precedent-analyzelisted

# Backfill the graph from an existing project The decisions were made already — they are sitting in the manifests, the CI config and the README, and nobody wrote them down anywhere they can be queried from. `<cli>` below is this package's `precedent.py`; the `precedent` skill says how to locate it. Work in this order. Do not skip to recording. **1. See what is already known.** ```bash uv run <cli> brief --project <path> ``` Everything under "decided here" is already recorded; re-recording it creates a second decision saying the same thing with a different id, and `maintain` will report the pair as a contradiction. Note the tags too — an untagged project's decisions surface nowhere, so tag it (the `precedent-tag` skill) before recording anything. **2. Read the evidence.** Not just the file listing — open them: - build manifests and lockfiles (what was picked, and what version policy) - CI and release workflows (where it builds, what gates a merge) - `Dockerfile`, compose files, deployment and infra config (where it runs) - `docs/adr/`, `decisions/`, `ARCHITECTURE.md` if they exist — these are decisions already written as decisions, with their rationale intact - `README.md` and `CONTRIBUTING.md`, especially any "why" sections - `git log --oneline | grep -iE 'switch|migrat|replac|drop|move (to|off)'` — a migration commit is a decision with a date and often a reason **3. Keep only what is worth quoting back.** The test for each candidate is: *if I started a comparable