Owns copy.yaml, the structured content payload that design consumes. The same payload must work with any visual identity, so this skill carries words, not design decisions. Authoring operations change copy; judging operations report on it.
Triggers
Author — produce or change copy:
- write ("write the headline", "new copy from this brief", "we need a value proposition") → write.md
- extract ("structure this page", "pull the copy from this URL", "turn this brief into copy.yaml") → extract.md
- refresh ("tighten this", "the copy reads weak", "polish before handoff") → refresh.md
- revoice ("make it playful", "make it sound premium", "drier, less salesy") → revoice.md
- reconcile ("sync copy from code", "the implementation drifted") → reconcile.md
Judge — a non-mutating verdict on existing copy:
- critique ("does this read as slop", "score the copy", "verdict before more editing") → critique.md
- audit ("is this copy ready to ship", "defect report before handoff") → audit.md
Classify the request by what it wants done to the copy, and infer from source and intent rather than asking. "Before handoff" matches two operations: a judging request with no implementation source is audit, while a sync request naming code or a live URL as the source of truth is reconcile.
Workflow
trigger → discovery (context, intent, register) → operation → copy.yaml or verdict
|
critique → refresh → critique again
Every operation starts by loading discovery. A judging verdict is applied by running the matching authoring operation, never by patching from the judgment itself.