claude-review

name	claude-review
description	[pr-review-focus-area: Claude Config] Holistic audit of .claude/ configuration, skills inventory, CLAUDE.md quality, settings, hooks, and memory.
user-invocable	true
disable-model-invocation	true
allowed-tools	["Read","Grep","Glob"]

Claude Configuration Review

Audit the project's Claude Code configuration for completeness, correctness, and coherence.

Invocation

/claude-review — reviews the entire .claude/ directory and CLAUDE.md.

Checks

1. Skills Inventory

Scan for all SKILL.md files in the project (including plugin skills).

• List each skill with: name, description, invocability, allowed tools.
• Flag skills with missing or malformed frontmatter.
• Flag skills with no description.
• Flag duplicate skill names across directories.
• Check that all skills referenced in CLAUDE.md actually exist.

2. CLAUDE.md Review

Evaluate the project's CLAUDE.md as an LLM context document:

• Completeness: Does it cover project structure, key commands, conventions, and architecture?
• Accuracy: Do referenced files, commands, and paths actually exist?
• Conciseness: Is it focused and free of redundant information?
• Structure: Is it well-organized with clear headers and sections?
• Actionability: Does it contain specific, actionable guidance (not vague principles)?

3. Settings & Hooks Review

Review .claude/settings.json (or equivalent):

• List all configured hooks (pre-commit, post-commit, custom).
• Verify hook commands are valid and executable.
• Flag hooks that reference missing scripts or tools.
• Check for conflicting settings.

4. Memory Review (Read-Only)

If a memory file exists (.claude/memory.md, .claude/memory.json, or equivalent):

• Read and summarize key stored preferences and decisions.
• Do NOT modify memory. This is read-only.
• Flag contradictions between memory and CLAUDE.md.

5. Holistic Coherence

Cross-check all configuration for consistency:

• Skills should align with project type (e.g., no database skills for a static site).
• CLAUDE.md guidance should match configured hooks and settings.
• Tool restrictions in skills should make sense (e.g., review skills shouldn't need Write).
• Flag configuration that appears copied from another project without adaptation.

Report Format

## Claude Configuration Audit

### Skills: 12 found
- preflight (user-invocable, 6 tools) ✓
- api-review (user-invocable, 3 tools) ✓
- [WARN] write-tests — missing description in frontmatter

### CLAUDE.md Quality: GOOD
- [WARN] References `npm run test:integration` but no such script exists
- [PASS] Well-structured, complete project overview

### Settings: OK
- 2 hooks configured, both valid

### Memory: 3 entries
- [PASS] No contradictions with CLAUDE.md

### Coherence: GOOD
- [WARN] dba-review skill present but no database configuration detected