Make a repository ready for the lifecycle skills

create-issue, implement-issue, merge-pr and auto-dev are generic workflows wrapped around a thin layer of repo-specific facts. This skill does the whole setup story in one pass: it records those facts in a committed profile, converges the configuration they describe (labels, issue forms, settings, topics, Pages source) from a declarative manifest, and refreshes the profile so it reports what now exists. Decision record: ADR 0018.

The profile is data, not a skill (ADR 0001): the lifecycle skills read it at their Step 1, and only reach for this skill when it is missing.

Three paths

Pick the path from the request, not only from the flag. A request to read, show or regenerate the profile takes --profile-only; one to check or list drift takes --plan. apply runs only on a request to set up, configure or converge the repository, or on a bare /tagout:init — the writer never fires from a read-only question (ADR 0018).

Do this

Two bundled scripts do the deterministic work; run them from anywhere in the target repo (each anchors to the git root). <skill-dir> is this skill's base directory — given when the skill loads.

1. The profile

bash "<skill-dir>/scripts/repo-profile.sh" show

• It printed the profile (and no --refresh was asked) → keep it; relay the headline values (repo slug, commit identity, build/test commands, integration style).
• It printed NO_PROFILE (exit 3), or --refresh was asked → the generation path (on --plan, which writes nothing locally, say the profile is missing instead and go on to step 2). Read references/generating.md and follow it: run repo-profile.sh detect, fill references/profile-template.md from the facts it emits, write .claude/skills/repo-profile.md, and list every TODO you left.
• Exit 4 → not inside a git repository: say so and stop.

On --profile-only, stop here and recap.

2. The manifest

bash "<skill-dir>/scripts/repo-setup.sh" plan

plan writes nothing. On --plan, show the delta and stop. Otherwise, when it found drift, converge:

bash "<skill-dir>/scripts/repo-setup.sh" apply

The manifest is the repo's own .github/repo-setup.yml when it has one, else the kit's templates/repo-setup.yml; --manifest <path> overrides both. The rules apply follows (additive, pruneKeep, never clobber a form, placeholders reported !TODO, topics additive, Pages never disabled) are spelled out in references/desired-state.md — relay them before a first run against a repo that already has labels.

3. Refresh the profile

After an apply that followed a plan exit 1 (drift — apply itself exits 0 either way), re-run step 1 on its generation path: the label axes and issue forms the profile reported as missing exist now, so the profile must record them before a lifecycle skill reads it. This step is part of the run, not advice to the user.

Autonomy contract

Run hands-off. show and plan are reads — no ceremony. The generation path is best-effort inference: fill what detect proves and write a marked <!-- TODO: … --> for the rest rather than inventing a value. Before apply, get a yes only when the repo is not the user's own or the run includes --prune. Never invent a taxonomy: a missing axis is a manifest edit and a re-run, never a hand-made gh label create. A refused surface is reported and the run moves on. Stop only for a real blocker: not inside a git repo, or gh unauthenticated when the request needs GitHub (tell the user to run ! gh auth login -h github.com).

Inputs

• --plan — profile, then plan; nothing is written on GitHub.
• --profile-only — the profile alone; with --refresh, regenerate it even if one exists.
• --prune — passed to apply: delete labels the manifest does not declare (never the pruneKeep ones). Never implied.
• --manifest <path> — a manifest other than the repo's own or the kit's default.
• A path argument — a repo other than the current directory, passed as the scripts' [dir].

Recap

Close with the shared recap shape — ../_shared/recap.md. It owns the four blocks (verdict · What happened · Artifacts · Assumed · skipped · unverified, where None is a required answer rather than an omission) and the Next line, which is read off this skill's row in that file's hand-off table instead of being decided again here. Everything below is only what init adds on top of them.

• Say which path ran (/tagout:init, --plan, --profile-only) and whether the profile was read back or generated — they carry very different confidence.
• Name each GitHub surface separately — labels, issue forms, settings, topics, Pages — and whether it converged, was already converged, or was refused (and by what).
• Every <!-- TODO: … --> left in the profile goes in Assumed · skipped · unverified, by name, and so does whether --prune ran.