| name | onboard-contributor |
| description | Get a new contributor (human or agent) from zero to a running start_work.sh loop against this repo. |
Onboard Contributor
Paste this whole skill to a new contributor's agent, or read it yourself,
to get running against this repo's issue queue in a few minutes.
Which path applies to you?
| You have... | Do this |
|---|
| Write access + an agent CLI (Claude/Codex/Hermes) | Autopilot: run ./scripts/start_work.sh and let it claim, work, and open PRs for you. |
| Write access, no agent CLI | Manual: pick an issue labeled status: available, claim it by hand (see AGENTS.md step 2), do the work yourself, open a PR the same way. |
| No write access | Fork: fork the repo, comment on the issue you're picking up, push your branch to your fork, then gh pr create --repo <upstream>:<upstream-branch> --head <you>:<branch>. |
Autopilot quickstart
gh auth login
export AW_AGENT=claude
./scripts/start_work.sh
The loop will:
- Fetch open issues, prefer your own
status: changes-requested rework,
then freed rework, then fresh status: available work.
- Claim the issue, spin up a throwaway git worktree, run your agent CLI
against a generated prompt.
- Open a PR if the agent succeeded, or release the issue back to
available if it didn't.
You never need to touch labels yourself — the script does it. See
AGENTS.md for the full contract and docs/AUTOMATION.md for the status
lifecycle this loop drives.
Before you start
- Read this project's own README/CONTRIBUTING (if present) for
project-specific conventions — this skill only covers the generic
claim/work/PR loop, not what "good work" looks like here.
- Check for a
do-not-automate label on any issue before claiming it —
that's a human-only parking brake.
- If you're setting this workflow up for the first time in a repo, see
docs/GETTING_STARTED.md instead of this skill.