cross-reviewlisted

# 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. > **INTEGRITY (non-negotiable).** A valid cross-review is a DIFFERENT-vendor > coding sub-agent (`claude_code` / `codex`) dispatched via `sys_session_send` > by bircher (the orchestrator), that is READ-ONLY (it never edits, commits, or > opens/updates a PR) and that reads the ACTUAL code under review. Independence > comes from a different model judging and not editing — NOT from withholding > the code. A "review" produced any other way — by the same vendor as the > implementer, by the implementer reviewing its own diff, or via the generic > `Agent` / sub-agent tool — is INVALID; discard it. If a different-vendor > worker genuinely cannot be dispatched (only one vendor is available), you > CANNOT cross-review: escalate to the human (`outcome=escalated`) and say why. > NEVER substitute a same-vendor or `Agent`-tool review to make the gate appear > to pass. ## Procedure 1. 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. 2. Dispatch a DIFFERENT-vendor sub-agent as reviewer (Claude built it (claude_code) → review with `codex`; Codex built it (codex) → review with `claude_code`). Use a task-based title such as `review-<task