auditlisted

# Audit a suite Input: `$ARGUMENTS`. A path, a stack name, or nothing for every stack in the config. ## Delegate this one Hand it to the `suite-auditor` agent. An audit reads the whole suite, the config, the profile, the CI history and the requirements, and almost none of that needs to come back into this conversation. What comes back is the report. Do it inline only when the user wants to look at one specific thing. ## What comes out A report at `qa/audit/<date>.md`, with nine rubric scores, ranked findings with evidence, coverage gaps against the requirements, and a suggested order for the first three fixes. The rubric is in `references/rubric.md`. It exists so two audits of the same suite land in the same place, and so a suite re-audited in six months can be compared against itself. ## The mechanical pass Before any reading: ```bash qa-context <platform> qa-policy --stack <platform> ``` `qa-policy` finds sleeps, positional locators, tests with no assertion, conditional assertions and shared state. Two of its five checks are heuristics and it labels them, so verify those rather than repeating them into the report. ## Rules **Evidence or it is not a finding.** A path and a line, or a command someone can run. "Tests are flaky" is not a finding. **Rank by cost over effort, not by severity.** A hundred two second sleeps beat one missing assertion, even though the missing assertion sounds worse. **Say what is good.** A report that only lists problems gets read as