| name | prompt-engineering |
| description | Designs, optimizes, and validates high-performance prompts for LLMs and AI agents. Use when improving vague prompts, building reusable prompt templates, enforcing structured outputs, or designing agent workflows with verification loops. |
| metadata | {"author":"nick-brennan-system","version":"2.0","category":"ai-core-skill"} |
🧠 Prompt Engineering Agent Skill
Purpose
Transform user intent into structured, high-performance prompts that produce reliable, repeatable, and verifiable outputs.
This skill follows modern prompt engineering principles:
- Clear instructions
- Structured formatting
- Iterative refinement
- Verification loops
Core Prompt Framework (MANDATORY)
All prompts MUST follow this structure:
[ROLE]
Define expertise and perspective
[TASK]
Explicit objective using action verbs
[CONTEXT]
Relevant background, constraints, or data
[INSTRUCTIONS]
Step-by-step execution rules
[OUTPUT FORMAT]
Exact structure required (JSON, table, bullets)
[CONSTRAINTS]
Restrictions and rules
[EXAMPLES] (optional)
Input → Output samples
[SELF-CHECK] (REQUIRED)
- Validate output format
- Validate constraints
- Fix errors automatically if needed
Execution Workflow
Step 1: Diagnose
- Identify user intent
- Detect ambiguity
- Identify missing constraints
Step 2: Classify Prompt Type
- Generation
- Transformation
- Extraction
- Reasoning
- Agent Workflow
Step 3: Construct Prompt
- Add structured framework
- Add explicit instructions
- Define output format
- Add constraints
Step 4: Apply Reasoning Strategy
- SIMPLE → direct execution
- STANDARD → structured steps
- COMPLEX → plan → execute → verify
Step 5: Add Verification Loop (CRITICAL)
All prompts must include:
[SELF-CHECK]
- Did output follow format? (YES/NO)
- Did output meet constraints? (YES/NO)
- If NO → regenerate corrected output
Output Modes
Mode A: Optimized Prompt Only
Returns production-ready prompt
Mode B: Full Breakdown (DEFAULT)
- Issues detected
- Optimized prompt
- Explanation
- Variants
Mode C: System Template
Reusable prompt template for workflows
Prompt Design Rules
- Use explicit verbs (Write, Build, Analyze, Extract)
- Never assume context — define it
- Avoid conflicting instructions
- Always define output structure
- Prefer constraints over suggestions
- Break complex tasks into steps
- Force deterministic outputs when possible
Advanced Capabilities
🔹 Multi-Step Prompt Systems
Split tasks into:
- Planning Prompt
- Execution Prompt
- Verification Prompt
🔹 Meta Prompting Loop
- Diagnose failure
- Rewrite prompt
- Validate output
🔹 Structured Output Enforcement
Always enforce:
- Tables
- JSON
- Bullet systems
🔹 Agent Integration
When used in agents:
- Define tool usage conditions
- Require explanation before actions
- Enforce step-by-step execution
Verification Block
[SELF-CHECK]
- Did output follow format? (YES/NO)
- Did it meet constraints? (YES/NO)
- If NO → fix automatically
Execution Block
[MODE]
- FAST → minimal reasoning
- BALANCED → default
- DEEP → multi-step + validation
Stop Conditions
[STOP CONDITIONS]
- Task is complete
- All constraints satisfied
- Output validated
Tool Awareness
[TOOLS]
- Use tools when needed
- Explain action before using tool
Example
Input
"Make this better"
Output
[ROLE]
You are a senior content editor specializing in clarity and engagement.
[TASK]
Rewrite the provided content to improve clarity, readability, and engagement.
[CONTEXT]
The content may contain vague language and weak structure.
[INSTRUCTIONS]
- Improve sentence clarity
- Remove redundancy
- Strengthen wording
- Maintain original meaning
[OUTPUT FORMAT]
Rewritten version followed by bullet list of improvements
[CONSTRAINTS]
- Do not change intent
- Do not add new ideas
[SELF-CHECK]
- Is the content clearer? (YES/NO)
- Are constraints followed? (YES/NO)
Edge Case Handling
- If prompt is vague → infer intent and state assumptions
- If no format provided → enforce structured output
- If task is complex → split into steps
- If already optimized → refine minimally
Integration Notes
Use this skill in:
- Copilot agents
- MCP systems
- Prompt libraries
- Workflow automation
Store reusable prompts in:
assets/templates.md
End of Skill
