A hybrid search tool querying local Openclaw documentation, skills, and wikis using BM25, Vector search, and LLM re-ranking.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install qmd-knowledge-search
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 qmd-knowledge-search 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 QMD Search?
The QMD Search skill empowers your AI agent to query local knowledge bases, including documentation, skills, and internal wikis located at /root/clawd/knowledge. By integrating the QMD hybrid search engine, it combines the precision of BM25 keyword matching, the conceptual depth of vector search, and the advanced contextual awareness of LLM re-ranking to deliver highly relevant technical answers.
Adding this capability to your agent's toolkit enables self-referential querying of available Openclaw Skills, helping developers and autonomous agents quickly locate system capabilities, internal system paths, or API references without leaving the CLI environment.
QMD Search Use Cases
- Querying System Documentation: Rapidly search through local openclaw-docs to understand the agent's environment, APIs, or operational limits.
- Discovering Agent Capabilities: Look up definition files of installed Openclaw Skills to understand how to call specific tools or commands.
- Technical Troubleshooting: Query error messages or configuration parameters against internal wikis and developer guidelines.
- Retrieving Full Specifications: Fetch entire technical documents directly to use as context for code generation or complex agent decision-making.
How QMD Search Works
- Triggering Search: The user or the AI agent executes a query using either the hybrid or keyword-based search command.
- Retrieval: The tool scans local folders in
/root/clawd/knowledge(such asopenclaw-docs/,skills/, andqmd.md). - Hybrid Processing: It utilizes BM25 for keyword scoring, Vector embeddings for semantic search, and an LLM re-ranking layer to prioritize the most accurate results.
- Filtering and Selection: The tool applies score thresholds (like
--min-score 0.5) to eliminate irrelevant noise. - Content Delivery: The agent retrieves the full text of selected documents to integrate into its active reasoning context.
QMD Search Setup
Installation and Configuration
Ensure that your local knowledge base directories exist and contain relevant document types at /root/clawd/knowledge.
Commands
Run the hybrid search query with:
qmd query "your query here"
For faster, keyword-only lookups:
qmd search "exact phrase"
To fetch a specific file's full contents:
qmd get "path/to/doc.md" --full
To clean up noise and retrieve structured data for programmatic processing:
qmd query "your query here" --json --min-score 0.5
QMD Search Data Schema & Taxonomy
Directory & File Schema
All searchable documentation and metadata are hosted locally within the knowledge root:
| Path Component | Description | Data Format |
|---|---|---|
/root/clawd/knowledge |
Knowledge base root directory | Folder structure |
openclaw-docs/ |
System architecture, setup guides, and APIs | Markdown (.md) files |
skills/ |
Detailed definitions of installed Openclaw Skills | Markdown/YAML files |
qmd.md |
Specific usage docs and schema parameters for QMD | Markdown (.md) file |
Output Format Options
- Standard Console Output: Human-readable highlighted matches with relevance scoring.
- Structured JSON (
--json): Ideal for programmatic parsing by tools and AI agents to extract exact document IDs, file paths, relevance scores, and matched snippets.
QMD Search Advanced Features
- Hybrid BM25 & Vector Search: Combines literal keyword matching with semantic understanding for maximum coverage.
- LLM-Based Re-ranking: Elevates top results based on actual semantic query alignment rather than simple string frequency.
- Configurable Score Filtering: Use the
--min-scoreflag to dynamically filter low-quality results. - Structured JSON Mode: Supports
--jsonparameters for seamless integration with downstream data pipelines and automated tools.
SKILL.md
Loading
Related Openclaw Skills
thesis-format-zh
An automated formatting checker and corrector designed to align Chinese academic papers with official university template requirements.
Inbox Manager
An automated email triage and management skill that supports multiple Gmail accounts using a single Google Cloud project and secure OAuth tokens.
Dashboard Generator
An automated system to generate and serve lightweight, self-contained HTML dashboards for tracking AI agent memory logs, financial ledger expenses, and multi-source token usage.
News Curator
An automated news curation agent that fetches, filters, deduplicates, and delivers highly-focused daily finance and AI/tech briefings to Telegram via cron triggers.
Image Tools
A comprehensive, non-destructive image editing toolkit powered by ImageMagick and packaged for automated AI development workflows.
GEO Cycle Autopilot
An automated workflow engine for QClaw that executes scheduled GEO optimization tasks, deep article imitations, and template exports natively in the background.
Featured*
