Skill Master
Minimum Model
Any model. This is a lookup table — any model can use it.
⚠️ Skill Count Rule
Current: 19 active skills. Sweet spot: 15–25. Above 30 = routing breaks down.
Start lean. Add a new skill only when there's a clear, recurring trigger that no existing skill covers.
How to Use This Skill
- 1. Read the owner's request.
- Find a match in the Quick Lookup table below.
- If not found, use the Decision Tree.
- ⚠️ Log the selection FIRST — before doing any work (see Analytics below).
- Load that skill's SKILL.md and follow it.
Do not improvise. If no skill matches, say so and ask the owner.
❌ Skipping the log is not allowed. Every skill invocation must be recorded, even for simple tasks. This is how the agent and the owner track what's working.
📊 Analytics — MANDATORY: Log Every Skill Use
This step is NOT optional. Before starting any skill, append one line:
CODEBLOCK0
Replace:
- -
SKILL_NAME → the skill selected (e.g. meetings) - INLINECODE2 → the phrase that matched (e.g.
schedule meeting with Daniel) - INLINECODE4 →
dm, group:<name>, or INLINECODE7
This is ~50 bytes/entry. Non-negotiable.
Why This Matters
- - The owner can ask "skill stats" at any time to see what's being used
- Unused skills get pruned, improving routing quality
- Shared across the PA network so every agent learns from usage patterns
- Enables weekly reports on which skills are carrying the most weight
To View Analytics
Ask: "skill stats" / "skill usage" / "which skills am I using?" → triggers
skill-analytics skill
Quick Lookup — By Trigger Phrase
| If the owner says... | Use skill |
|---|
| "schedule a meeting with X" | meetings |
| "summarize meeting notes" / "action items from meeting" |
meetings |
| "what's on my calendar today" | owner-briefing |
| "send me a morning briefing" | owner-briefing |
| "billing error" / "API out of credits" | billing-monitor |
| "connect my calendar" / "can't write to calendar" | calendar-setup |
| "connect Gmail" / "set up email" | calendar-setup |
| "set up a new PA" / "onboard a new agent" | pa-onboarding |
| "contact [person]'s PA" / "find PA phone number" | ai-pa |
| "set up monday.com" / "create a board item" / "monday question" | monday-for-agents |
| "I made a mistake" / "owner corrected me" | self-learning |
| "backup workspace" / "push to git" | maintenance |
| "update openclaw" / "update skills" / "run maintenance" | maintenance |
| "what was discussed in [group]" | whatsapp |
| "find new skill ideas" / "what skills are trending" | skill-scout |
| "skill usage" / "skill stats" / "skill report" | skill-analytics |
| "which skills am I using" / "unused skills" | skill-analytics |
| "security check" / "check for vulnerabilities" | self-monitor |
| "health check" / "check infrastructure" | self-monitor |
| "how are all the PAs doing" / "PA network status" | supervisor |
| "מה הסטטוס" / "what's the status" | supervisor |
| "how am I doing" / "review my performance" | eval |
| "run eval" / "מה עובד ומה לא" | eval |
| "summarize this YouTube video" | youtube-watcher |
| "add nikud to this" / "Hebrew vowel points" | hebrew-nikud |
| "compact memory" / "organize memory" | memory-tiering |
Decision Tree
CODEBLOCK1
Full Skill Library
| Skill | Category | When to Use |
|---|
| ai-pa | Coordination | Find PA contacts, group JIDs, coordination protocols |
| billing-monitor |
Health | Detect and respond to API billing failures |
|
calendar-setup | Setup | Calendar connection with write access + Gmail/email setup |
|
eval | Self-improvement | Full performance audit — scores tasks, checks PA health, reviews memory |
|
hebrew-nikud | Utility | Add nikud (vowel points) to Hebrew text, especially for TTS |
|
maintenance | Infrastructure | Workspace git backup (every 6h) + OpenClaw updates (weekly) |
|
meetings | Operations | Schedule meetings AND summarize meeting notes/transcripts |
|
memory-tiering | Memory | HOT/WARM/COLD memory compaction and archiving |
|
monday-for-agents | Operations | All monday.com operations: API, MCP, boards, items |
|
owner-briefing | Operations | Daily morning/evening summaries |
|
pa-onboarding | Setup | Full new agent setup from zero |
|
self-learning | Self-improvement | Log corrections and apply lessons; maintain HOT.md |
|
self-monitor | Health | Infrastructure + security checks, disk/memory/service health |
|
skill-master | Routing | Pick the right skill (this file) |
|
skill-scout | Discovery | Weekly search for new skill ideas |
|
supervisor | Operations | Network-wide status dashboard — all PAs, tasks, system health |
|
whatsapp | Memory | Per-conversation context, unanswered tracking, loop prevention |
|
youtube-watcher | Utility | Fetch and summarize YouTube video transcripts |
|
skill-analytics | Analytics | Track skill usage, generate daily reports, find unused skills |
Multi-Skill Workflows
Some tasks need multiple skills in sequence:
New PA Setup
CODEBLOCK2
PA Network Health Check
CODEBLOCK3
After a Mistake
CODEBLOCK4
Schedule a Meeting
CODEBLOCK5
Weekly Maintenance
CODEBLOCK6
After Important Group Chat
whatsapp (log decisions) → maintenance (push to GitHub)
Where to Run (Complexity Guide)
Run inline (main session)
- - ai-pa, billing-monitor, owner-briefing, supervisor, self-learning, maintenance
Consider subagent for heavy operations
- - calendar-setup, meetings (scheduling flow), monday-for-agents (bulk ops)
Spawn subagent (recommended)
- - pa-onboarding (20+ steps), eval (full monthly analysis), batch operations, skill-scout
Model Guidance
| Skill | Minimum Model |
|---|
| ai-pa, billing-monitor, supervisor, maintenance, owner-briefing | Any |
| calendar-setup, pa-onboarding, whatsapp, memory-tiering |
Small–Medium |
| meetings, monday-for-agents, skill-scout | Medium |
| eval (trend analysis), self-learning (writing rules) | Medium–Large |
Adding New Skills
When a new skill is added:
- 1. Add a row to the Full Skill Library table.
