goalrunlisted
Install: claude install-skill xtieume/testcase
# Goal Run — Plan It, Build It, Prove It
> You declare done. The user asks. You find more. Repeat.
The thing you checked was not the whole thing. **Violating the letter of the rules is
violating the spirit of the rules.**
## The rule
**Done is the script's exit code, not your judgement.** No "done" until `goalrun.py` exits 0
on a current whole-ledger `--verify`, with `last_proof.whole_ledger_verified: true` and
`proof_stale: false` in the selected run. Until then: "not yet" and the table. Git-ignored
files do not stale a proof, except `.env`, `.env.*` and the patterns in `.goalrun-include`.
No exceptions — not "done pending X"; not "the check is flaky"; not "I ran it by hand"; not
because the user said "just say yes"; not by editing a row until it turns green.
## Files in this skill
Read each one when the workflow calls for it, not upfront.
| File | Read when |
| ---- | --------- |
| `references/run-context.md` | Before you create, resume, hand off or take over a run. A status question needs only `inspect` and the rule above. |
| `references/ledger-design.md` | Writing or auditing ledger rows (step 4). |
| `references/pressure-test.md` | `SKILL.md` changed materially and the pressure test must be re-run. |
## Setup
From the repo root, `GOALRUN=<path to>/.agents/skills/goalrun/scripts/goalrun.py`.
Select an explicit run. `init` before edits;
`resume` as controller; save the top-level `token` as `TOKEN`. Existing work resumes its run
and keeps the original baseline.