菜单
Agent Skill for creating new skills according to the official specification. Use when asked to create or update a skill, and follow specifications from https://agentskills.io/.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Agent Skill for planning implementation work into comprehensive, testable steps. Use before implementation to create a clear roadmap for execution.
Agent Skill for writing and reviewing durable design specs. Use when the user asks to write, formalize, save, review, or approve a spec, design document, RFC, or implementation-ready proposal. If the direction is still vague or unsettled, use `brainstorming` first before producing the spec. If the direction is already clear and no spec is needed, use `plan`.
Agent Skill for spawning new AI agents with interactive TUI in tmux. Use when work can be offloaded to new or existing AI agents (exploration, code review, second opinions, parallel work, etc.). Can also be used if spawning new agents with different model families (e.g. Spawn GPT models when you are Claude Code, Spawn Claude models when you are Codex, Or spawn any other models (e.g. DeepSeek, Gemini) with Pi). Prefer harness's own subagent tools when no need for a different model.
Agent Skill for OpenCLI. Use when the user wants to control websites through the `opencli` CLI, run or inspect OpenCLI site/app/external adapters, use the OpenCLI browser bridge/CDP automation
Agent Skill for collaborative brainstorming and design exploration. Use when the user wants to think through an idea, compare options, clarify trade-offs, or get a quick recommendation, without forcing a written artifact. If the user wants a saved, reviewed design spec, use `design-spec`.
Agent Skill for working with Python. Use when writing, editing, or reviewing Python code and scripts.
| name | skill-creator |
| description | Agent Skill for creating new skills according to the official specification. Use when asked to create or update a skill, and follow specifications from https://agentskills.io/. |
The procedure is very straightforward. Use curl or web crawl tool to fetch and follow guides from various pages in official documentation site. Strictly follow the following:
First, follow the official specification at specification.
For improving description triggering accuracy, follow the guide at optimizing descriptions, while ensure the following user conventions are followed:
Description field format: 3 parts:
After creating or updating a skill, evaluate its output quality. Follow the eval framework at evaluating skills.
For bundling scripts in skills, follow the guide at using scripts.
If you are adapting a skill from an external upstream source, also follow the instruction in porting-existing-skills.
If no web access, local copies are available but may be outdated. Prefer web access if possible.
