Best for
- Getting more reliable answers from the same model
- Rewriting long or messy requests into a clean structure
- Choosing between zero-shot, few-shot, step-by-step and other techniques
What you give it
- The request or task you plan to send to your AI model
- Anything the answer must follow: format, length, audience
- Optionally: an example of a good answer
What you get back
- The technique chosen and a one-line reason
- A rough cost estimate: is the rebuilt prompt cheaper, the same or dearer?
- The rebuilt prompt, ready to use, with gaps marked [ask]
How it works
- Classifies the request: type, clarity, size, depth of reasoning, whether tools or files are needed.
- Uses the Prompt Frameworks skill (30 techniques with selection rules).
- Chooses the cheapest technique that is accurate enough; combines at most two.
- Rebuilds the prompt in that structure and marks gaps instead of inventing facts.
- A clear, common request gets no rewrite at all.
Example
You: Write release notes for the new login flow.
Result: Technique: CO-STAR (writing where tone and audience matter). Rebuilt: Context — v2.3 adds passkey login. Objective — announce it. Style — plain, short. Tone — friendly. Audience — existing customers. Response — 5 bullet points. Gaps: [ask] which browsers support passkeys?
Limits — please read
- The guidance is a rule of thumb, not a measurement: when two techniques are close, it says so.
- It does not do the task itself.
- It needs the Prompt Frameworks skill installed alongside it.