Ticket pipeline
Run a ticket from discovery through a draft pull request. safety-overrides beat every step.
Put evidence in the PR body under the headings this skill names. Then run the status script. Do not argue in prose about which steps are done:
python <this-skill>/scripts/pipeline_status.py --pr-body BODY.md
Re-run until the script prints missing: 0.
Sequence
- how — build a mental model before changing code. Heading:
## How. - writing-plans — plan in the pull request body. Heading:
## Plan. - verify-loop — define the deterministic check, run it red (
VERIFY_FAIL), then iterate. - test-driven-development — if the check is a test: failing test, then minimal code. Heading:
## TDD(includered-green). - verification-before-completion — paste fresh
verify-looprunner output. Heading:## Evidence. - code-review — Standards and Spec axes on the branch diff. Heading:
## Code review. - interrogate — required for one-way doors (money movement, schema or data migrations, auth, data deletion, infra or prod config, licences). Heading:
## Interrogate, or writeN/A (two-way door). - pr — draft pull request body. Run
pr/scripts/check_pr_body.py. Ticket links areRefs #Nonly.
Hard rules (also in safety-overrides)
- Draft PRs. Never mark ready. Merge only as
trust_levelin safety-overrides allows. - Push only this branch. Rewrite with
--force-with-leaseonly. - No production credentials, live infrastructure, runtime package installs, or secret-using CI jobs.
- No tracker or chat writes from a ticket.
- Skip a blocking approval gate only with
Ruling: what / whyin the PR body.
Stacked pull requests
When one ticket is too large for a single reviewable PR, split it into layers:
- One concern per layer. Dependencies belong in the same layer or a lower one.
- Open the next layer while the lower one is still in review.
- Merge bottom-up. When merging through the API, use the asynchronous merge endpoint for stacks.
- Prefer
gh stackfor cascading rebases. If stacks are unavailable, chain pull requests with plain base-branch targeting.