Use when adding new error messages to React, or seeing "unknown error code" warnings.
npx skills add shuiyuan1223/skillfix_xuetang --skill "add-skill"
Install specific skill from multi-skill repository
# Description
Use when adding a new Agent Skill to PHA. Covers SKILL.md format, triggers, MCP tool integration, and Git tracking.
# SKILL.md
name: add-skill
description: Use when adding a new Agent Skill to PHA. Covers SKILL.md format, triggers, MCP tool integration, and Git tracking.
Add New Agent Skill to PHA
What is a Skill?
Skills are expert knowledge modules that guide the Agent's behavior. They are NOT code — they are structured Markdown documents with YAML metadata.
- MCP Tools = hands (do things: fetch data, execute operations)
- Skills = brain (judge things: interpret data, evaluate quality, guide behavior)
Checklist
1. Create Skill Directory & File
- [ ] Create
src/skills/<skill-name>/SKILL.md - [ ] Skill name should be kebab-case (e.g.
sleep-coach,heart-monitor)
2. Write SKILL.md
Follow the OpenClaw format:
---
name: skill-name
description: "一句话描述这个 Skill 做什么"
metadata:
pha:
emoji: "icon-name" # Icon name (NOT emoji), e.g. "heart", "brain"
triggers: # Keywords that auto-activate this skill
- "关键词1"
- "关键词2"
config: {} # Optional config (usually empty)
---
# Skill Title
## 角色定位
描述 Agent 在使用此 Skill 时扮演的角色。
## 评估框架 / 行为指导
Skill 的核心内容:评分标准、解读规则、建议模板等。
## 与 MCP 工具的配合
说明需要调用哪些 MCP 工具获取数据,以及如何解读数据。
3. Design Triggers
Triggers are keywords in user messages that auto-inject this Skill's guide into the Agent's context.
triggers:
- "睡眠" # Matches messages containing "睡眠"
- "sleep" # English variant
- "失眠" # Related terms
Design principles:
- Use 3-8 specific keywords, avoid overly broad terms
- Include both Chinese and English variants
- Avoid overlap with other Skills' triggers
- Triggers are substring matches, not exact matches
4. Integrate with MCP Tools (if needed)
If the Skill needs data from MCP tools, describe the integration in the Skill body:
## 数据获取
使用以下 MCP 工具获取所需数据:
- `get_sleep`: 获取睡眠数据(入睡时间、深睡比例等)
- `get_heart_rate`: 获取心率数据用于睡眠质量评估
## 解读规则
根据获取的数据,按以下标准评估:
- 深睡比例 > 20%: 良好
- 深睡比例 15-20%: 一般
- 深睡比例 < 15%: 需改善
5. Verify
bun run check # Ensure no TypeScript errors
bun run build # Build succeeds
After deployment, test by sending a message containing one of the trigger keywords and verify the Agent uses the Skill's guidance.
File Structure
src/skills/
├── sleep-coach/
│ └── SKILL.md # 睡眠教练
├── evolution-driver/
│ └── SKILL.md # 进化方法论
├── benchmark-evaluator/
│ └── SKILL.md # 评测框架
└── your-new-skill/
└── SKILL.md # 新 Skill
Git Tracking
- Skills are stored in the Git-tracked
src/skills/directory - When edited via Settings > Skills UI, changes are auto-committed with
gitCommitFiles() - Commit message format:
feat(skill): update <skill-name>
When to Use a Skill vs MCP Tool
| Need | Solution |
|---|---|
| Fetch data from API | MCP Tool (src/tools/) |
| Read/write database | MCP Tool |
| Expert judgment / scoring rubric | Skill |
| Behavior guidance / response template | Skill |
| Both data + interpretation | MCP Tool provides data, Skill provides interpretation |
UI Management
Skills can be managed through the web UI:
- Settings > Skills — View, edit, enable/disable skills
- The UI reads from src/skills/*/SKILL.md and provides a code editor
- Saving in the UI writes the file and creates a Git commit
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.