Stop Me Init

stop-me-init

Write .stop-me.yml by reading the repo, never by interviewing the user. What the fields mean is defined in the stop-me skill (gate 1 overrides, gate 3 horizon) — this skill only decides their values for one repo. Guess every field, show each guess with the evidence behind it, ask once, write.

Guess horizon

Read before you guess: git log (age, contributor count via git log --format=%ae | sort -u), deploy files (Dockerfile, fly.toml, Procfile, compose, k8s manifests), CI workflows, publish config (package.json publishConfig/not-private, pyproject classifiers), schema files for tenancy columns.

• prototype — repo under ~30 days old, one deploy target or none, no CI
• platform — publishes a package, exposes a documented API, or carries tenancy columns
• product — everything else. When torn between two, pick product; the horizon only weights gate 3, and a wrong guess costs one line to fix.

Guess irreversible / reversible

Only list what the built-in surfaces get wrong for this repo:

• published SDK, proto, or wire-format directories the defaults would miss → irreversible
• local-only throwaway paths (dev migrations recreated on boot, scratch dirs) → reversible

Most repos need neither list. An empty guess is a real answer — say "the defaults already fit this repo" instead of padding the file to look thorough.

Confirm and write

Show the proposed YAML in one block, one line of evidence per field, then exactly one question: write it as-is, or with which edits? Write .stop-me.yml to the repo root only after the answer. Never write DECISIONS.md — stop-me creates that on its first objection.