refine-prompt
npx machina-cli add skill PaulRBerg/agent-skills/refine-prompt --openclawContext
- Working directory: !
pwd - Request: $ARGUMENTS
Task
You are an expert prompt engineer. Create an optimized prompt based on $ARGUMENTS.
1. Craft the Prompt
Apply relevant techniques:
- Few-shot examples (when helpful)
- Chain-of-thought reasoning
- Role/perspective setting
- Output format specification
- Constraints and boundaries
- Self-consistency checks
Structure with:
- Clear role definition (if applicable)
- Explicit task description
- Expected output format
- Constraints and guidelines
2. Display the Result
Show the complete prompt in a code block, ready to copy:
[Complete prompt text]
Briefly note which techniques you applied and why.
3. Save to .ai/PROMPT.md
First ensure the directory exists: mkdir -p .ai
If .ai/PROMPT.md exists:
Read current contents and append:
---
## [Brief title from $ARGUMENTS]
[The optimized prompt]
If .ai/PROMPT.md does not exist:
Create with:
# Optimized Prompts
## [Brief title from $ARGUMENTS]
[The optimized prompt]
Confirm: "Saved to .ai/PROMPT.md"
Source
git clone https://github.com/PaulRBerg/agent-skills/blob/main/skills/refine-prompt/SKILL.mdView on GitHub Overview
An expert prompt engineer refines user prompts for LLMs to improve clarity, constraints, and results. It applies techniques like few-shot examples, chain-of-thought, role/perspective settings, and explicit output formats to produce a ready-to-use prompt and an updated PROMPT.md entry.
How This Skill Works
The skill takes ARGUMENTS to craft an optimized prompt. It structures the prompt with a clear role, explicit task, defined output format, and constraints, then displays the complete prompt in a code block and notes the techniques used. Finally it saves the result to .ai/PROMPT.md according to the presence or absence of the file.
When to Use It
- You want to optimize or rewrite a prompt for an LLM to improve clarity and results
- You need explicit output formats, constraints, or role-based prompts
- You want to add few-shot examples or chain-of-thought guidance to improve accuracy
- You want to append the optimized prompt to PROMPT.md for reuse
- You are prompting for prompt engineering or appending to PROMPT.md
Quick Start
- Step 1: Provide the prompt you want refined as ARGUMENTS
- Step 2: The assistant crafts an optimized prompt with a clear role, task, output format, constraints, and self-checks, then displays it in a code block
- Step 3: The assistant saves or appends the result to .ai/PROMPT.md per the SKILL rules
Best Practices
- Define a clear role and task for the model
- Specify explicit output format and required fields
- Embed constraints and boundaries, plus validation steps
- Include few-shot examples when helpful
- Document which techniques were used and why
Example Use Cases
- Rewrite a user prompt to produce a JSON summary with fields: title, date, author, and key_points
- Create a prompt that instructs the model to compare two products and output results as YAML with pros/cons
- Ask the model to extract structured data from customer messages, e.g., email, order_id, date
- Provide a role-based prompt: act as a senior developer debugging Python code and output a concise plan
- Request an outline for a blog post on prompt engineering with clear H2 sections and a final CTA
