profile-repo and setup-repo stay a reader and a writer

Context and Problem Statement

profile-repo reads a repository’s facts into the committed profile; setup-repo writes labels, issue forms and settings from a manifest. They are two front doors to one story — “make this repo ready for the lifecycle skills” — and the kit’s own trigger bench once sent “configure this repo for the issue skills” to the half that cannot create a label (#279). On 2026-09-02 the owner asked whether the two should be merged. (Recorded from that question; the answer below is a proposal for the owner, not a decision taken.)

Considered Options

  • Merge them into one skill, configure-repo, with three verbs — detect · plan · apply.
  • Keep two skills and sharpen the boundary: the eval sets’ near-miss negatives, the hand-off from profile-repo’s recap to /setup-repo when an axis is missing, and the description sentence “It WRITES what profile-repo only READS.”
  • Keep two skills and add one command, /configure-repo, that runs profile-repo then setup-repo plan in sequence.

Decision Outcome

Keep two skills, for the current major: one is a reader that runs anywhere without rights and degrades to TODOs, the other a writer that needs an admin token, refuses by name and is idempotent — the same report/decide split the kit applies everywhere (survey.sh reports, the supervisor decides), and a merge would put the writer inside the reader’s trigger surface. A rename or merge is a breaking change under ADR 0012, and 2.0 has just shipped one. If the boundary keeps confusing users after the sharpened evals, the merge into configure-repo with three verbs is the shape to take, in the next major.

Consequences

Two front doors remain, so the which one? question stays answerable only by the descriptions and the README’s table; tests/skills/test.sh’s eval structure and the implement-issue ↔ merge-pr style boundary bench are what keep it answerable. Reopens when: the trigger bench measures setup-repo’s near-miss negatives firing profile-repo (or the reverse) above the 0.5 threshold after a description edit, or when a 3.0 is planned for another breaking reason.