بنقرة واحدة
openclaw-config
OpenClaw 配置文件(openclaw.json)的安全修改规则、字段参考和错误学习记录。 在修改任何 OpenClaw 配置前必须先读此文件和 lessons.md。
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
القائمة
OpenClaw 配置文件(openclaw.json)的安全修改规则、字段参考和错误学习记录。 在修改任何 OpenClaw 配置前必须先读此文件和 lessons.md。
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Full-stack development workflow skill. Guides the complete cycle from understanding requirements to shipping code via PR with Preview validation. USE FOR: - Bug fixes, feature development, UI changes - "修复这个 bug", "帮我实现这个功能", "改一下这个页面" - "fix this", "implement", "build", "add feature", "change the UI" - Any code modification that needs Dev Server → PR → Preview → Merge flow REQUIRES: git, gh (GitHub CLI), pnpm/npm, agent-browser
Apple Silicon local ASR via Qwen3-ASR + MLX. Server-mode for stable memory. USE FOR: - "转文字", "转录", "transcribe this", "speech to text" - 任何音频/视频文件需要转文字时 - SRT/VTT 字幕生成 - 用户发音频/视频文件说"这个说的什么" - douyin-dl 下载后自动转文字 - Agent 需要处理语音内容时 ⚠️ ALWAYS use this skill for any audio/video transcription on Apple Silicon. Do NOT use whisper-transcribe (disabled). Apple Silicon only. No API key needed. No cloud dependency. Uses server-mode (model loaded once, ~4GB memory stable).
Convert Markdown to high-quality PNG/JPG images with zero whitespace, syntax highlighting (Shiki), and optional KaTeX math support. Uses Playwright (Chromium) for rendering with GitHub-style default theme. USE FOR: - Any request to render Markdown as an image - Code snippet screenshots with syntax highlighting - Markdown with math formulas (use --math flag) - Generating styled images for Discord/Telegram/Slack - "markdown转图片", "把这段md变成图片", "render markdown as image", "screenshot this code" - Converting README, changelogs, or docs to shareable images REQUIRES: md2img (npm global), playwright-core, chromium browser
Unified self-reflection and self-improvement. Three-layer memory: Golden Rules (SOUL.md) → Active Lessons (self-review.md) → Raw Log (daily). Auto-captures corrections, reflects at session end, promotes patterns that work. USE FOR: - User corrects agent behavior → auto-log and learn - Session end → "land the plane" reflection - Heartbeat timer → periodic self-check - "what have you learned", "show your lessons", "memory stats" - "remember this rule", "promote this pattern", "demote this" - Agent notices own mistake → self-log REPLACES: self-improving, self-reflection (merged into this skill) COMPANION: soul-keeper (independent — pushes lessons to workspace files) REQUIRES: jq, date
Transcribe audio/video to text using the best available ASR backend. Backend priority: MLX Qwen3-ASR (local, best for Chinese) → Groq API (fast, cloud) → whisper.cpp (local fallback) USE FOR: - "转文字", "转录", "transcribe this", "speech to text" - Any audio file (wav, mp3, m4a, flac, ogg, wma, aac) needs transcription - Video file needs transcription (extracts audio first via ffmpeg) - Douyin/video downloaded → auto transcribe - Need SRT/VTT subtitles from audio/video - "这个视频说的什么", "听一下这个音频" REPLACES: openai-whisper, openai-whisper-api, mlx-whisper, douyin-transcript (for ASR step) REQUIRES: ffmpeg OPTIONAL: mlx-qwen3-asr (pip), GROQ_API_KEY, whisper-cli (brew)
系统化研究 skill — 螺旋收敛模型。把任何问题(模糊或清晰)分解成子问题,迭代搜索,越搜越清晰,直到每个子问题都有答案。 Triggers: "帮我研究", "研究一下", "调研", "分析对比", "research", "investigate", "look up" 也触发: 竞品分析、市场调研、技术选型对比、趋势了解 Does NOT trigger: - 已经知道答案的简单事实问题 - 用户直接给了 URL 要抓取的 - 代码调试、写代码任务 Output: 结构化研究报告(结论 + 子问题答案 + 来源 + 争议点 + 未解决缺口)
| user-invocable | false |
| name | openclaw-config |
| description | OpenClaw 配置文件(openclaw.json)的安全修改规则、字段参考和错误学习记录。 在修改任何 OpenClaw 配置前必须先读此文件和 lessons.md。 |
| triggers | ["config","openclaw.json","配置","改配置","config.patch","添加agent","添加channel","修改模型","gateway配置"] |
| metadata | {"clawdbot":{"emoji":"⚙️"}} |
~/.openclaw/openclaw.jsongateway config.patch 工具修改配置(内置校验,格式错直接拒绝)lessons.md,检查是否有相关的历史教训lessons.md1. 读本文件(规则)
2. 读 lessons.md(历史教训)
3. 告诉用户打算改什么 → 等确认
4. 用 gateway config.patch 执行修改
5. 如果成功 → 告诉用户改了什么
6. 如果失败 → 记录到 lessons.md → 修复 → 再试
| 陷阱 | 说明 |
|---|---|
tools 是对象不是数组 | { "tools": { "profile": "coding" } } |
tools.allow 会覆盖全部工具 | 用 tools.alsoAllow 追加 |
agents.list 是数组 | [{ "id": "xxx", ... }] 不是 { "xxx": {...} } |
allowFrom 必须是字符串数组 | ["6883367773"] 不是 "6883367773" |
.strict() 模式 | 任何未知字段直接拒绝启动 |
dmPolicy="open" 需要 allowFrom: ["*"] | 否则全部 DM 被丢弃 |
gateway.auth.mode | 有 token + password 时必须显式指定 |
heartbeat.every | 必须是时长字符串如 "30m"、"1h" |
{
"id": "new-agent", // string,必填,kebab-case
"workspace": "/path/to/ws", // string,必填
"model": {
"primary": "anthropic/claude-sonnet-4-6" // string
},
"tools": {
"deny": ["gateway", "cron"], // string[],可选
"elevated": { "enabled": false } // object,可选
},
"subagents": {
"allowAgents": ["researcher"] // string[],可选
}
}
{
"agentId": "new-agent", // 必须匹配 agents.list 中的 id
"match": {
"channel": "discord", // telegram | discord | whatsapp | slack
"accountId": "new-agent" // 必须匹配 channels.<provider>.accounts 中的 key
}
}
{
"name": "Display Name",
"enabled": true,
"token": "BOT_TOKEN",
"groupPolicy": "allowlist",
"streaming": "off",
"guilds": {
"GUILD_ID": {
"channels": {
"CHANNEL_ID": {
"allow": true,
"requireMention": false
}
}
}
}
}
{
"providerName": {
"baseUrl": "https://api.example.com/v1",
"api": "openai-completions",
"models": [{
"id": "model-id",
"name": "Display Name",
"reasoning": true,
"input": ["text"],
"cost": { "input": 0, "output": 0, "cacheRead": 0, "cacheWrite": 0 },
"contextWindow": 128000,
"maxTokens": 8192
}]
}
}
需要更多信息时读 references 目录:
references/field-reference.md — 完整字段手册references/common-errors.md — 社区 TOP 10 错误 + 修复references/examples.md — 5 个真实配置例子references/schema-rules.md — 校验机制详解