setup-agency-skills

Write .agents/config.yml so the other agency skills resolve your tracker and move issues into states that actually exist on your board. Run once per project. Re-run to revise: read the existing config first and edit it, rather than starting over.

Detect and derive first; ask the user only for what the environment can't reveal. Every state you record must be detected from the board, never invented: the runtime skills only set a state that already exists, so a value the board doesn't have silently no-ops.

Resolve the tracker

Resolve the tracker and the user's identity per references/issue-tracker.md. This names the host and confirms you can reach the board.

• Remote names the tracker (GitHub, GitLab) → leave tracker and project out of the config; the skills read them from git remote every run. Writing them caches a lookup that can go stale.
• Remote can't name it (Jira, Linear, self-hosted) → ask the user for tracker and project, and write both.

Read the board

Fetch the workflow states the board actually uses: the status field's values, or the workflow labels when the tracker models state as labels. This live set is the point of running setup. It's what a static example can't know, and what every state you write must match.

Map the states

Fill three keys from the real board states, offering the closest match for each and confirming it with the user: