Guide · Agent skill
How to use the LFG Autonomous Pipeline agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The lfg skill is an orchestrator from the Compound Engineering plugin. When you explicitly ask for hands-off, end-to-end work, it routes your request to the right Compound Engineering skills and aims to finish a code change as an open pull request. This guide explains what it does, how to install it, how to trigger it, and when to choose a step-by-step skill instead.
LFG Autonomous PipelineEveryInc/compound-engineering-plugin · MIT Download the skill Quick start
- 1Install the skill. Download the lfg ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 2Confirm the child skills are installed. Check that the Compound Engineering skills lfg calls, such as ce-plan, ce-work and ce-debug, are available in the same host.
- 3Check git and the remote. Make sure the repository has a configured git remote if you want pushes and a pull request. Without one, the run makes local commits only.
- 4Make an explicit request. Ask for autonomous, end-to-end work or name lfg, and include a feature, bug, issue reference or plan path.
- 5Let the pipeline run. lfg routes the request, implements or debugs, simplifies, reviews, commits and pushes without pausing for approval.
- 6Check the result. Review the open pull request, the recorded residual findings, and the DONE report. Merge it yourself unless you granted merging for the run.
What the lfg skill makes the agent do
lfg is a pipeline controller. It does not do the work itself. It reads your request, picks the Compound Engineering skill whose job matches it, and calls that skill. For a code change, the run goes through implementation, simplification, code review, applying the review fixes, committing, pushing, and opening a pull request. It then watches the pull request until CI is decided.
If the result is not a code change, such as an explanation or a prototype, the owning skill produces it and the run ends there. Merging stays with you unless you granted it for that run. The skill stops and reports when a required step is blocked or when an irreversible action falls outside what you granted.
- Routes plan paths and plans written earlier in the session to ce-work for implementation.
- Sends concrete bug reports to ce-debug, which reproduces, fixes, and commits.
- Sends unsettled judgment calls to ce-pov, and product questions to ce-brainstorm when a human is present.
Install the lfg skill
Download the ZIP for the skill from this page. You can install it in one of three ways.
For a personal install, unzip the download into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal: unzip into ~/.claude/skills/
- Per project: unzip into .claude/skills/ inside the project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
What lfg needs before it runs
lfg depends on other skills and on git. The skill calls the Compound Engineering child skills, so they must be installed in the same host. Those include ce-plan, ce-work, ce-debug, ce-simplify-code, ce-code-review, ce-commit-push-pr, ce-brainstorm, ce-pov and others. Without them, the pipeline cannot complete.
Pushing and opening pull requests requires git with a configured remote. If the repository has no remote, the skill still makes commits but skips every push, pull request, and CI-watch action. The source does not specify your CI or pull request tooling, so check that your project's shipping process works with an unattended push.
- git with a configured remote for pushing and pull requests
- The Compound Engineering child skills installed in the host
- CI and pull request tooling, if your project's shipping process uses them
Trigger lfg with a natural request
The skill is meant to run only when you explicitly ask for autonomous, end-to-end work or call lfg by name. A request that only describes a problem will not trigger it by design. Phrase the request so it clearly asks for hands-off completion, and give it a feature, a bug, an issue reference, or a path to a plan.
The argument hint for the skill accepts a feature, bug, issue reference, or plan path, and can optionally assign planning or implementation to a model or harness.
- Run lfg on the checkout bug: fix it and open a pull request, no check-ins.
- Use lfg to implement docs/plans/your-plan.md end to end.
- Invoke lfg for the issue reference in the tracker, and finish with a PR.
Use lfg only where pushing unattended is acceptable
lfg pushes and opens pull requests without waiting for approval. Do not use it in repositories where pushes or pull requests need human sign-off that the skill cannot see. If you want to review each step, use ce-plan, ce-work, ce-debug, or ce-commit-push-pr directly.
The skill runs no scripts, but it does use the network to push and open pull requests. It stops before any irreversible action you did not grant, such as a merge, a force-push, or deleting data. A stop leaves nothing pushed that was not already pushed.
- Runs scripts: no
- Uses network: yes, for pushes, pull requests and CI checks
- Stops for merges you did not grant, force-pushes, and data deletion
What people use it for
- Turning a well-specified plan file into an open pull request without stopping for check-ins.
- Taking a concrete bug report through reproduction, fix, commit and pull request.
- Running scheduled or nested agent workflows that cannot pause to ask questions.
- Producing a non-code result, such as an explanation, through the Compound Engineering skill that owns it.
Frequently asked questions
Is the lfg skill free, and what license is it under?
The skill is published under the MIT license. The source material on this page does not state pricing, so check the license and the Compound Engineering plugin source for any terms that apply to you.
How do I install the lfg agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
What does lfg require to run?
It requires git with a configured remote for pushes and pull requests, and the Compound Engineering child skills installed in the host. CI and pull request tooling are needed only if your project's shipping process uses them.
Does lfg merge my pull requests?
No. Merging stays with you unless you granted it for that run. The skill stops and reports before any merge you did not authorize.
Why did lfg not start when I described a bug?
The skill is meant to run only when you explicitly ask for autonomous end-to-end work or invoke lfg by name. For step-by-step review of a bug, use ce-debug directly.
What is the difference between lfg and ce-work?
ce-work implements a plan with you in the loop. lfg is an orchestrator that routes your request across several Compound Engineering skills and continues through review, commit, push and pull request without waiting for approval.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use lfg if you already trust the surrounding Compound Engineering skills and want an unattended run from a request to an open pull request. Choose ce-plan, ce-work, ce-debug or ce-commit-push-pr if you want to review each step, or if your repository needs human sign-off before pushes or pull requests.
Download the skill