What problem does it solve? Claude Code projects accumulate permission prompts, bloated CLAUDE.md files, leaked secrets in settings, and stale documentation. This Skill centralizes project setup, health auditing, permission tidying, and learning capture into one workflow so configuration stays clean over a project's lifetime. ## Core Features & Use Cases - Full Health Audit: Launches parallel sub-agents to check permissions (missing MCP servers, leaked secrets, legacy syntax, junk entries) and context quality (CLAUDE.md scoring, stale references, missing docs). - New Project Setup: Detects project type from indicator files (wrangler.toml, package.json, Cargo.toml, etc.) and generates settings.local.json, CLAUDE.md, and .gitignore from curated presets. - Permission Presets & Capture Learnings: Merges language/deployment permission presets without duplicates and saves session discoveries to the right context file. - Use Case: After weeks of development, run a health check to find three leaked API keys in settings.local.json, an oversized 400-line CLAUDE.md, and missing ARCHITECTURE.md — then apply all fixes with one confirmation. ## Quick Start Ask the agent to run a project health check and fix any permission or context issues it finds.