AgentSkillsCN

prompt-authoring

运用CAGEERF方法论,打造高效提示词、链式流程与智能门控机制的指导方案

SKILL.md
--- frontmatter
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 agents-prompts-mcp system.

CAGEERF Methodology

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

PhasePurposeOutput
ContextGather domain knowledgeSituational understanding
AnalysisBreak down the problemStructured insights
GoalsDefine success criteriaMeasurable objectives
ExecutionImplement the solutionConcrete deliverables
EvaluationValidate against goalsQuality assessment
RefinementIterate based on feedbackImproved output
FinalizationComplete and documentFinal deliverable

Prompt Structure

yaml
---
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 -->:

code
>>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:

yaml
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:

typescript
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