| name | flywheel |
| description | Outer loop for Laboratory research work. Picks a backlog item, shapes it,
delivers it, settles it, ships it through PR, monitors evidence, and reflects.
Trigger: /flywheel.
|
| argument-hint | [--max-cycles N] |
/flywheel
Run bounded Laboratory research cycles. This repository ships experiments,
analysis, reports, and harness improvements; it does not deploy an app by
default.
Cycle
- Pick the next ready item from
backlog.d/.
/shape if the item lacks a falsifiable oracle./deliver the item to merge-ready./yeet only when the user requested commits/pushes./settle until review, make ci-smoke, refactor, and QA are clean./ship through the repo's branch/PR workflow when the user asked to land./monitor the relevant evidence surface after shipping or a run./reflect and capture any follow-up backlog or harness changes.
Laboratory Invariants
- Scientific workflow is the product: hypothesis, harness, data, analysis,
traces, and deliverables.
- Simulated data is QA evidence only. It never supports publishable claims.
- Live runs require explicit
--live, provider preflight, budget cap, and
trace artifacts.
make ci-smoke is the default gate unless a touched surface names a narrower
additional command.- Spellbook owns workflow primitives. Laboratory owns experiment semantics.
Non-Goals
- No semantic workflow engine or hidden scheduler.
- No deployment phase unless the user points at an actual site/service change.
- No live API spend merely because a cycle reached an experiment item.
- No backlog archival without file or command evidence.
Output
For each cycle, report the item, phase verdicts, commands run, artifacts
inspected, residual risk, and the next concrete step.
