A Claude Code skill that systematically reviews your project's setup using the six-layer framework: CLAUDE.md → rules → skills → hooks → subagents → verifiers. It detects project complexity, runs two parallel diagnostic agents, and outputs a prioritized report telling you what to fix first.
Recommended, install globally for Claude Code:
npx skills add hiclaude/health -a claude-code -s health -g -yInstall only in the current project:
npx skills add hiclaude/health -a claude-code -s health -yClaude Plugin:
claude plugin marketplace add hiclaude/health
claude plugin install healthRestart Claude Code after installation. Then in any Claude Code session, run /health or just say:
"Run a health check on my Claude Code config"
The skill automatically detects your project tier (Simple / Standard / Complex) and calibrates checks accordingly. It won't flag missing layers that aren't needed for your project size.
Unknown skill: health: install to Claude Code explicitly with-a claude-code. Use-gif you want the skill available in every project. Restart Claude Code after installation.- Installer summary shows
./.agents/skills/health: that summary is generic. For Claude Code, the installed path should end up at./.claude/skills/healthor~/.claude/skills/health. - Scope: this repository targets Claude Code only. Codex and OpenClaw are not supported here.
| Layer | Checks |
|---|---|
| CLAUDE.md | Signal-to-noise ratio, missing Verification/Compact Instructions, prose bloat |
| rules/ | Language-specific rules placement, coverage gaps |
| skills/ | Description token count, trigger clarity, auto-invoke strategy, frequency-based optimization |
| skill security | Prompt injection, data exfiltration, destructive commands, hardcoded credentials, obfuscation, safety overrides |
| hooks | Pattern field presence, file-type coverage, stale entries |
| MCP | Server count, token cost estimation, context pressure detection, filesystem allowlist failures |
| allowedTools | Dangerous or stale one-time commands |
| Prompt Cache | Dynamic timestamps, tool reordering, mid-session model switching |
| Three-Layer Defense | Critical rules covered by CLAUDE.md + Skill + Hook together |
| Behavior | Rules violated in practice, repeated corrections, context hygiene habits |
Results are grouped into three priority levels:
- Critical: Fix now: rule violations, dangerous permissions, cache-breaking patterns, MCP overhead >12.5%, skill security issues (prompt injection, data exfiltration, etc.)
- Structural: Fix soon: misplaced content, missing hooks, single-layer critical rules, skill quality issues (bloated content, broken references)
- Incremental: Nice to have: context hygiene, HANDOFF.md adoption, skill frequency tuning, skill provenance (symlink sources, version tracking)
Built around the six-layer framework. Run /health to quickly see how your configuration measures up.
MIT License, feel free to enjoy and participate in open source.