- Add trigger phrases to Quick Lookup.
- Update the Decision Tree if it fits a new category.
- Add to any relevant Multi-Skill Workflows.
- Check skill count — stay under 32 active skills.
- Add the skill name to the
KNOWN_SKILLS list in skill-analytics/SKILL.md.
Supervisor (Status Dashboard)
| Trigger | Action |
|---|
| "מה הסטטוס" / "what's the status" | supervisor |
| "supervisor" |
supervisor |
| "מה קורה" / "give me a summary" | supervisor |
The supervisor skill aggregates: active tasks, billing issues, group activity, pending follow-ups, and system health into one structured report.
Eval
| Trigger | Action |
|---|
| "תעשי eval" / "run eval" | eval |
| "מה עובד ומה לא" |
eval |
| "בדקי הכל" | eval |
The eval skill scores performance, audits tasks, checks PA network health, verifies integrations, and reviews memory quality — all in one report.
Cost Tips
- - This skill itself: Very cheap — it's a lookup table, any model works.
- Routing decision: If unsure, lean toward a smaller, cheaper skill first.
- Don't over-spawn: Use subagents only when the task would actually block the main session.
技能精通
最低模型要求
任何模型均可。这是一个查找表——任何模型都能使用。
⚠️ 技能数量规则
当前：19个活跃技能。最佳区间：15–25个。超过30个=路由失效。
从精简开始。只有当出现现有技能无法覆盖的明确、重复触发条件时，才添加新技能。
如何使用本技能
- 1. 阅读所有者的请求。
- 在下方快速查找表中寻找匹配项。
- 如未找到，使用决策树。
- ⚠️ 先记录选择——再开始任何工作（参见分析部分）。
- 加载该技能的SKILL.md文件并遵循其指示。
不要即兴发挥。如果没有匹配的技能，请告知所有者。
❌ 不允许跳过记录。 每次技能调用都必须记录，即使是简单任务也不例外。这是智能体与所有者追踪工作成效的方式。
📊 分析——强制要求：记录每次技能使用
此步骤不可省略。 在开始任何技能之前，追加一行记录：
bash
mkdir -p /opt/ocana/openclaw/workspace/data
echo {\ts\:\$(date -u +%Y-%m-%dT%H:%M:%SZ)\,\skill\:\技能名称\,\trigger\:\触发条件\,\context\:\上下文\} \
>> /opt/ocana/openclaw/workspace/data/skill-analytics.jsonl
替换：
- - 技能名称 → 所选技能（例如 meetings）
- 触发条件 → 匹配的短语（例如 schedule meeting with Daniel）
- 上下文 → dm、group:<名称> 或 cron
每条记录约50字节。不可协商。
为何重要
- - 所有者可随时询问技能统计以查看使用情况
- 未使用的技能会被修剪，提高路由质量
- 在PA网络间共享，使每个智能体都能从使用模式中学习
- 支持每周报告，显示哪些技能承担了最多工作
查看分析
询问：技能统计 / 技能使用情况 / 我在使用哪些技能？ → 触发 skill-analytics 技能
快速查找——按触发短语
| 如果所有者说... | 使用技能 |
|---|
| 安排与X的会议 | meetings |
| 总结会议记录 / 会议待办事项 |
meetings |
| 我今天有什么日程 | owner-briefing |
| 给我发送晨间简报 | owner-briefing |
| 账单错误 / API额度用尽 | billing-monitor |
| 连接我的日历 / 无法写入日历 | calendar-setup |
| 连接Gmail / 设置邮箱 | calendar-setup |
| 设置新的PA / 接入新智能体 | pa-onboarding |
| 联系[某人]的PA / 查找PA电话号码 | ai-pa |
| 设置monday.com / 创建看板项目 / monday问题 | monday-for-agents |
| 我犯错了 / 所有者纠正了我 | self-learning |
| 备份工作区 / 推送到git | maintenance |
| 更新openclaw / 更新技能 / 运行维护 | maintenance |
| [群组]里讨论了什么 | whatsapp |
