cross-review

# cross-review — independent verification The implementer never signs off on its own work — a different model does, and review is a sub-agent that returns a structured report, not a transcript anyone needs to read through. ## Procedure 1. Reuse the supplied diff file, or save the task's diff directly with a shell redirect (`gh pr diff <pr> > <absolute shared diff path>` or `git -C .worktrees/<task_id> diff <base-sha>...<head-sha> > <absolute shared diff path>`). Use a unique snapshot outside the implementer's worktree, record its base and head commit SHAs, and leave it unchanged until review finishes. The reviewer must be able to read that path. For a different host or sandbox, transfer the snapshot first or arrange access to both commits so it can compute the diff; never regenerate the diff through model output. 2. Run the deterministic gates first — tests / lint / typecheck via `sys_os_shell`. If red, re-dispatch the implementer to drive it green first; don't involve the reviewer yet. If a pytest result's count must be recorded or reconciled, collect ground truth with `python -m pytest --collect-only -q <same files>` against the exact file set/command/commit the implementer reported. Never use `grep -c 'def test_'` as a pytest count: it counts functions, not collected cases, and misses parametrized case expansion. 3. Dispatch a DIFFERENT-vendor sub-agent as reviewer: pick any AVAILABLE worker whose vendor differs from the imple...