user-walkthroughlisted

# User walkthrough Attended, **operator-driven** UAT. You (the operator) drive; the agent rides along. Each turn you ask a question or point at a problem; the agent answers from PRIMARY SOURCE (reads the actual code/artifact, cites `file:line` — never vibes), marks any coverage the exchange earned, routes issues to fix-now or log-for-later, then **stops and hands the turn back**. Coverage accrues as a byproduct of poking the build, not as a script you march through. The ledger it reads and writes is defined once in [`.claude/references/shakedown-engine.md`](../../references/shakedown-engine.md) — this skill never redefines the row grammar, slug rule, or status vocabulary; it cites the engine. ## When to use / not - **Use:** a tool or feature was just built (post-`/build-phase`, post-`/build-step`, or a fresh `/user-debug`) and you want to poke it yourself — ask how it works, try things, surface issues — with the agent answering from source, fixing small breakage live, and tracking what got covered. You want to be in the driver's seat. - **Don't use** when you want the agent to close out UAT autonomously (that is `/user-shakedown`), to execute an already-written UAT block for you (`/user-uat`), or to refine a fuzzy UAT script (`/review-uat`). If you have no build to poke yet, there is nothing to walk through. ## Invocation ```text /user-walkthrough <tool/feature> # attended walkthrough of the named build /user-walkthrough export-button # slug resolve