Unslop Copywriting

AI models write public UI text in two failure modes: marketing slop (empty empowerment verbs, interchangeable SaaS claims) and boundary failure (instructions, criteria, schema, and process notes rendered as if they were product copy). This skill bans both and rewrites until every user-visible string reads as plain language for a human completing a task.

Core Mandate

1. Public UI is for humans. Never ship copy that sounds like a system prompt, ticket, schema, stack trace, or implementation note.
2. Ban first, then rewrite. Do not polish slop in place. Strip banned phrases and leaked internals, then write specific plain language.
3. Prefer delete. If a string adds no task value, remove it. Models always produce text; humans often should not.
4. Outcome over fluff. Name the concrete action, destination, or result. Vague verbs and ghost CTAs are failures.
5. Pivot rule. If a surface hits 3+ banned patterns, rewrite the whole message set for that surface — do not synonym-swap.

Hard Bans (never ship)