AgentSkillsCN

prompt-engineering

在编写、版本控制或测试 LLM 提示词与技能定义时，应用 VCS-001 规范。

SKILL.md
--- frontmatter
name: prompt-engineering
description: Apply PES-001 when writing, versioning, or testing LLM prompts and skill definitions for agents.

Prompt Engineering Standard (PES-001)

Use this skill when creating or updating agent prompts, skill definitions, or prompt tests so behavior is reproducible and resilient to model changes.

When to Use

  • •Writing or refactoring system prompts for agents
  • •Defining skill files (.md) that encode agent capabilities
  • •Structuring prompts for clear instruction-following and injection resistance
  • •Versioning and testing prompts (e.g., evals, A/B tests)

Instructions

  1. •Structure: Use the mandatory XML-style tags—<system_role>, <engineering_principles>, <instructions>, <reference_example>, <hitl_protocol>—so parsing and governance are consistent.
  2. •Skill format: Store each skill as a .md file with Context (role, input, output, tools) and a system prompt block. Keep prompts in the repo and treat them as code.
  3. •Testing: Version prompts and run evals (e.g., Promptfoo) so Agent behavior can be regression-tested.

For the full normative standard and skill definition format, see references/prompt-engineering.mdx.