Guidance for creating effective prompts, chains, and gates using CAGEERF methodology

121 stars
28 forks
4 views

SKILL.md


name: prompt-authoring description: Guidance for creating effective prompts, chains, and gates using CAGEERF methodology version: 1.0.0

Prompt Authoring Guide

This skill provides patterns for creating prompts in the claude-prompts system.

CAGEERF Methodology

Prompts follow the C.A.G.E.E.R.F framework:

Phase Purpose Output
Context Gather domain knowledge Situational understanding
Analysis Break down the problem Structured insights
Goals Define success criteria Measurable objectives
Execution Implement the solution Concrete deliverables
Evaluation Validate against goals Quality assessment
Refinement Iterate based on feedback Improved output
Finalization Complete and document Final deliverable

Prompt Structure

---
id: my-prompt
name: My Prompt
description: What this prompt accomplishes
category: development
execution_hint: single  # or 'chain'
arguments:
  - name: input
    type: string
    description: Primary input
---
system_message: |
  You are an expert at [domain].

user_message_template: |
  Task: {input}

  Requirements:
  - Requirement 1
  - Requirement 2

Chain Definition

Chains connect prompts with -->:

>>analyze --> >>design --> >>implement --> >>test

Chain features:

  • Sequential execution with context passing
  • Gate validation between steps
  • Resume capability with chain_id

Quality Gates

Gates validate output quality:

id: code-quality
name: Code Quality Gate
severity: high
criteria:
  - "Code follows established patterns"
  - "Error handling is comprehensive"
pass_criteria:
  - "All criteria met with evidence"

Severity levels: critical | high | medium | low

Creating New Prompts

Use the resource_manager tool:

resource_manager({
  resource_type: "prompt",
  action: "create",
  id: "new-prompt",
  name: "New Prompt",
  description: "Purpose",
  category: "development",
  system_message: "...",
  user_message_template: "..."
})

Best Practices

  1. Clear objectives: Define what success looks like
  2. Structured output: Specify expected format
  3. Validation criteria: Include checkable requirements
  4. Composability: Design prompts that chain well
  5. Gate integration: Add quality checks for critical outputs