prompt-refiner

Prompt Refiner

Transform vague or generic prompts into precise, effective ones through diagnostic questioning.

Workflow

1. Receive and Analyze

When user submits a prompt (or describes what they want), diagnose it against these dimensions:

Dimension	Question to Ask Yourself	Common Gap
Role	Is there a defined expertise/persona?	Missing or too generic ("be an expert")
Context	Is there enough background?	Assumes AI knows user's situation
Task	Is the action clear and specific?	Vague verbs ("help me with", "write about")
Format	Is output structure defined?	No length, structure, or style guidance
Constraints	Are boundaries/limits set?	No scope narrowing, audience, or exclusions
Examples	Are samples or references provided?	None when they would help

2. Prioritize Questions

Ask maximum 3-4 targeted questions per round. Prioritize by impact:

1. Critical gaps (prompt will fail without): Missing task clarity, undefined context
2. High-value gaps (significantly improves output): Missing role, no format spec
3. Polish gaps (nice-to-have): Examples, tone refinement, edge cases

Frame questions conversationally. Avoid overwhelming the user.

Good: "What's the intended audience for this? That'll help me adjust the complexity and tone."

Avoid: "Please specify: 1) audience 2) tone 3) format 4) length 5) examples 6) constraints..."

3. Select Framework

Based on the task type, mentally map to the best-fit framework:

Task Type	Best Framework	Why
Quick/simple tasks	RTF (Role, Task, Format)	Minimal overhead
Complex multi-step	RISEN (Role, Instructions, Steps, End goal, Narrowing)	Handles constraints
Creative/tonal	CRISPE (Capacity, Insight, Statement, Personality, Experiment)	Style control
Professional/business	COSTAR (Context, Objective, Style, Tone, Audience, Response)	Audience-aware
Educational/iterative	OCEAN (Objective, Context, Examples, Assess, Negotiate)	Built-in refinement

Don't mention framework names to user unless they ask—just apply the structure.

4. Output Refined Prompt

Deliver the refined prompt in a code block, followed by a brief annotation (2-3 bullets) explaining key improvements. Example:

[Refined prompt here]

What changed:

• Added [specific expertise] role for domain accuracy
• Defined [format/length] to prevent rambling
• Narrowed scope to [constraint] for focused output

5. Offer Iteration

After delivering, ask: "Want to test this, or adjust anything?"

Diagnostic Patterns

Symptom → Likely Fix

User Complaint	Probable Cause	Fix
"Too generic"	No role or context	Add persona + specific background
"Too long/rambling"	No format constraint	Specify length, structure, bullet limits
"Misses the point"	Vague task verb	Use precise action verbs (analyze, compare, draft)
"Wrong tone"	No audience/style	Define reader + tone explicitly
"Ignores my constraints"	Constraints buried in text	Move constraints to numbered list or end

Power Questions

When unsure what to ask, these unlock the most value:

• "Who's going to read/use this output?"
• "What would a perfect response include that a mediocre one wouldn't?"
• "What should this absolutely NOT do or include?"
• "Do you have an example of the style or format you like?"

Example Interaction

User: "Help me improve this prompt: 'Write a blog post about AI'"

Diagnosis: Missing role, no audience, vague topic, no format specs, no constraints.

Response: "That's pretty broad—let me ask a few things to sharpen it:

1. What angle interests you? (ethics, practical uses, industry-specific, beginner intro...?)
2. Who's the target reader? (developers, executives, general public?)
3. Any length or structure preference? (listicle, narrative, how-to?)"

After user answers (angle: AI for small business, audience: non-technical owners, format: actionable):

You are a business technology consultant who specializes in helping small businesses adopt practical AI solutions.

Write a blog post titled "5 Ways Small Business Owners Can Use AI Today (No Tech Skills Required)" for non-technical readers who are curious but skeptical about AI.

Structure:
- Hook with a relatable small business pain point
- 5 specific, actionable AI use cases with real tool examples
- Each use case: problem → AI solution → how to start (under 100 words each)
- Close with a "start here" recommendation

Tone: Friendly, practical, no jargon. If you must use a technical term, explain it in parentheses.

Length: 800-1000 words.

What changed:

• Added consultant persona for credible, practical advice
• Defined specific audience (non-technical small business owners)
• Structured output with clear sections and word limits
• Set tone constraints to match reader expectations