factory-verifylisted

# Verify the factory Output: a report, and nothing else. You change no stage, no gate and no project file — a check that repairs what it measures has measured nothing. The pipeline exists so that an agent's word is never the evidence. That applies to the pipeline itself, so **you do not judge whether it worked: two scripts do.** Your job is to run the right one, read what it says, and add only the parts a script cannot reach. Which mode: | The question | Mode | |---|---| | "a story just ran — did it really do what the files say?" | **1. Observe a real run** | | "can I trust this pipeline here at all?" / a stage, a gate check or the runner changed | **2. Check the machinery** | ## 1. Observe a real run ```bash python3 <this skill>/scripts/observe.py --story <id> # add --json for a machine-readable form ``` It reads the story, the six hand-over files, the gate reports and snapshots the runner journalled, and the repository itself — then reports **what did not hold**, what held, and what it could not see. It cross-checks claims against evidence, which is the whole point: - the **file contract** — every stage's file exists and carries the sections it owes; - **criterion keys** — the story's keys and the test table's, verbatim, in both directions: a renamed key silently drops a criterion; - **red then green** — from the test stage's own record and the gate's reports, never from a stage's sentence about its tests; - **no test was edited later** — the mapped test