Quality Reviewer

Deep quality review with web research to verify code against the latest ecosystem state.

Primary differentiator: Web research (WebSearch, WebFetch) to verify against current versions, documentation, and best practices.

Triggers:

• Explicit web research request: "double check against latest docs", "verify we're using latest version", "check for security issues"
• Deep dive needed: User wants analysis beyond automatic hook (performance, architecture alternatives, trade-offs)
• No SAFEWORD.md/CLAUDE.md: Projects without context files (automatic hook won't run, manual review needed)
• Pre-change review: User wants review before making changes (automatic hook only triggers after changes)
• Model-invoked: Claude determines web research would be valuable

Relationship to automatic quality hook:

• Automatic hook: Fast quality check using existing knowledge + project context (guaranteed, runs on every change)
• This skill: Deep review with web research when verification against current ecosystem is needed (on-demand, 2-3 min)

Review Protocol

1. Identify What Changed

Understand context:

• What files were just modified?
• What problem is being solved?
• What was the implementation approach?

2. Read Project Standards

ls CLAUDE.md SAFEWORD.md ARCHITECTURE.md .claude/

Read relevant standards:

• CLAUDE.md or SAFEWORD.md - Project-specific guidelines
• ARCHITECTURE.md - Architectural principles
• @./.safeword/guides/code-philosophy.md - Core coding principles

3. Evaluate Correctness

Will it work?

• Does the logic make sense?
• Are there obvious bugs?

Edge cases:

• Empty inputs, null/undefined, boundary conditions (0, -1, max)?
• Concurrent access, network failures?

Error handling:

• Are errors caught appropriately?
• Helpful error messages?
• Cleanup handled (resources, connections)?

Logic errors:

• Off-by-one errors, race conditions, wrong assumptions?

4. Evaluate Anti-Bloat

• Are all dependencies necessary? Could we use stdlib/built-ins?
• Are abstractions solving real problems or imaginary ones?
• YAGNI: Is this feature actually needed now?

5. Evaluate Elegance

• Is the code easy to understand?
• Are names clear and descriptive?
• Is the intent obvious?
• Will this be easy to change later?

6. Check Standards Compliance

Project standards (from CLAUDE.md/SAFEWORD.md/ARCHITECTURE.md):

• Does it follow established patterns?
• Does it violate any documented principles?

Library best practices:

• Are we using libraries correctly?
• Are we following official documentation?

7. Verify Latest Versions ⭐ PRIMARY VALUE

CRITICAL: This is your main differentiator from automatic hook. ALWAYS check versions.

WebSearch: "[library name] latest stable version 2025"
WebSearch: "[library name] security vulnerabilities"

Flag if outdated:

• Major versions behind → WARN (e.g., React 17 when 19 is stable)
• Minor versions behind → NOTE (e.g., React 19.0.0 when 19.1.0 is stable)
• Security vulnerabilities → CRITICAL (must upgrade)
• Using latest → Confirm

Common libraries: React, TypeScript, Vite, Next.js, Node.js, Vitest, Playwright, Jest, esbuild

Check even if dependencies didn't change - User might be using outdated patterns.

8. Verify Latest Documentation ⭐ PRIMARY VALUE

CRITICAL: This is your main differentiator from automatic hook. ALWAYS verify against current docs.

WebFetch: https://react.dev (for React)
WebFetch: https://vitejs.dev (for Vite)
WebFetch: https://www.electronjs.org/docs (for Electron)

Look for:

• Are we using deprecated APIs?
• Are there newer, better patterns?
• Did the library's recommendations change since training data?

Cache results: If you checked docs recently in this session, don't re-fetch.

Output Format

Simple question ("is it correct?"):

**Correctness:** ✓ Logic is sound, edge cases handled, no obvious errors.

Full review ("double check and critique"):

## Quality Review

**Correctness:** [✓/⚠️/❌] [Brief assessment]
**Anti-Bloat:** [✓/⚠️/❌] [Brief assessment]
**Elegance:** [✓/⚠️/❌] [Brief assessment]
**Standards:** [✓/⚠️/❌] [Brief assessment]
**Versions:** [✓/⚠️/❌] [Latest version check with WebSearch]
**Documentation:** [✓/⚠️/❌] [Current docs check with WebFetch]

**Verdict:** [APPROVE / REQUEST CHANGES / NEEDS DISCUSSION]

**Critical issues:** [List or "None"]
**Suggested improvements:** [List or "None"]

Use structured format for "double check"/"critique". Use brief format for specific questions.

Example: Full Review

## Quality Review

**Correctness:** ✓ Logic sound, edge cases covered, error handling adequate
**Anti-Bloat:** ✓ Minimal dependencies, appropriate abstractions
**Elegance:** ✓ Clear code, good naming, well-structured
**Standards:** ✓ Follows CLAUDE.md patterns
**Versions:** ✓ React 19.0.0 (latest stable), TypeScript 5.7.2 (latest)
**Documentation:** ✓ Using current React patterns per https://react.dev

**Verdict:** APPROVE - Production ready

**Critical issues:** None
**Suggested improvements:** None

Critical Reminders

1. Primary value: Web research - Use WebSearch/WebFetch to verify against current ecosystem (versions, docs, security)
2. Complement automatic hook - Hook does fast check with existing knowledge, you do deep dive with web research
3. Explicit triggers matter - "double check against latest docs", "verify versions", "check security" = invoke web research
4. Projects without SAFEWORD.md - Automatic hook won't run, you're the only quality check
5. Always check latest docs - Verify patterns are current, not outdated (WebFetch)
6. Always verify versions - Flag outdated dependencies (WebSearch)
7. Be thorough but concise - Cover all areas but keep explanations brief
8. Provide actionable feedback - Specific line numbers, concrete suggestions
9. Clear verdict - Always end with APPROVE/REQUEST CHANGES/NEEDS DISCUSSION
10. Separate critical vs nice-to-have - User needs to know what's blocking vs optional

Non-Obvious Edge Cases

User requests review after automatic hook ran:

• Acknowledge hook ran: "The automatic quality hook already did a fast check. I'll now do deeper analysis with web research..."
• Focus on what automatic hook doesn't do: fetch latest docs, verify versions, security checks, performance analysis

WebSearch/WebFetch fails:

• Continue review without version/docs checks
• Note: "Couldn't verify latest versions/docs, skipping that check"

Project has no CLAUDE.md/SAFEWORD.md:

• Use @./.safeword/guides/code-philosophy.md as fallback
• Note: "No project-specific standards found, using general best practices"