Skills
Prompt improver skill
Turns vague asks into structured prompts before answering: goal, inputs, output shape.
Last reviewed July 17, 2026
A single-file skill template that makes agents restructure vague requests: state the goal, name missing inputs, and confirm output shape before the work.
SKILL.md
# Prompt improver Apply when a request is vague enough that two reasonable readings would produce different work. ## The check Before answering a vague request, restate it structured: - Goal: what done looks like, in one sentence. - Inputs: what you have been given, and what is missing that would change the answer. - Output shape: format, length, and audience you will assume. Then EITHER ask the single most valuable clarifying question (only one; more is stalling) OR proceed with stated assumptions, whichever is cheaper for the user to correct. ## Rules - Never ask about things you can decide reasonably; state assumptions instead. - If the request contains an example of the desired output, treat it as the spec. - For recurring vague patterns, propose saving the structured version as a reusable prompt. - Skip the whole ritual for simple factual questions; ceremony on trivial asks is worse than vagueness.
Set it up
- 1
Copy the skill body
Works as written; teams sometimes tighten the one-question rule to zero (assume-and-state only).
- 2
Create the skill
Add it in the workspace Skills page; every agent now restructures fuzzy asks the same way.
- 3
Harvest the structured prompts
When the same restructured ask recurs, save it via the prompt librarian agent so it becomes a team asset.