Use to run an interactive human↔agent QA pass on mod-bot ("let's QA"). The agent monitors logs/DB and drives multi-action verification steps; the human performs only the Discord-client actions that require a human. Produces a findings-only QA report; shown to the human if testing a local branch, posted to a PR if available.
Use when QA-testing mod-bot during active local development against a dev bot you start and drive yourself — typically a feature branch in a git worktree, before there's an RC. Covers standing up the dev environment, deciding per finding whether to fix-now/log/defer, and authoring or re-validating checklists. For release/RC verification against a deployed pod, use qa-session instead.