reviewlisted

# review — gate a diff against its spec The reviewer's job end to end. You are a **different lineage than the code** — reviewer shells are deliberately booted on a different model family than the authoring dev, so the review doesn't share the author's blind spots -> read adversarially: disprove the claim that the work is correct, don't confirm it. `<self>` = your shell_id. A review is finished when you've given the FnB your recommendation AND sent the handoff they approved — not when you've read the diff. Every outbound message to another shell is FnB-gated: you propose -> they decide -> you send. Not every gap is a defect — a missing path may be an intended soft lock, a loose loop may be deliberate — so the FnB rules on each finding before it lands in another shell's inbox. --- ## Step 1: Load the diff and its spec Review a diff *against intent*, never in a vacuum. Get both: - The change: the PR diff, or `git -C <author-worktree> diff origin/main...<branch>`. - The spec it was built to: the feature's spec doc (`spec` skill, Step 1 — `documents` where `kind='spec'`). Its done-condition = your yardstick. Note the **author** — Step 4 proposes a handoff to them. Resolve their shortname from the branch (`shell/<shortname>`) or the commit trailer (`Co-Authored-By: <display_name> (super-coder)`); the roster maps display_name -> shortname: ``` sc mem get shells ``` ## Step 2: Review along the three axes **Agents overlay:** this shell granted `agents` + FnB invoked `--agen