plan-reviewlisted

# Plan review — the second planner - **Can:** challenge every assumption in a plan written by another agent, and return a verdict. - **Must:** refuse if you wrote the plan, verify for yourself what you can, and declare what you did not check. - **Cannot:** rewrite the plan — it reports back to the author — and cannot write code. `plan` decides what to do. `codereview` judges finished code. **This sits between them**, and it is the cheapest review in the whole chain: a finding here costs a paragraph, the same finding after implementation costs the implementation. ## 🔴 Rule zero — you did not write this plan **Never run this skill on a plan you wrote.** The value here is entirely in the second pair of eyes; an author reviewing their own plan re-reads their own reasoning and finds it convincing, because it convinced them once already. | Plan written by | This skill is run by | |---|---| | Claude | Codex | | Codex | Claude | If you wrote the plan under review, stop and say so. See [`../../workflows/agent-roles.md`](../../workflows/agent-roles.md). **In `solo` mode** (`{{RECORDS_DIR}}/info.md`) there is no other agent. Reviewing your own plan is then permitted, on two conditions: do it as a **separate pass** after the plan is fully written, not while writing it — and **state in the verdict that it was a self-review**. A reader who does not know will assume a second agent signed off. 🔴 **Delegating any part of this?** Start it and then `*_await` — never poll `*_status` i