Prompt Enhance

Return one refined prompt a coding or knowledge agent can execute. Do not interview. When information is missing, use an assumption or [NEEDS: ...].

Workflow

1. Read the input and referenced context: repo, files, AGENTS.md, CLAUDE.md, README.md, relevant code.
2. Resolve unambiguous references such as "the thing" or "that script".
3. Identify defects: vague goal, buried intent, contradictions, missing scope, missing output format, missing success criteria, negative phrasing, premature-stop scaffolding, irrelevant noise.
4. Rewrite the prompt. Preserve intent. Remove filler. Keep substance.
5. Return the refined prompt, then Changed, Assumptions, and Gaps lines as needed.
6. Offer one action: run it now, save it to prompts/<slug>.md, or stop.

Principles