports into a single review:

## Team Code Review

**Scope:** [what was reviewed]
**Reviewers:** [N] subagents
**Overall Assessment:** [Excellent / Good / Needs Work / Significant Issues]
**Quality Score:** [N]/100

### Critical Issues (must fix before commit)
[Merged from all reviewers, deduplicated, attributed to source reviewer]
- **[reviewer]** [file:line] — [issue]

### Major Issues (should fix before PR)
[Same format]

### Minor Issues (polish)
[Same format]

### Reviewer Summaries
| Reviewer | Assessment | Critical | Major | Minor |
|----------|-----------|----------|-------|-------|
| Code | [status] | [N] | [N] | [N] |
| Security | [status] | [N] | [N] | [N] |
| Test | [status] | [N] | [N] | [N] |
| Architecture | [status] | [N] | [N] | [N] |

### Recommended Actions (priority order)
1. [Most critical fix]
2. [Next priority]

---

Adversarial Guarantee

This skill enforces complete separation of review and implementation:

• Review subagents are spawned as read-only — they cannot edit files
• If the reviewed code was written by subagents, the review subagents MUST use different agent definitions
• The lead does NOT review — it only synthesizes
• Findings are reported with severity; the lead does not downgrade severity without justification