Reviews with teeth
Cross-agent review, typed verdicts.
One keystroke spawns a second agent — even from the other vendor — beside the author. It reads the task's recorded history first, reviews the code where relevant, and posts a structured verdict. Not prose: a contract.
- ⌘R or right-click → Review with Claude Code / Codex — author and reviewer side by side under grouped tabs
- The reviewer is briefed from the board: brief, decisions, turn outcomes — then verifies the repo
- Typed verdict: approve or request_changes, a summary, and concrete findings
- request_changes findings become open review findings on the task — and the server refuses Done while any stay open
- /twd-address-review walks the author through fixing each finding and logs the outcome back
⌘R · Review DB-0142 with Codex
AUTHOR (claude) ····· REVIEWER (codex)
⏺ get_work_item · diff vs main · judging…
🧪 Verdict: request_changes — missing regression test
→ 2 findings opened on DB-0142
Board: cannot move to Done — 2 open findings
A second pair of eyes, always available
Author and reviewer, side by side.
Cross-vendor review is the point: a Claude author gets a Codex reviewer and vice-versa — different models, different blind spots. The grouped split keeps both agents (and the artifact under review) on one screen.
