AgentSkillsCN

Llm Prompt Engineer

LLM 提示词工程师

SKILL.md

Skill: LLM Prompt Engineer

Description

This skill guides the creation and modification of LLM prompts stored in the backend/llm/prompts/ directory.

Rules

  1. •File Format: All prompts must be stored as JSON files.
  2. •Structure: The JSON object must have at least two keys: system and human.
    • •system: The instructions for the AI behavior.
    • •human: The template for the user's input.
  3. •Variables: Preserve existing variables wrapped in curly braces (e.g., {input}, {steps}). Do NOT remove them unless explicitly instructed.
  4. •JSON Validity: Ensure the file is valid JSON. Escape quotes within strings (e.g., "). Use \n for newlines.
  5. •Naming: Filenames should be snake_case (e.g., plan_refinement.json).

Example

File: backend/llm/prompts/example.json

json
{
    "system": "You are a helpful assistant.\n\nRules:\n1. Be concise.\n2. Use markdown.",
    "human": "Please help me with: {user_query}"
}

Procedure

  1. •Read: If modifying, read the existing file first to understand the context and variables.
  2. •Edit: Apply changes to the system or human fields.
  3. •Verify: Check that all brackets match and JSON is valid.