Agent Skills: Create Meta-Prompts

Create Meta-Prompts

Quick Start

Workflow

1. Intake - Determine purpose (Do/Plan/Research/Refine)
2. Generate - Create prompt using purpose-specific templates
3. Execute - Run prompt with dependency-aware execution
4. Validate - Verify output and create summary

Directory Structure

.prompts/
├── 001-topic-research/
│   ├── 001-topic-research.md     # Prompt
│   ├── topic-research.md         # Output
│   └── SUMMARY.md                # Human summary
├── 002-topic-plan/
│   ├── 002-topic-plan.md
│   ├── topic-plan.md
│   └── SUMMARY.md

Usage

Create Research Prompt

/create-meta-prompt research authentication options for the app

What it does:

• Determines this is a Research task
• Scans for existing research files to reference
• Creates prompt with research-specific structure
• Saves to .prompts/{number}-{topic}-research/
• Executes with validation

Create Planning Prompt

/create-meta-prompt plan the auth implementation approach

What it does:

• Detects existing auth-research.md
• Asks if it should reference the research
• Creates plan prompt with research context
• Runs after research completes

Create Execution Prompt

/create-meta-prompt implement JWT authentication

What it does:

• References both research and plan
• Creates implementation prompt
• Includes verification steps

Prompt Types

Research Prompts

For gathering information with structured output.

Structure:

<objective>
Research {topic} to inform {purpose}
</objective>

<context>
{Background and requirements}
</context>

<output>
Save to: .prompts/{num}-{topic}-research/{topic}-research.md
Include: findings, recommendations, metadata
Create: SUMMARY.md
</output>

Plan Prompts

For creating approaches and strategies.

Structure:

<objective>
Create implementation plan for {topic}
</objective>

<context>
Research: @.prompts/{num}-{topic}-research/{topic}-research.md
</context>

<output>
Save to: .prompts/{num}-{topic}-plan/{topic}-plan.md
Include: phases, tasks, dependencies
Create: SUMMARY.md
</output>

Do Prompts

For executing tasks and producing artifacts.

Structure:

<objective>
{What to build/create/fix}
</objective>

<context>
Plan: @.prompts/{num}-{topic}-plan/{topic}-plan.md
</context>

<output>
Create/modify: specified files
Verify: tests and checks
Create: SUMMARY.md
</output>

Execution Engine

Single Prompt

Straightforward execution of one prompt:

1. Create folder: .prompts/{number}-{topic}-{purpose}/
2. Write prompt to: {number}-{topic}-{purpose}.md
3. Execute with Task agent
4. Validate output
5. Archive prompt

Sequential Execution

For chained prompts:

1. Build execution queue from dependencies
2. Execute each prompt in order
3. Validate after each completion
4. Stop on failure
5. Report results

Parallel Execution

For independent prompts:

1. Spawn all Task agents in single message
2. Wait for completion
3. Validate all outputs
4. Report consolidated results

Chain Detection

Automatically detects dependencies:

• Scans for existing *-research.md and *-plan.md
• Matches by topic keyword
• Suggests relevant files to reference
• Determines execution order

Validation

After execution, verifies:

• Output file created and not empty
• Required XML metadata present
• SUMMARY.md created with all sections
• One-liner is substantive

Summary Format

Every execution creates SUMMARY.md:

# {Topic} {Purpose} Summary

**{Substantive one-liner describing outcome}**

## Key Findings
- {Finding 1}
- {Finding 2}

## Files Created
- `path/file.ext` - Description

## Decisions Needed
{Specific items or "None"}

## Next Step
{Concrete forward action}

References

Templates:

• references/research-patterns.md - Research prompt templates
• references/plan-patterns.md - Planning prompt templates
• references/do-patterns.md - Execution prompt templates

Supporting:

• references/metadata-guidelines.md - Metadata structure
• references/question-bank.md - Intake questions
• references/summary-template.md - Summary format

Success Criteria

Prompt Creation:

• Purpose identified correctly
• Chain detection performed
• Prompt structure matches purpose
• Output location specified
• SUMMARY.md requirement included

Execution:

• Dependencies correctly ordered
• Output validated
• SUMMARY.md created
• Results presented clearly