| name | deliver |
| description | Take one Laboratory backlog item to merge-ready. Composes /shape,
/implement, /code-review, /ci, /refactor, and /qa as needed. Stops before
merge. Trigger: /deliver.
|
| argument-hint | [backlog-item] |
/deliver
Deliver one shaped Laboratory work item to merge-ready state. This repository is
an experiment lab, not an app platform: delivery means the touched experiment,
harness, analysis, docs, or Spellbook harness surface is implemented,
reviewable, and verified by the repo's actual gate.
Inputs
- A
backlog.d/<id>.md item, experiment path, or explicit user request.
- Repo brief:
.spellbook/repo-brief.md.
- Gate:
make ci-smoke.
Loop
- Read the target item and the relevant repo anchors.
- Shape any missing oracle before editing. The oracle must name the behavior,
files, commands, and evidence that prove done.
- Implement narrowly. For code changes, preserve simulate mode and add focused
tests for new harness/analysis behavior.
- Run
/code-review for semantic issues.
- Run
/ci; it must execute make ci-smoke.
- Run
/refactor when the diff is broad or repeated logic appears.
- Run
/qa against the changed scientific surface:
- simulate run for harnesses
- fixture/committed-data analysis for analyzers
- provenance check for report or publication artifacts
- bridge/link audit for Spellbook harness changes
Loop back after any fix. Stop when review is clean, make ci-smoke passes, and
QA has inspected the relevant artifact.
Non-Goals
- Do not merge, push, or publish.
- Do not run live API experiments unless the target item explicitly requires a
bounded live smoke and the user has approved the spend.
- Do not archive backlog items without evidence in files or command output.
- Do not introduce platform scaffolding unrelated to the current experiment.
Output
Return a concise delivery brief:
- target item
- what changed
- commands run
- artifacts inspected
- residual risk
- next step for merge or follow-up
