mustslisted

# Musts `musts` is the validation loop. The task is not done until `musts validate` exits clean. ## When to use this skill - After making code changes in a repo with a `MUSTS.yml` at the root. - Before declaring work complete, opening a PR, or asking for review. - Whenever the user asks "is it ready?" / "is this validated?". If `musts validate` prints "No MUSTS.yml files found.", this skill does not apply. ## The loop ```bash musts validate # → prints every pending task, exit code 1 if pending, 0 if clean ``` 1. Run `musts validate`. It emits **every** dirty task (no batching), so it is idempotent — re-running it never invalidates the ids it just issued. 2. Close each task: - **Runnable checks** (`do:` is a plain command — `cargo/*`, `bazel/build`): just `musts run <task-id>`. musts executes the command itself, checks the real exit code, and records the evidence for you. You never re-run the build to satisfy the loop. - **Judgment checks** (`agent`, `mav` — verify facts, drive a UI): do the work yourself, then `musts evidence <task-id>` (see below). 3. Re-run `musts validate`. 4. Repeat until exit code `0`. A task shown as `(unchanged since last validate …)` is one you already saw — its full body isn't reprinted; act on it the same way. ## Reading a task Each task in the report tells you: - **Task id** — passed to `musts run` or `musts evidence`. - **do** — exactly what to run. - **evidence** — what to attach when recording (judgment ch