Claude Code's Missing Context Layer

What it does

This kit scaffolds a .claude/ directory and four focused context documents into your project, turning a static CLAUDE.md into a living system that evolves with your code. It is essentially the organizational layer you’d build yourself after a few weeks of Claude Code usage, packaged for day one. The installer drops slash-command templates, safety hooks, and optional skills directly into your repo without touching existing code.

The interesting bit

The project treats documentation as working memory rather than reference material. A built-in deny list blocks destructive commands like git push --force, while optional hooks spawn parallel sub-agents to review diffs or pull in Google’s Gemini CLI for an independent second opinion when Claude hits a dead end.

Key highlights

• Four-file context system in docs/ai-context/ — spec.md, project-structure.md, progress.md, and deployment-infrastructure.md — that /prime loads at session start
• /update-docs skill keeps the context files synchronized with code changes, skipping trivial diffs
• Core safety hooks block destructive commands and scan for API keys before they leak to MCP plugins
• Optional quality gates include parallel “Bug Hunter” and “Rules Auditor” sub-agents, plus a review-on-stop hook with a three-tier advisory nudge
• Clean uninstall removes only the scaffolding, leaving your source code untouched

Caveats

• Skills and hooks are Claude Code-specific; the markdown templates work elsewhere, but the workflow does not
• /second-opinion and /plan-feature require a separate Gemini CLI installation
• Explicitly designed for individual developers and small teams, not large-scale multi-agent API orchestration

Verdict

Pick this up if you subscribe to Claude Code and want your project context to survive the weekend. Pass if you already orchestrate bespoke multi-agent workflows or prefer keeping your AI assistant stateless.