Objective

Create effective prompts for both direct execution and multi-stage AI-to-AI workflows. This skill handles standalone prompts and meta-prompts (chains where outputs from one prompt become inputs for the next).

Prompt Categories

Single Prompts

Standalone prompts for direct execution by users or agents.

• Analysis: Examine and interpret information, extract insights
• Generation: Create new content or artifacts
• Review: Quality checks, validation, critique
• Transformation: Conversion, extraction, summarization
• Q&A: Answer specific questions
• Creative: Brainstorming, ideation, problem-solving

Meta-Prompts

Multi-stage prompt chains for Claude-to-Claude communication where outputs feed into subsequent prompts.

• Research Prompts: Gather information, analyze options, produce findings
• Plan Prompts: Create implementation roadmaps and strategies
• Do Prompts: Execute tasks, produce artifacts
• Refine Prompts: Improve existing prompts with versioning

Output Structure

Single Prompts

.cattoolkit/prompts/
├── {number}-{name}.md    # The prompt file
└── SUMMARY.md             # Summary of created prompt

Meta-Prompts

.cattoolkit/metaprompt/
├── {number}-{topic}-{purpose}/
│   ├── {number}-{topic}-{purpose}.md  # The prompt file
│   ├── {topic}-{purpose}.md            # Output file (for Research/Plan)
│   ├── completed/                      # Archive after running
│   └── SUMMARY.md                      # Required: executive summary

Prompt Numbering

Auto-increment from existing prompts:

# Single prompts
find .cattoolkit/prompts -name "*.md" | wc -l

# Meta-prompts
find .cattoolkit/metaprompt -type d -name "*-*" | wc -l

Key Methodology

XML/Markdown Decision Framework

The Two-Pattern Model

For AI Agents, use simplified patterns designed for structuring AI cognition:

Pattern	Tags	When to Use
Pure Markdown	0	Default for simple, linear workflows
Hybrid XML/Markdown	1-5	For structuring AI cognition, state tracking, safety constraints

Decision Criteria

Choose Pure Markdown if ALL are true:

• ✓ Simple, linear workflow
• ✓ AI can execute in single pass
• ✓ No state tracking needed
• ✓ No critical safety constraints
• ✓ No large data to isolate

Choose Hybrid XML/Markdown if ANY are true:

• ⚡ AI needs to organize multiple pathways/approaches (routing)
• ⚡ Multi-phase execution with state tracking
• ⚡ Critical safety constraints (NEVER/MUST NOT)
• ⚡ Large data that could confuse instructions
• ⚡ Non-negotiable workflow sequences

XML Tags for AI Cognition

Available Tags (max 3-5):

Tag	Purpose	Use When
<intent_analysis>	Analyze context to determine appropriate approach	AI needs to route based on task analysis
<routing>	Decision tables mapping context to workflows	Natural language logic for AI to route between paths
<workflow>	Enforce step sequences	Order matters, steps cannot be skipped
<state>	Track multi-phase context	Multi-phase execution with continuity
<context>	Separate data from instructions	Large data blocks that could confuse
<constraints>	Safety boundaries	NEVER/MUST NOT rules critical to safety

Note: <intake> is deprecated. Use <intent_analysis> instead for analyzing context.

Critical Rules:

• Limit to 3-5 XML tags maximum
• Never nest XML tags
• All content between tags should be Markdown
• Single prompts should use Pure Markdown (0 tags)

Prompt Creation Steps

Single Prompts

1. Understand the Task: What inputs, outputs, context?
2. Select Pattern: Use Pure Markdown (0 tags) for simple prompts
3. Structure: Use clear Markdown headings and sections
4. Include Examples: Concrete examples where helpful
5. Add Verification: How to confirm successful output

Meta-Prompts

1. Design the Chain: Research → Plan → Do sequence
2. Structure Each Prompt: Use hybrid XML/Markdown (1-5 tags)
3. Add Required Tags: Include <output_format> for chain integration
4. Set Dependencies: Reference previous outputs with @ syntax
5. Add YAML Frontmatter: Include confidence, dependencies, questions
6. Create SUMMARY.md: Required for each prompt in the chain

Chaining Patterns

• Sequential: Each prompt depends on previous output
• Parallel: Multiple prompts run simultaneously
• Mixed: Parallel layer feeds into sequential step

Templates Index

Prompt templates in assets/templates/:

• assets/templates/single-prompt.md - Standalone analysis/generation prompt
• assets/templates/research-metaprompt.md - Multi-phase research chain prompt
• assets/templates/plan-metaprompt.md - Structured planning prompt
• assets/templates/do-metaprompt.md - Execution workflow prompt
• assets/templates/refine-metaprompt.md - Iteration and improvement prompt

WHY: Templates ensure prompts follow consistent XML/Markdown hybrid structure and include quality checks.

Reference Files

Single Prompts

• references/prompt-types.md - Detailed type guidance
• references/single-prompt-templates.md - Template patterns
• references/optimization-guide.md - Quality improvement

Meta-Prompts

• references/research-patterns.md - Research prompt structure
• references/research-pitfalls.md - Common research mistakes to avoid
• references/plan-patterns.md - Plan prompt structure
• references/do-patterns.md - Do prompt structure
• references/refine-patterns.md - Refine prompt structure
• references/metadata-guidelines.md - YAML frontmatter requirements
• references/summary-template.md - SUMMARY.md structure

General Reference

• references/intelligence-rules.md - Prompt complexity guidelines
• references/question-bank.md - Clarification questions for requirements gathering