Use when authoring or reviewing agent prompts that need gated execution over repos, local files, runners, or pipelines. Establishes audience separation, executable gates (discovery, staging, rollback, validation), and strict claim discipline (no PASS without evidence).
