Copy-paste AI agent workflows for real-world developer tasks.
A community-driven collection of ready-to-use AI agent workflows. Each recipe is a self-contained prompt you can drop into Claude Code, Cursor, Aider, or any AI coding assistant.
| Recipe | Description |
|---|---|
| PR Review | Comprehensive pull request review with security, performance, and style checks |
| Architecture Review | High-level codebase architecture analysis and recommendations |
| Dependency Audit | Check for outdated, vulnerable, or unnecessary dependencies |
| Performance Review | Identify performance bottlenecks and optimization opportunities |
| Multi-Repo PR Check | Check all your open PRs across multiple repos in parallel |
| Recipe | Description |
|---|---|
| Generate Unit Tests | Auto-generate unit tests for uncovered functions |
| E2E Test Writer | Generate end-to-end tests from user stories |
| Coverage Gap Finder | Find untested code paths and edge cases |
| Recipe | Description |
|---|---|
| JavaScript to TypeScript | Migrate JS files to TypeScript with proper types |
| React Class to Hooks | Convert React class components to functional + hooks |
| CJS to ESM | Convert CommonJS requires to ES module imports |
| Upgrade Dependencies | Safely upgrade all project dependencies with automated testing |
| Recipe | Description |
|---|---|
| README Generator | Generate comprehensive README from repo analysis |
| API Docs Generator | Generate OpenAPI/Swagger docs from code |
| Changelog Writer | Generate changelog from git commits |
| Recipe | Description |
|---|---|
| Secret Scanner | Find hardcoded secrets, keys, and credentials |
| OWASP Top 10 Audit | Check code against OWASP Top 10 vulnerabilities |
| Input Validation | Add input validation and sanitization to API endpoints |
| Recipe | Description |
|---|---|
| Dockerfile Generator | Generate optimized multi-stage Dockerfiles |
| CI/CD Pipeline | Generate GitHub Actions / GitLab CI from project analysis |
| Monitoring Setup | Add health checks, metrics, and alerting |
| Background Deploy Monitor | Monitor a deployment in the background while continuing other work |
Run the prompt, not the whole file. A recipe also contains a worked example, and if you pipe the entire file into an agent it reads that example's output as part of its instructions. scripts/extract-prompt.sh prints only the fenced block under ## The Prompt.
# Run a recipe's prompt directly
claude -p "$(bash scripts/extract-prompt.sh recipes/code-review/pr-review.md)"
# Or save it as a custom command
mkdir -p .claude/commands
bash scripts/extract-prompt.sh recipes/code-review/pr-review.md > .claude/commands/review.md
# Then use: /review- Open the recipe
.mdfile - Copy the fenced block under
## The Promptinto your AI chat - Run on your codebase
Each recipe does one thing, so chain them when the task is bigger than one prompt. Run the first, then feed its report to the second:
- Coverage Gap Finder to find the untested paths, then Generate Unit Tests for the gaps it named
- Secret Scanner to clear hardcoded credentials first, then OWASP Top 10 Audit on the rest of the surface
- Dependency Audit to get the license and maintenance picture, then Upgrade Dependencies to actually move the versions
- Architecture Review before any migration recipe, so the migration lands in the structure you want rather than the one you have
Every recipe is one markdown file with the same sections in the same order:
| Section | Purpose |
|---|---|
# Title and a > line |
Recipe name plus a one-line description |
## When to Use |
The scenario it fits, the problem it solves, any prerequisites |
## The Prompt |
The fenced, copy-paste prompt. This is the recipe |
## Example |
### Input and ### Output showing realistic results |
## Customization Tips |
How to adapt it to other languages, stacks, or stricter rules |
## Cost |
Optional: a rough cost range for running the recipe |
## Tags |
Backtick-quoted tags for discoverability |
CONTRIBUTING.md has the full template and is the source of truth for the format. bash scripts/lint-recipes.sh checks every recipe against it, and CI runs the same check on every pull request.
Have a useful AI workflow? Share it!
- Fork this repo
- Create a recipe in the appropriate
recipes/subdirectory - Follow the recipe format
- Add a row for it to the index above
- Run
bash scripts/lint-recipes.shto check both - Submit a PR
Ideas for new recipes? Open an issue!
See CONTRIBUTING.md for full guidelines.
Most open issues are requests for recipes that do not exist yet rather than bug reports. Five of them, all labelled roadmap:
- Accessibility audit recipe
- Database schema reviewer recipe
- Python-specific recipes
- Refactoring recipes category
- Infrastructure recipes (K8s, Terraform, Helm)
Browse the full backlog by label: good first issue, help wanted, roadmap.
| Project | Description |
|---|---|
| claude-code-recipes | 47 copy-paste recipes for Claude Code - commands, subagents, hooks, skills, MCP integration, and workflow patterns |
| claude-cost-optimizer | An installable skill, CLI tools, and deep-dive guides for cutting Claude Code token spend |
| ai-git-hooks | AI-powered git hooks - auto-review diffs, generate commit messages, scan for secrets. Supports Claude, OpenAI, and Ollama |
| mcp-toolkit | TypeScript middleware toolkit for MCP servers - authentication, caching, rate limiting, CORS, logging (beta) |
| claude-skills | Custom Claude Code plugin marketplace with skills for FARM stack, open source contributions, repo maintenance, and portfolio management |