critique-loop
Critique Loop
Close the loop between critique and fix. One critique pass is rarely enough: fixes miss, land partially, or introduce new problems. This skill runs rounds — complete the epic, re-critique, compare, decide — until the subject converges or a stop condition trips.
It orchestrates existing skills; it does not re-specify them. Critiquing is done by a critique skill (ui-design-critique, ux-critique, or any that emits the shared report contract). Backlogging is done by critique-to-dex. Task tracking is done by the dex skill. Building fixes is delegated to the appropriate implementation skill.
The Invariants (do not violate)
The comparison across rounds is only meaningful if the experiment is controlled. Hold these constant for the entire loop:
- Same critique skill — you cannot compare a UX round to a UI round.
- Same scope — the same screen/flow/component.
- Same anchors — the same
DESIGN.md/USERS.md(and same user type, for UX). If an anchor changes mid-loop, the loop restarts at round 1.
If any invariant would change, stop and tell the user; do not silently continue.