Announce: "Using audit-fix-loop to plan a scored iterative improvement loop."

What this skill is (and isn't)

/goal is the mechanism: it re-fires turns and uses a separate-model evaluator to gate exit on a condition it reads from the transcript. This skill is the doctrine that tells you how to set a loop up so /goal enforces the right thing — /goal will just as happily enforce a wrong condition.

It is also the generic fallback. The domain workflows below are specialized, pre-wired instances of this same doctrine — prefer them when one fits. Use this skill directly only for ad-hoc "improve " requests that don't map to any of them:

Domain workflow	Use it instead when…
visual-verify	improving slides/charts/rendered output (Gemini vision, zero-blocking-defects gate)
bluebook-audit	correcting Bluebook citations in a DOCX
source-verify	verifying citations exist + quotes match sources
writing-review / writing-revise	improving a prose draft
workflow-creator Mode 3	hardening a skill/workflow (wc-audit substrate gate)