AgentSkillsCN

agent-creator

提供 AI Agent 的创建、配置与优化指南。当用户希望定义新的 Agent 角色、生成系统提示词，或为特定领域的 Agent（如“创建 QA Agent”或“设计安全审计 Agent”）定制专属技能与工作流程时，可使用此技能。

SKILL.md
--- frontmatter
name: agent-creator
description: Guide for creating, configuring, and refining AI Agents. Use this skill when users want to define a new agent persona, generate a system prompt, or assemble a specific set of skills/workflows for a specialized agent (e.g., "Create a QA Agent" or "Design a Security Auditor Agent").

Agent Creator

This skill provides a structured process for designing and configuring specialized AI Agents.

When to Use

Use this skill when you need to:

  1. •Create a New Agent: Define a purpose-built agent with specific expertise (e.g., "Make a Frontend Specialist Agent").
  2. •Generate System Prompts: Create robust, effective system instructions for an agent.
  3. •Assemble Capabilities: Select the right combination of Skills, Workflows, and Rules for a specific domain.
  4. •Refine Agent Behavior: specialized tuning of an existing agent's operational guidelines.

Agent Architecture

An Agent in the Antigravity system is defined by a markdown file in .agent/agents/{name}.md containing:

1. Frontmatter (Metadata)

  • •name: Kebab-case identifier (e.g., backend-specialist).
  • •description: Short summary and trigger keywords.
  • •tools: List of tools the agent has access to (e.g., Read, Write, Bash).
  • •model: The model usage strategy (usually inherit).
  • •skills: Comma-separated list of skills from .agent/skills/ this agent needs.

2. Identity & Charter

  • •Role: Who the agent is.
  • •Philosophy: Core beliefs driving decisions.
  • •Mindset: Operational mode and priorities.

3. Critical Guidelines (The "Stop & Ask" Protocol)

  • •CRITICAL: CLARIFY BEFORE CODING: A mandatory section forcing the agent to ask clarifying questions before making assumptions about stack, runtime, or tools.

4. Decision Frameworks

  • •Tables and logic guides to help the agent make technical decisions (e.g., "Node vs Python", "SQL vs NoSQL").

5. Capabilities & specialized Lists

  • •Expertise Areas: Deep dive into specific techs.
  • •Quality Control Loop: Mandatory steps to run after every edit.

Workflow: Creating an Agent

Follow these steps to create a new Agent.

Step 1: Define the Goal

Ask the user for the Agent's primary purpose.

  • •Prompt: "What is the primary goal of this agent? What domain does it specialize in?"

Step 2: Select Capabilities (Skills)

Analyze the available Skills in .agent/skills/ to recommend the best set to include in the skills frontmatter.

  • •Example: A Backend Agent needs nodejs-best-practices, database-design.

Step 3: Draft the Agent Definition

Use the Agent Template in assets/agent_template.md as the mandatory base.

  1. •Frontmatter: Fill in name, tools, and required skills.
  2. •Philosophy & Mindset: Define how the agent thinks, not just what it does.
  3. •Critical Clarifications: Define what the agent MUST ask users before starting (e.g., "Which framework?", "Which DB?").
  4. •Decision Frameworks: Populate tables with current best practices for the domain.

Step 4: Save the Artifact

Save the file to .agent/agents/{name}.md.

  • •Ensure the filename matches the name in frontmatter.

Tools & Resources

Agent Template

Use assets/agent_template.md to structure the agent definition. Strictly follow this structure.

Best Practices for specialized Agents

  • •Opinionated Defaults: Agents should have strong opinions (Philosophy) but flexible execution (Clarification).
  • •Mandatory Checks: Include a "Quality Control Loop" that forces the agent to validate its own work (Lint, Test, Security).
  • •Anti-Patterns: Explicitly list what the agent should AVOID.