An intelligent Q&A skill that leverages team overview documents to guide and ground external web search queries.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install kb-web-query
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 kb-web-query 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 kb_web_query?
The kb_web_query skill bridges the gap between static internal knowledge bases and dynamic internet research within the ecosystem of Openclaw Skills. Unlike general web search clients, it first analyzes your team's contextual overview documents—such as research maps, roadmaps, and open questions. This process creates a detailed profile of the team's current focus area, allowing the agent to formulate precise, highly relevant search terms.\n\nOnce external information is gathered, the skill synthesizes the findings to directly address the user's query while framing findings within your team's real-world environment. It provides a structured output highlighting external citations and team internal references, all without persisting the conversational data back to your permanent knowledge base.
kb_web_query Use Cases
- Resolving deep technical research questions that require current external data while aligning with the team's existing technology stack.\n- Investigating open technical questions listed in a team's roadmap without polluting the permanent knowledge repository with unverified search scraps.\n- Synthesizing recently published academic papers, industry standards, or vendor documentation in direct relation to active project streams.
How kb_web_query Works
- Initialize Team Context: The workflow starts by executing a preparation script to parse six core team overview files, including
team-overview.md,research-map.md, androadmap.md. This step establishes a team profile containing current objectives, gaps, and technical directions.\n2. Formulate Search Queries: Guided by both the user query and the established team context, the system constructs a precise web search strategy, preferring authoritative sources like academic papers, official specifications, and documentation.\n3. Execute Search and Fetch: Uses available web-searching tools to query the internet. If specialized search endpoints are configured, it runs targeted queries; otherwise, it employs fallbacks to gather web resources.\n4. Optional KB Fetch: If the baseline overview context is insufficient, the agent can optionally select and parse supplementary internal knowledge base pages to resolve ambiguities.\n5. Synthesize Findings: Integrates external data with internal constraints, drafting a unified answer where external facts are mapped to verified team contexts.\n6. Validate & Apply: Runs an validation task to normalize citation formats, filter invalid URLs, append a formatted source reference section, and output the final response payload.
kb_web_query Setup
To deploy this skill within your Openclaw Skills directory, configure the necessary environmental variables and run the core execution scripts.\n\n### Environment Variables\nConfigure your execution environment with access to your knowledge base repository and search providers:\nbash\nexport GITEA_URL="https://your-gitea-instance.com"\nexport GITEA_BOT_TOKEN="your-gitea-token"\nexport GITEA_BOT_USERNAME="claw-bot"\nexport GITEA_ORG="research-group"\nexport TEAM_KB_REPO="team-knowledge-base"\nexport OPENCLAW_SHARED_DIR="/path/to/shared/dir"\nexport TAVILY_API_KEY="your-tavily-key" # Optional for enhanced search\n\n\n### Step-by-Step Execution\n\n1. Prepare Team Context\nRun the preparation script to build your context profile:\nbash\npython3 scripts/run_task.py prepare --input payload.json --context-output context.json\n\n\n2. **Fetch Supplementary Pages (Optional)**\nIf additional background files are required, call the fetch stage with your page selection:\nbash\npython3 scripts/run_task.py fetch --input payload.json --context context.json --selection page-selection.json --evidence-output evidence.json\n\n\n3. Apply and Generate Final Results\nOnce the agent generates the unified answer payload, apply the references and write the final result JSON:\nbash\npython3 scripts/run_task.py apply --input payload.json --context context.json --evidence evidence.json --answer answer.json\n
kb_web_query Data Schema & Taxonomy
The skill manages its context, intermediate outputs, and results through structured files in the designated shared directory. Below is the structure of the primary output JSON (answer.json):\n\n| Field | Type | Description |\n| :--- | :--- | :--- |\n| answer | String | The Markdown-formatted final response addressing the user's query. |\n| teamSources | Array | Objects mapping the path, title, and contextual role of internal files utilized. |\n| webSources | Array | External metadata references, including publisher, author, URL, and accessed date. |\n| usedSearchQueries | Array | List of raw query strings sent to search engines during execution. |\n| highValue | Boolean | Declares whether the answer is of exceptional high value (defaults to false for web queries). |\n| qa | Object | Control directives indicating whether to write back to the KB repository (e.g., {"write": false}). |\n\n### Expected Citation Format\nEvery response dynamically appends a reference block structured as follows:\nmarkdown\n## References\n\n### Team Context\n- Team Overview (overview/team-overview.md)\n\n### External Sources\n- [paper] [Paper Title](https://example.com/item) — publisher; date; accessedAt\n
kb_web_query Advanced Features
- Targeted Context Grounding: Automatically limits search intent based on six specific system overview files to eliminate generic, out-of-context search answers.\n- Robust Fallback Strategy: Automatically degrades gracefully from API-based search tools to native HTTP request procedures when primary APIs are unavailable.\n- Validation Engine: Sanitizes outbound references, systematically filtering out local file paths, raw attachment references, and non-HTTP protocols from external citations.\n- Zero Persistence Guarantee: Specifically bypasses repository modifications, ensuring conversational search cycles remain ephemeral without altering the permanent QA knowledge database.
SKILL.md
Loading
Related Openclaw Skills
X (Twitter) Auto-Posting & Content Operations
An automated X (Twitter) content operations workflow that takes target keywords from research to published, tracked posts within a real browser session.
Obsidian Best Practices & Master Resources
A curated knowledge-base skill that helps users find the best Obsidian guides, PKM methods, and workflow resources fast.
Terraform State Inventory
A powerful DevOps automation tool designed to extract managed infrastructure metadata, map dependencies, and catch deployment drift straight from your state files.
Shieldz Crypto Payments
A keyless and non-custodial payment tool that generates crypto payment links or reusable tip jars directly from user-controlled wallets.
StablePay Agent Bootstrap
A dedicated setup skill to bootstrap, validate, and integrate the StablePay MCP server and CLI utility for AI coding agents.
Scrapfly Cloud Browser
Automate cloud browsers using the Scrapfly Cloud Browser API with Python Playwright for residential proxy rotation and anti-bot fingerprint management.
Featured*
