一键导入
skill-creation
Guide to skill creation. Use this skill when the user requests to make, create, or write a new skill.
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。
菜单
Guide to skill creation. Use this skill when the user requests to make, create, or write a new skill.
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
评估方法论 — 指导 agent 如何系统性评估代码库状态
已并入 implement skill,请勿新用
实现阶段主操作手册 — 指导 agent 完成改码与局部验证,并把提交留给独立 commit phase
规划规范 — 将评估结果收敛为结构化任务计划
流水线选择规范 — 根据任务和事实选择最合适的 pipeline
Evaluate another skill by inspecting its SKILL.md, referenced assets, and runnable helpers, then write a concise audit report plus a machine-readable score summary. Use when the caller asks to evaluate a skill, assess its readiness, or review quality/safety/testability. Do NOT use when the task is to create a new skill from scratch.
| name | skill_creation |
| description | Guide to skill creation. Use this skill when the user requests to make, create, or write a new skill. |
This document explains how to design and build effective skills for agents.
Skills are a manual of information that gives an agent focused capabilities in a specific domain. It includes instructions and reuseable resources
The context window is a limited space. Write only what an agent truly needs.
Every skill directory contains:
skill-name/
├── SKILL.md (required)
├── scripts/ (optional, for executable code, e.g. Python, Bash, etc.)
├── references/ (optional, documentation or schemas loaded only when needed to keep SKILL.md short)
└── assets/ (optional)
Do NOT create unrelated documentation on the creation of this skill. DO NOT create any README, changelog, installation guide, creation summary etc. files. The skill-name directory should contain only execution-relevant content.
name and description.Skills minimize context usage through three layers of progressive disclosure:
name and description of the skill is always in context (~100 words)Keep SKILL.md focused and short, move detailed explanations or lengthy codes/examples into other files. Always reference other files inside SKILL.md so the agent knows where they are.