capturelisted

If the project has no knowledge system yet, run `bootstrap` first — this skill maintains what that one sets up. For whole-knowledge-base health, independent of any one session, run `lint` instead — this skill only reasons about what just happened. ## Artifact model This skill's sibling `../../shared/artifact-model.md` defines the four artifact types, their fields, the derivation recipe, and the economy rule for anything drafted here — read it before drafting anything. ## Checks, one pass per invocation 1. **New decision?** Draft a stub: next id, title, one-line motivation, `track: product | process`, `status: draft`. Don't finalize without the owning domain's sign-off. If it explicitly replaces an existing decision, also update that decision's `status` to `superseded` and add `superseded-by: <new-decision-id>` — don't leave two `active` decisions on the same topic for a later `lint` pass to catch. 2. **Fact changed?** Update the fact file and its `last-verified`. If the governing decision is no longer current, surface the contradiction now. 3. **New automatable rule?** Log as debt in the fitness-function inventory: rule text, governing decision, why not automated yet. 4. **Contradiction found?** Block. Don't close the session until resolved or explicitly deferred with a written note (decision, fact, or stub). 5. **Human-doc drift?** If watched paths overlap what changed this session, propose the specific update; if the doc's still accurate, bump its verified date. 6. **Ro