| 寻找新技能创意 / 哪些技能流行 | skill-scout |
| 技能使用情况 / 技能统计 / 技能报告 | skill-analytics |
| 我在使用哪些技能 / 未使用的技能 | skill-analytics |
| 安全检查 / 检查漏洞 | self-monitor |
| 健康检查 / 检查基础设施 | self-monitor |
| 所有PA运行如何 / PA网络状态 | supervisor |
| מה הסטטוס / 状态如何 | supervisor |
| 我表现如何 / 审查我的表现 | eval |
| 运行评估 / מה עובד ומה לא | eval |
| 总结这个YouTube视频 | youtube-watcher |
| 给这个加注音 / 希伯来语元音符号 | hebrew-nikud |
| 压缩记忆 / 整理记忆 | memory-tiering |
决策树
这是什么类型的任务？
│
├─ 沟通/协调
│ ├─ 查找PA的联系方式 → ai-pa
│ ├─ 安排会议 → meetings
│ ├─ 总结会议记录 → meetings
│ └─ 向所有PA广播 → ai-pa
│
├─ 设置/接入
│ ├─ 从零开始新PA → pa-onboarding
│ ├─ 连接Google日历或Gmail → calendar-setup
│ └─ 连接monday.com → monday-for-agents
│
├─ 监控/健康
│ ├─ 检测到账单错误 → billing-monitor
│ ├─ 基础设施/安全检查 → self-monitor
│ └─ 同时检查所有PA → supervisor
│
├─ 日常运营
│ ├─ 晨间/晚间简报 → owner-briefing
│ ├─ monday.com看板任务 → monday-for-agents
│ ├─ 备份工作区或更新OpenClaw → maintenance
│ └─ WhatsApp对话上下文 → whatsapp
│
└─ 自我改进
├─ 所有者纠正了我 → self-learning
├─ 表现审查/审计 → eval
├─ 寻找新技能创意 → skill-scout
└─ 记忆压缩 → memory-tiering
完整技能库
| 技能 | 类别 | 使用时机 |
|---|
| ai-pa | 协调 | 查找PA联系方式、群组JID、协调协议 |
| billing-monitor |
健康 | 检测并响应API账单失败 |
|
calendar-setup | 设置 | 带写入权限的日历连接 + Gmail/邮箱设置 |
|
eval | 自我改进 | 全面表现审计——评分任务、检查PA健康、审查记忆 |
|
hebrew-nikud | 工具 | 为希伯来语文本添加元音符号，尤其用于TTS |
|
maintenance | 基础设施 | 工作区git备份（每6小时）+ OpenClaw更新（每周） |
|
meetings | 运营 | 安排会议并总结会议记录/转录 |
|
memory-tiering | 记忆 | HOT/WARM/COLD记忆压缩与归档 |
|
monday-for-agents | 运营 | 所有monday.com操作：API、MCP、看板、项目 |
|
owner-briefing | 运营 | 每日晨间/晚间摘要 |
|
pa-onboarding | 设置 | 从零开始完整设置新智能体 |
|
self-learning | 自我改进 | 记录纠正并应用教训；维护HOT.md |
|
self-monitor | 健康 | 基础设施+安全检查、磁盘/内存/服务健康 |
|
skill-master | 路由 | 选择正确技能（本文件） |
|
skill-scout | 发现 | 每周搜索新技能创意 |
|
supervisor | 运营 | 网络范围状态面板——所有PA、任务、系统健康 |
|
whatsapp | 记忆 | 按对话上下文、未回复追踪、循环预防 |
|
youtube-watcher | 工具 | 获取并总结YouTube视频转录 |
|
skill-analytics | 分析 | 追踪技能使用、生成每日报告、发现未使用技能 |
多技能工作流
某些任务需要按顺序使用多个技能：
新PA设置
pa-onboarding → calendar-setup → monday-for-agents → ai-pa（添加到目录）
PA网络健康检查
supervisor → billing-monitor（标记的PA）→ self-monitor（基础设施问题）
犯错之后
self-learning（记录）→ eval（更新评分）→ SOUL.md（如为模式则添加规则）
安排会议
ai-pa（查找对方PA的联系方式）→ meetings（协调+预订）
每周维护
whatsapp（每周摘要）→ owner-briefing（包含亮点）→ maintenance（推送到git）
重要群聊之后
whatsapp（记录决策）→ maintenance（推送到GitHub）
运行位置（复杂度指南）
内联运行（主会话）
- - ai-pa、billing-monitor、owner-briefing、supervisor、self-learning、maintenance
考虑使用子智能体处理繁重操作
- - calendar-setup、meetings（安排流程）、monday-for-agents（批量操作）
生成子智能体（推荐）
- - pa-onboarding（20+步骤）、eval（完整月度分析）、批量操作、skill-scout
模型指南
| 技能 | 最低模型要求 |
|---|---
