testbook-validatelisted
Install: claude install-skill QAIA-Project/QAIA
# testbook-validate — audit any test book
Follow the shared contract in `../README.md`. This is the **Validate intent**: it audits, it
never rewrites. It works on any `.feature` set — including books QAIA did not generate, which is
a first-class use case, not an afterthought.
## Steps
1. **Collect the pieces.** Use the inputs already designated, or ask for the `.feature` files (or
directory) and — if available — the source US/requirements and any coverage matrix.
The report's depth **adapts honestly to what exists**: without a source US,
business-correctness and coverage checks are marked `not assessable`, never guessed.
2. **Deterministic structural pass FIRST** — not an LLM impression. Compute the reproducible
structural score `/100` on every collected `.feature` file, with its three forced-FAIL
detectors (hollow AC, no expected result, fabrication sniffer) and the redundancy finding.
Full algorithm, how to run it, and the trap of running the sniffer blind:
`references/structural-pass.md`.
**This skill's whole point is auditing books QAIA did not write — so decide the mode first.**
A book QAIA generated carries at least one `@QAIA-*` tag; a foreign one carries none. On a
foreign book the structural pass must run in **third-party mode** (`--third-party` when using
the shipped scorer of `qaia-score`): the tag-based traceability budget is unwinnable by
construction, and the priority and technique tags are QAIA conventions, not quality