Elves skill
A skill that hands planned work to a separate worker: the driver plans and reviews, the worker writes code while you sleep.
- 225 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The worker writes code and commits to the repository autonomously while you are not watching.
- The skill opens draft pull requests and works with review bots.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/elvesDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The worker writes code and commits to the repository autonomously while you are not watching.
- The skill opens draft pull requests and works with review bots.
This is third-party code. Review the repository files before installing.
What it does
Elves separates roles. A capable driver (Claude Code, Codex, Grok Build or Oh My Pi) plans and reviews the work, a subscription-native or external worker implements it, and run state files survive context compaction. You write the plan and decide on the merge: the agent takes the middle. For implementation Elves opens a draft pull request at the first useful commit, checks whether configured bots review drafts and passes their feedback to the driver at safe boundaries. The worker gets no pull request authority and the final independent review stays. Read-only audits open no PRs. The Claude Code install command clones the repository and runs `scripts/sync_installed_skills.py --apply --target claude`, which installs `~/.claude/skills/elves/` and managed aliases such as `/cobbler`, `/council`, `/grok` and `/devin`. Codex, Grok Build and Oh My Pi have their own `--target` values, without aliases.
Who it is for. Developers with a written plan who want to hand implementation to a worker and get a draft pull request by morning.
Good fit when
- A multi-batch plan is ready and you want another agent to write the code.
- An overnight run must leave a draft PR with bot review checked.
- Work must not be tied to one model provider.
- Research or an audit runs without writing changes to the repository.
Not a fit when
- There is no plan: the skill executes it and does not invent one for you.
- You need an instant answer in one session without a separate worker.
- The merge decision cannot go unreviewed by you.
Example request
Run Elves on the plan in docs/plan.md: the worker implements the batches, the driver reviews, and the result goes into a draft pull request.Limitations
Elves is built for a prewritten plan, and result quality depends on it. The install through `sync_installed_skills.py` writes to the agent skills directory and creates managed aliases, and it does not overwrite user aliases of the same name but reports a conflict. The worker runs on your subscription or an external key, and long runs spend limits. The merge stays with a person. The license is MIT, the last commit dates from October 2026.
How to disable. Delete `~/.claude/skills/elves/` and the managed aliases (`cobbler`, `council`, `grok`, `devin` and others) from `~/.claude/skills/`, then restart the agent.
FAQ
How do I install Elves in Claude Code?
Clone the repository into a temporary folder and run `python3 scripts/sync_installed_skills.py --apply --target claude`. The skill appears in `~/.claude/skills/elves/`.
Which agents work as a driver?
Claude Code, Codex, Grok Build and Oh My Pi. Each has its own `--target` value.
Do aliases work in Codex?
No, Codex gets only the main bundle, and the skill is invoked through `$elves` or a plain phrase.
What are the run files?
Durable run files: through them the work survives context compaction and continues after a pause.
Does the worker open a pull request?
No, the worker has no PR authority. The driver opens the draft PR at the first useful commit.
Discussion
No questions or reviews yet.
Related
SkillsOne agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.
A process kit for agents: a skill router, project memory in a DEV folder and result verification before saying done.
A set of skills that hand a task to a separate CLI agent while you review the diff and land the commit yourself
A meta-skill that folds rules scattered across AGENTS.md, CLAUDE.md and Cursor files into one project skill with task routes.
A CLI that detects your project's stack and installs matching agent skills from a vetted registry in one command
A harness around a coding agent: supervisor, reviewer, adversary and local log distiller, each on its own model
README in short
The Elves skill for Claude Code, Codex, Grok Build and Oh My Pi: a driver plans, a worker implements the plan in batches, and the result goes into a draft pull request.