setup

Write factory.toml once per repository so the other skills do not ask the same questions on every run.

What it does

/factory:setup configures a repository for the factory skills by writing factory.toml: the spec directory, the base branch, the issue tracker, and the verify commands. It detects the stack and remote host first and asks only what it cannot detect.

Only the user may start this skill. If the model reaches it on its own, it stops and asks before writing anything.

Reads

• Stack marker files: pnpm-lock.yaml, package.json, pyproject.toml, uv.lock, Cargo.toml, go.mod.
• The remote host from git remote -v and the default branch from git symbolic-ref refs/remotes/origin/HEAD.
• lawbook.yaml, an existing coverage file, tsconfig.json.
• An existing factory.toml, which it updates in place and keeps user edits in.

Steps

1. Detect everything in the list above. Never ask for it.
2. Report the detected values in a few lines.
3. Ask, one question at a time, only for what detection left open: the tracker kind when the remote is ambiguous or local files or Jira might be preferred; the Jira site, project key, and work item type; a test command when no stack marker matched. Recommend an answer with every question.
4. Check the tracker tool is usable with gh auth status, glab auth status, or acli jira auth status. Report a failure; never run a login.
5. Show the draft factory.toml. Write it after the user accepts, with default-valued keys omitted.
6. Add .verify/ and .factory/ to .gitignore if they are not already ignored.

Done when

factory.toml exists, every key in it was either detected or chosen by the user, and the reply lists the verify commands the stack default will run. See factory.toml for the schema.

Upstream

Wraps the explore-then-confirm pattern from mattpocock-skills/setup. Upstream writes docs/agents/*.md and a CLAUDE.md block; factory writes factory.toml only.