review-planlisted

# Review Plan The depth pre-pass for `/review-fix`. It answers one question — **how deep should this particular review go, and which lenses does this diff actually need** — and returns a small structured verdict. It does not review anything. Today every review runs at effort `high` with the full finder roster, whatever the input. That is right for a 900-line new subsystem and wrong for a one-line `/fix-checks` repair; it is also wrong in the other direction for a migration or a credential change, where `high` is not deep enough. This skill makes depth a per-input decision inside a band the author fixed. ## How it is invoked `/review-fix` Step 1a launches it as a **subagent with `model: opus` set explicitly on the Agent call**. The pin is not optional and not inherited: a nested run does not inherit the launching session's model, so omitting it silently accepts a default and leaves `strategy-token-economy` clarification 46's cost measurement uninterpretable — the same argument that pins `dispatch-code-review` to `--model opus`. It runs **before** both the detached `/code-review` pre-stage (Step 1b) and the Workflow fan-out (Step 2), because its verdict sets the effort the first passes and the roster the second uses. ## Output contract — a small structured verdict, never prose Return exactly this object and nothing else. The caller parses it; a prose answer is an unparseable verdict and takes the fail-open path. ```json { "effort": "low|medium|high|xhigh|max", "irre