Critique

Use this when the next risk is not implementation detail alone, but locking the wrong change or carrying the wrong fear into the next slice.

critique is the structured before-the-fact counterpart to retro. The substrate is the same across targets — pre-lock-in stress with anchor- discriminated multi-angle review followed by one counterweight pass — and the target reference shapes the angle distribution and output. Decision pre-mortem (Klein lineage) is one of those targets; code/PR critique, release critique, rename critique, and spec critique reuse the same substrate.

Task-completing repo work always records critique before closeout. Scale the pass, not the obligation: use the risk boundary or meaningful slice as the review unit, not every commit. See references/cadence.md.

When this standalone critique skill runs, it always means a fresh bounded subagent review. bounded limits scope and time box, not execution mode. There is no same-agent or local standalone critique variant.