Find agent skills by outcome

- Balanced context: enough to be accurate

not so much that it wastes tokens.

- Good structure (sections

headings

- Replace long phrases with precise

compact ones.

- If issues are found

revise once

- Prefer short

sharp rules over long descriptive paragraphs.

- Things to avoid (hallucinations

fabrications

- Limits (max length

language

- Output format (Markdown sections

JSON schema

- Remove fluff

repetition

- Define the role

expertise

- Match the task (e.g. senior engineer

legal analyst

- The user only wants a direct answer/content

not a prompt for another AI.

If in doubt

assume they want a better

- The user wants actions executed (running code

calling APIs) instead of prompt design.

- “Tạo prompt chuẩn cho việc X (code

viết bài

- A long

noisy

- Wants to design

improve

for another AI system (GPT

Claude

- Robust – low hallucination

resilient to edge cases.

for systems like GPT

Claude

- Turn them into a single

clean

- Take raw

messy

user requests into concise

token-efficient

Prompt Refiner