claude-code-enhance

Scar-tissue retrieval. Loads relevant past lessons from ~/.claude/lessons/ before producing any architectural plan, so you don't repeat mistakes that already cost time/money. Use this skill at the START of any task involving: authentication, JWT, OAuth, RBAC, sessions, multi-tenancy, permissions, caching (any layer), cache invalidation, TTLs, database migrations, schema changes, ORM upgrades, indexes, distributed locking, payment flows, refunds, webhook handling, queue processing, retry logic, error recovery, rate limiting, deploys, infrastructure changes, API contract design, response wrappers, response formats, frontend-backend type contracts, NestJS modules, NestJS DI, Prisma client setup, Next.js layouts, Server Components, seed data, demo data, fake data generation. Triggers on prompts containing "design", "refactor", "migrate", "introduce", "implement", "build", "set up", "fix", "upgrade", "add caching", "add auth", "ship", "deploy", "schema change", "data model", "memory system", "agent architecture", "tune Claude", "context optimization", "improve Claude" or any combination of action verbs with the topics above. ALSO use when user asks to plan, spec, or architect anything non-trivial. The cost of skipping this skill is repeating a known-broken pattern; the cost of running it is ~30 seconds and ~500 tokens.

Augment-style prompt enhancer for Claude Code. Takes a rough developer intent and produces a production-quality, codebase-aware, lesson-cross- referenced, prompt-engineering-optimized prompt that the user reviews and approves before execution. USE THIS SKILL when the user types "/enhance <rough idea>", "enhance: <idea>", "improve this prompt: <idea>", "structure this: <idea>", "expand this: <idea>", "make this prompt better:", "rephrase as a proper prompt:", or asks for a better-written/structured version of their request before any actual work begins. ALSO use when the user gives a vague high-level request with multiple possible interpretations and you want to align with them BEFORE coding — the enhancement output surfaces ambiguities as explicit questions. The skill loads project context (codemap, lessons, CLAUDE.md, recent git activity) and applies real prompt engineering principles (clear goal, scope, constraints, success criteria, suggested approach, anti-patterns, verification). The user can submit, edit, or scrap the enhanced version. This is a USER-FACING enhancement tool. The user must approve the enhanced version before it becomes the actual instruction. Never auto- apply the enhanced prompt without explicit user confirmation.

Regenerates the project codemap files (hotfiles.md, recent.md) from git history. Run weekly or before starting major architectural work to keep the SessionStart-injected codemap fresh. Triggers on user prompts like "refresh codemap", "update hotfiles", "regenerate codemap", "what's been hot lately", "what changed this month", "rebuild project map", or after significant rebases / merges that may have shifted the change-frequency picture. Does NOT regenerate overview.md — that's hand-curated and should only be edited when the architecture genuinely changes.