feature-healthlisted
Install: claude install-skill brianwestphal/hotsheet
Rebuild [`docs/feature-health.md`](../../../docs/feature-health.md) from current evidence.
## The question this answers
**Does each feature actually work?** — NOT "was it built?"
`docs/ai/requirements-summary.md` already tracks built-ness (Shipped / Partial / Design only). This
skill tracks **trustworthiness**. A feature can be 100% Shipped there and `Unknown` here (built,
never realistically exercised). That gap is the entire point.
**Enhancements are out of scope.** Every feature could do more; saying so is noise. A note earns its
place ONLY if it records:
- missing **core** functionality (the feature can't do its stated job),
- a **known bug** or a burst of fix commits,
- an **unsettled decision** the feature is resting on,
- **absent verification** (no automated coverage AND no evidence of real use).
A feature with nothing wrong gets **no note**. Resist the urge to fill every cell.
## Method
Work from evidence, in this order of weight. Don't skip to conclusions from the docs — the docs have
been wrong in both directions before.
1. **Enumerate the features.** Use `CLAUDE.md`'s requirements-doc list (§1–§N) as the spine, then
merge doc numbers into coherent user-facing features. Also sweep for features with no doc
(check `src/client/` modules and the settings dialog tabs). Keep the existing table's grouping
unless the product has genuinely changed shape.
2. **Map automated coverage.** `ls e2e/*.spec.ts` and `find src plugins -name '*.test.ts'`. For each