Just describe
your goal.

Chainer lets you automate sequences of actions in Claude Code by just describing what you want in plain English.

The simple mental model: You tell Chainer what you want to build. Chainer figures out the steps, shows you what it'll do, and asks for confirmation. Then it runs everything automatically.

How it works:

1. You type something like /chainer:run Add user login
2. Chainer shows you: "I'll run feature-dev to plan, then ralph-loop to implement"
3. You click Start (or Edit if you want to change something)
4. Chainer runs each step automatically
5. You come back to working code

Example: Instead of remembering commands and syntax, just say /chainer:run build payments in a new worktree and Chainer handles creating the isolated workspace, planning, and implementing.

Very early beta. Chainer is experimental and should be considered a proof-of-concept. Only a handful of people have tested it so far.

Expect rough edges, breaking changes, and incomplete features. Some features shown on this site (like Answer History and interactive Start/Edit/Cancel buttons) are designed but not yet implemented.

Use at your own risk. This notice will be updated when the project matures.

You need Claude Code installed. That's it for basic use.

The built-in chains use these plugins, so install them to use those chains:

• Worktree Manager - For the worktree-plan-implement chain
• feature-dev - For planning chains
• ralph-loop - For implementation chains

Missing a plugin? Chainer will detect missing dependencies and show exactly how to install them.

Chainer officially supports and is tested with:

• feature-dev - Planning and architecture
• ralph-loop - Autonomous implementation
• Worktree Manager - Isolated workspaces
• frontend-design - UI/UX design

Many other plugins should work in custom chains. See SUPPORTED_PLUGINS.md for the full compatibility matrix and feature support details.

Add a chainer.config.yaml file to your project. Define patterns (words that trigger your workflow) and chains (what steps to run):

patterns:
  - match: ["with tests", "and test"]
    chain: plan-implement-test

chains:
  plan-implement-test:
    steps:
      - skill: feature-dev
      - skill: ralph-loop
      - script: npm test

Now when you say /chainer:run Add OAuth with tests, Chainer detects "with tests" and runs your custom workflow.

See the community patterns for more examples.

No syntax to remember: Just describe what you want. Chainer figures out the right plugins and order.

Transparency: See exactly what will happen before it runs. Edit or cancel if it's not what you want.

Consistency: The same steps happen every time, in the same order.

Shareability: Teams can share patterns so everyone uses the same workflows.

Yes! Chainer supports parallel execution using tmux.

When you try to run a second chain while one is already running, Chainer detects the conflict and offers to spawn a new tmux window or session. Each chain runs independently in its own isolated environment.

Use /chainer:status to see all running and recently completed chains.

No, they're optional. But, Chainer was designed to work best with other plugins like Worktree Manager.

The power comes from pairing Chainer with other plugins for workflow automation. Chainer chains together feature-dev (planning) and ralph-loop (implementation) so you can describe what you want in plain English and come back to a completed feature in an isolated worktree.

Don't worry about missing plugins: Chainer automatically detects missing dependencies before running a chain and shows you exactly how to install them.

This is a known quirk of ralph-loop's design. To exit the loop, Claude must output the literal text <promise>DONE</promise>. Sometimes Claude completes the work but says things like "Done", "Complete", or "Finished" instead of outputting the exact escape tag.

When this happens, you'll see iterations increment (e.g., "Ralph iteration 25... 26... 27...") with Claude repeatedly confirming completion but not exiting. The loop will run until it hits the max iterations limit (default: 50).

This is harmless but wasteful. Your work was completed successfully—the extra iterations just burn through API credits. You can manually stop Claude (Ctrl+C) once you see the work is done, or wait for max iterations.

Update: We've added explicit exit instructions to the task prompt, and testing shows this significantly improves Claude's exit behavior. Most tasks now exit cleanly with the proper tag.

If you still see excessive looping:

1. Lower default_max_iterations in your config file (e.g., to 20) to cap wasted iterations.

2. Manually stop Claude (Ctrl+C) once you see the work is done.

We'd love to see what workflows you create! To contribute a chain to the community library:

1. Create your chain in a YAML file (e.g., my-workflow.yaml)
2. Fork the Chainer repository
3. Add your chain to the community-chains/ directory
4. Include a README explaining what it does and what plugins it requires
5. Submit a pull request

Check out existing chains in the community-chains library for examples.

Not from the same directory. Ralph-wiggum's state file (.claude/ralph-loop.local.md) is tied to your Claude session's current working directory, not the worktree. If you run multiple Claude sessions from the same parent directory, they'll all see the same state file and interfere with each other.

Workaround: Run each Claude session from within the worktree directory instead of your main project directory:

cd ~/worktrees/feature-a
claude

This way each worktree has its own isolated .claude/ directory, and ralph loops won't cross-contaminate. You can then have multiple terminals, each running Claude from a different worktree, all working independently.