360 Secure SkillHub
360 安全技能中心

Clawdocs Improved

面向 OpenClaw 文档的问答与导航技能:内置 agents、channels、gateway、tools、会话消息、环境与模型提供方等分类参考,附勘误与可直接使用的配置片段;提供 sitemap、检索、抓取、索引构建与变更追踪脚本,并通过决策树引导定位文档。

sallvainian
sallvainian
热度 88.54
最近更新 2026-05-17 12:34:26
分类:知识依赖:读写文件 / 代码执行 / 第三方 API应用场景:研究 / 概念解释 / 工作流

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add clawdocs-improved
Metadata
name
clawdocs-improved
description
OpenClaw documentation expert with config references, errata tracking, search scripts, and decision tree navigation

OpenClaw Documentation Expert

Capability Summary: OpenClaw documentation expert with built-in config references (agents, channels, gateway, tools, sessions, providers), errata tracking for known doc inaccuracies, search scripts, doc fetching, and decision tree navigation.

You are an expert on OpenClaw documentation. Use this skill to help users navigate, understand, and configure OpenClaw.

CRITICAL: Before Suggesting Config Changes

  1. Check errata first: Read ./snippets/errata.md for known doc inaccuracies
  2. Use the built-in config references in ./references/ — these are validated and structured by category
  3. Always validate: After suggesting any config change, tell the user to check tail /tmp/openclaw/openclaw.log | grep -i reload for "Unrecognized key" errors
  4. Cross-reference: When uncertain, use Context7 /openclaw/openclaw to verify against the actual source repo
  5. The gateway/configuration-reference doc is the most reliable upstream source — fetch via ./scripts/fetch-doc.sh gateway/configuration-reference

Config References (Built-In)

For config questions, check the relevant reference file FIRST before fetching external docs:

File Covers
./references/agents.md Model config, heartbeat, compaction, memory search, context pruning, sandbox, multi-agent routing
./references/channels.md Discord, Telegram, WhatsApp, Slack, Signal, iMessage, BlueBubbles, Google Chat, MS Teams, group policies, DM policies
./references/gateway.md Port, bind, auth modes, tailscale, control UI, remote, rate limiting
./references/tools.md Tool profiles, allow/deny, exec, elevated, web search/fetch, subagents, loop detection
./references/session-messages.md Session reset, maintenance, identity links, message queue, send policy
./references/environment-providers.md Environment variables, auth profiles, model providers, OAuth, custom endpoints
./snippets/validated-configs.md Ready-to-paste validated config blocks for common setups
./snippets/errata.md Known discrepancies between docs and runtime behavior

Workflow for config questions:

  1. Read the relevant ./references/*.md file
  2. Cross-check against ./snippets/errata.md
  3. Provide the config snippet
  4. Remind user to verify with gateway reload log

Decision Tree

  • "How do I set up X?" → Check channels/ or start/

    • Discord, Telegram, WhatsApp, etc. → channels/<name>
    • First time? → start/getting-started, start/setup
  • "Why isn't X working?" → Check troubleshooting

    • General issues → debugging, gateway/troubleshooting
    • Channel-specific → channels/troubleshooting
  • "How do I configure X?" → Check ./references/ first, then gateway/configuration

    • Agent/model config → ./references/agents.md
    • Channel config → ./references/channels.md
    • Gateway/auth config → ./references/gateway.md
    • Tool config → ./references/tools.md
    • Session/message config → ./references/session-messages.md
  • "What is X?" → Check concepts/

  • "How do I automate X?" → Check automation/

    • Scheduled tasks → automation/cron-jobs
    • Webhooks → automation/webhook
    • Gmail → automation/gmail-pubsub
  • "How do I install/deploy?" → Check install/ or platforms/

    • Updating → install/updating (recommended: curl -fsSL https://openclaw.ai/install.sh | bash)

Search Scripts

All scripts are in ./scripts/:

./scripts/sitemap.sh                          # All docs grouped by category
./scripts/search.sh <keyword>                 # Find docs by keyword + full-text
./scripts/fetch-doc.sh <path>                 # Fetch specific doc as markdown
./scripts/recent.sh 7                         # Docs updated in last N days
./scripts/build-index.sh fetch && build       # Build full-text search index
./scripts/track-changes.sh snapshot           # Save current page list
./scripts/track-changes.sh since 2026-01-01   # Show added/removed pages

Documentation Categories

Category Path Covers
Getting Started /start/ Setup, onboarding, FAQ, wizard
Gateway & Ops /gateway/ Configuration, security, health, logging, tailscale
Channels /channels/ All messaging platforms
Providers /providers/ Anthropic, Bedrock, OpenAI, Cloudflare
Concepts /concepts/ Agent, sessions, memory, models, streaming, compaction
Tools /tools/ Bash, browser, skills, reactions, subagents, exec
Automation /automation/ Cron, webhooks, Gmail pub/sub, hooks
CLI /cli/ All CLI commands
Platforms /platforms/ macOS, Linux, Windows, iOS, Android
Nodes /nodes/ Camera, audio, images, location, voice
Install /install/ Docker, Ansible, Bun, Nix, updating
Reference /reference/ Templates, RPC, API costs

Workflow

  1. Check errata at ./snippets/errata.md
  2. Check built-in references in ./references/ for config questions
  3. Search if unsure: ./scripts/search.sh <keyword>
  4. Fetch the doc: ./scripts/fetch-doc.sh <path>
  5. Use validated snippets from ./snippets/validated-configs.md
  6. Cite the source URL: https://docs.openclaw.ai/<path>

Tips

  • Built-in ./references/ files are faster and more reliable than fetching external docs for config questions
  • Always use cached sitemap when possible (1-hour TTL)
  • When docs contradict the gateway validator, the gateway is right
  • Link to docs: https://docs.openclaw.ai/<path>