Setup

Setup

Scaffold the per-repo configuration that the engineering skills assume:

• Markdown tracker: where maps, specs, and tickets live as files under .scratch/
• Triage status: the status: frontmatter strings used for the five canonical triage roles
• Domain docs: where CONTEXT.md and ADRs live, and the consumer rules for reading them

This is a prompt-driven skill, not a deterministic script. Explore, present what you found, confirm with the user, then write.

Process

1. Explore

Read whatever exists; don't assume:

• AGENTS.md / CLAUDE.md at the repo root: does an ## Agent skills section already exist?
• CONTEXT.md at the repo root, docs/adr/
• .agents/: does this skill's prior output already exist (domain.md, issue-tracker.md, triage-labels.md)? These sit next to .agents/skills/; they are not skills.
• docs/agents/: leftover from an older setup. If those files exist and .agents/ does not, move them.
• .scratch/: is the convention already in use, and at which levels?
• Is the triage skill installed? This decides whether Section B runs at all.

2. Present findings and ask

Summarise what's present and missing, then take the sections in order; one section, one answer. Lead each with the recommended answer so the user can accept it in a word. Use AskUserQuestion where available; otherwise ask in chat and wait.

Section A: Issue tracker. Markdown files under .scratch/ are the only tracker these skills use: it needs no auth and keeps every map, spec, and ticket greppable next to the code. There is no choice to make here, so ask nothing: write .agents/issue-tracker.md from this skill's issue-tracker.md seed. Ask one question only if the repo already keeps this work somewhere other than .scratch/, and then only to confirm that path.

Section B: Triage status vocabulary. Skip entirely if the triage skill isn't installed. If it is, ask exactly one question: keep the default status strings? (recommended: yes). Defaults: needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix. Collect overrides only on "no".

Section C (Domain docs. Default to single-context) one CONTEXT.md + docs/adr/ at the repo root. Write it without asking. Offer multi-context (a root CONTEXT-MAP.md pointing at per-module CONTEXT.md files) only when exploration found real module boundaries (app-modules/, modules/, a monorepo layout).

3. Confirm and write

Show a draft of the ## Agent skills block and each .agents/*.md file; let the user edit before writing. If docs/agents/ still holds a prior copy, move those files into .agents/ instead of writing a second set.

Edit CLAUDE.md if it exists, else AGENTS.md; if neither exists, ask which to create. If an ## Agent skills block already exists, update it in place, never append a duplicate.

## Agent skills

### Issue tracker

[one-line summary]. See `.agents/issue-tracker.md`.

### Triage status

[one-line summary]. See `.agents/triage-labels.md`.

### Domain docs

[one-line summary: "single-context" or "multi-context"]. See `.agents/domain.md`.

Omit the ### Triage status sub-block (and its file) when Section B didn't run.

Seed templates in this folder:

• issue-tracker.md: the markdown layout for maps, specs, and tickets
• triage-labels.md: status role mapping (only if triage is installed)
• domain.md, domain doc consumer rules + layout

4. Done

Tell the user setup is complete and which skills now read these files (wayfinder, to-spec, to-tickets, implement, code-review, triage). They can edit .agents/*.md directly later: re-running this skill is only needed to restart.