A vital agent behavior that forces AI coding assistants to consult official, live documentation before answering queries about libraries, frameworks, or APIs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install docs-first
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install docs-first using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Docs-First?
The Docs-First behavior is a critical mechanism designed to prevent AI hallucinations caused by stale training data. In modern software development, frameworks, CLI tools, SDKs, and libraries evolve at a rapid pace. A confident but outdated answer from an LLM can lead to hours of wasted debugging for deprecated flags or changed API signatures. By integrating this pattern into your AI agent workflows using Openclaw Skills, you ensure that the agent prioritizes live, authoritative sources over static memory.
This workflow mandates that whenever a user asks about installing, configuring, or calling specific software, the AI must fetch the official documentation or repository README before formulating its response. By making real-time search the default behavior for technical questions, Openclaw Skills empower developers with verified, up-to-date code and precise API references.
Docs-First Use Cases
- Querying active API signatures, method names, arguments, and return types for fast-evolving libraries.
- Finding specific configuration flags, environment variables, or schema specifications for tools like vLLM, Docker, or Kubernetes.
- Resolving version compatibility issues or verifying breaking changes introduced in recent software updates.
- Troubleshooting runtime errors, stack traces, or exception messages thrown by third-party packages.
- Guiding setup, installation, and deployment steps for hardware SDKs, CLI tools, or cloud runtimes.
How Docs-First Works
- Analyze User Request: The AI agent evaluates if the user's query references a specific third-party library, framework, SDK, or tool.
- Identify Target Fact: The agent determines the exact technical details needed, such as a function name, flag, or configuration parameter.
- Search Official Sources: The agent executes a web search, prioritizing the official hosted documentation portal (e.g., docs.project.io) or the official code repository.
- Fetch and Read Content: Rather than relying on search engine snippets, the agent retrieves the actual webpage or README contents to ensure thorough understanding.
- Draft and Verify: The agent compares the retrieved document with its internal knowledge, prioritizing the official documentation in case of discrepancies.
- Ground Output with Source Footnote: The response is presented in the user's language, complete with a clean, low-impact sub-text footer link directly pointing to the source documentation page.
Docs-First Setup
To enable this capability in your local development environment, make sure your agent is configured to use Openclaw Skills. You can clone and add the skill configuration directly to your project's custom system prompt directory:
# Navigate to your agent workspace
cd my-ai-agent-workspace
# Create a system prompts directory if it doesn't exist
mkdir -p custom_prompts/skills
# Copy the Docs-First markdown prompt into your active agent profiles
cp /path/to/docs-first-skill.md custom_prompts/skills/docs-first.md
Once added, register the skill in your agent configuration file (e.g., agent_config.json):
{
"skills": [
"docs-first"
]
}
Docs-First Data Schema & Taxonomy
The Docs-First skill does not require external database persistence but operates via structured systemic instructions and outputs. Below is the metadata taxonomy and formatting guidelines:
| Attribute | Description | Formatting Rule |
|---|---|---|
| Trigger Match | Softwares, libraries, SDKs, runtime environments, and CLI tools | Triggers on specific named entities; skips generic/conceptual queries |
| Authoritative Priority | Web-search target priority queue | 1. Official Documentation Portal 2. Official Code Repository (README, CHANGELOG) 3. General search (fallback only) |
| Output Footer Format | Hyperlink metadata referencing the exact source page | Formatted in HTML sub-text (<sub>[Anchor Text](URL)</sub>) |
Example Footer Output:
<sub>📄 [vLLM docs](https://docs.vllm.ai/en/stable/features/structured_outputs/)</sub>
Docs-First Advanced Features
- Context-Aware Triggering: Intelligently ignores purely conceptual questions (such as "what is a hash map") while aggressively stepping in on library-specific APIs.
- Multi-Language Adaptability: Searches global sources in the most authoritative language, while translating final explanations to match the user's active locale.
- Direct Git Repo Introspection: Automatically falls back to parsing raw repository markdown files, READMEs, or CHANGELOGs when official documentation portals are missing or offline.
- Fine-Grained Link Footers: Employs minimized sub-text formatting rules to maintain professional, clutter-free output while keeping source URLs accessible for instant verification within Openclaw Skills integrations.
SKILL.md
Loading
Related Openclaw Skills
Car Consult
An expert automotive advisory skill designed to guide buyers through evaluating, pricing, and selecting new energy vehicles.
Finance Daily Report
An automated AI agent skill that fetches real-time global market data and generates interactive, beautifully styled HTML financial daily reports.
h5-wechat-deploy
An automated workflow skill to build, deploy, and troubleshoot responsive single-page H5 marketing landing pages hosted on GitHub Pages and optimized for WeChat sharing.
IT Home Rank
A lightweight Python skill designed to fetch IT Home trending hot-lists (daily, weekly, monthly) and push them directly to messaging channels.
Google Social Media Finder
An automated search and intelligence tool designed to scan Google and locate social media profiles across dozens of platforms for any individual or brand.
Instagram Post Comments
An automated browser-driven skill to extract comments, user handles, timestamps, and reaction metrics from any Instagram post using cursor-based pagination.
Featured*
