A skill that teaches agents how to design, write, and review agent skills — covers trigger design, progressive disclosure, common patterns, anti-patterns, and quality gates.
概览
A skill that teaches agents how to write better skills — and is itself the best example of its own rules. Drawn from Anthropic and OpenAI official documentation, the agentskills.io specification, and real-world community experience. When you ask an agent to create a skill, it often produces a wall of text that looks professional but breaks down in practice: * descriptions that never trigger; * bodies cluttered with obvious explanations, burning through context; * directory structures blindly copied from software projects. This skill does one thing: it tells usable apart from good. And it uses its own design as the example — a Task Routing table up front (Activation Over Storage), a body that bluntly explains it's teaching one structure while also showing another (Progressive Disclosure), and every design principle paired with an anti-pattern check that puts the rule, the test, and the failure modes in one reading block. Copy SKILL.
README
English | 中文
Skill Authoring
A skill that teaches agents how to write better skills — and is itself the best example of its own rules. Drawn from Anthropic and OpenAI official documentation, the agentskills.io specification, and real-world community experience.
What this is
When you ask an agent to create a skill, it often produces a wall of text that looks professional but breaks down in practice:
- descriptions that never trigger;
- bodies cluttered with obvious explanations, burning through context;
- directory structures blindly copied from software projects.
This skill does one thing: it tells usable apart from good. And it uses its own design as the example — a Task Routing table up front (Activation Over Storage), a body that bluntly explains it’s teaching one structure while also showing another (Progressive Disclosure), and every design principle paired with an anti-pattern check that puts the rule, the test, and the failure modes in one reading block.
Installation
Copy SKILL.md into your agent’s skills directory:
| Agent | Install path |
|---|---|
| OpenCode / Codex | ~/.agents/skills/skill-authoring/ |
| Claude Code | ~/.claude/skills/skill-authoring/ |
| Cursor | .cursor/skills/skill-authoring/ |
Or clone the repository:
git clone https://github.com/Ronifue/skill-authoring.git ~/.agents/skills/skill-authoring
What is inside
A single-file skill, 328 lines in total. Opens with a Task Routing table so you can jump straight to what you need. A Quickstart gives you a minimal copy-paste skill skeleton.
Six design principles, each paired with a Check question and an inline anti-pattern warning. The rule, its test, and the common failure modes sit together in one reading block. A Decision Tree maps whether to create a skill, save a reference, or use a one-time prompt. A Degrees of Freedom framework teaches you to pick the right skill shape based on task fragility.
Four common patterns, each with concrete examples (gotchas, templates, checklists, validation loops). A P0-P2 quality checklist you can use for self-testing. Also covers cross-harness thin shells, anti-drift maintenance, and auditing third-party skills.
Relationship to prompt-writing skills
Skills that handle prompt authoring take care of language: phrasing, XML sections, safety boundaries. This skill takes care of architecture: file placement, routing design, progressive disclosure. The two are orthogonal — use both, just trigger them for different tasks.
Credits
This skill draws from:
- Anthropic Agent Skills documentation
- OpenAI Codex Skills guide
- agentskills.io specification
- COMPASS Skill Writing Tutorial by dongshuyan
- Linux DO — How to Write a Good Skill by woji_666
- Community patterns from agentpatterns.ai, Perplexity research, and mgechev’s best practices
Learn more on Linux DO
License: MIT
推荐工具
换一个关键词，或者移除筛选条件。
安装
npx skillfish add ronifue/skill-authoring