review-uatlisted

# Review UAT Take a draft UAT script (manual smoke list, M-step in a plan, "things to test before ship" doc) and produce a tightened version where: 1. Every step is **unambiguous** — one valid reading, not three. 2. Every step has **explicit prerequisites** (what state the system must be in first). 3. Every step has **explicit pass criteria** (what the operator looks for, in observable terms). 4. Only the steps that **genuinely need a human** remain on the human's plate. Setup, teardown, and anything verifiable from logs / DB / HTTP gets split out so an agent can do it. Iteration goal: a future operator (or a fresh Claude session) can execute the UAT cold, without asking clarifying questions or guessing. --- ## Invocation ``` /review-uat # ask user for the draft /review-uat path/to/plan.md#step-21 # refine the M-step at this anchor /review-uat --apply # after refinement, write the result back to the source doc /review-uat --exec # after refinement, delegate execution of the refined script to /user-uat /review-uat --exec --dry-run # delegate with --dry-run: /user-uat prints what WOULD run; nothing executes ``` `--apply` and `--exec` are opt-in. Default behavior is read-only describe-only: produce the refined version in chat, do NOT modify the source and do NOT run anything. --- ## When to use - A plan has a "Manual M*" step or "bundled UI smoke" handoff that reads as a terse bullet list. - The user