DeepthinkLite is a local-first research tool that generates durable markdown artifacts and enforces a time-budgeted workflow for deep thinking.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install deepthinklite
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 deepthinklite 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 DeepthinkLite?
DeepthinkLite provides a repeatable framework for deep research, inspired by high-end deep thinking workflows. Unlike ephemeral chat sessions, it produces permanent, version-controlled artifacts including a detailed investigation map and a structured final response. This skill ensures that complex queries are handled with technical depth and clarity while keeping the user in control of the research process.
By leveraging the Openclaw Skills architecture, DeepthinkLite prioritizes security and local-first data. It implements a strict time budget contract and treats all external information as untrusted data, making it an ideal choice for developers who need reliable, prompt-injection resistant research capabilities directly within their local development environment.
DeepthinkLite Use Cases
- Researching complex technical documentation or recent library changelogs.
- Deep code inspection and architectural analysis within local repositories.
- Generating decision-ready reports from multiple untrusted web sources.
- Creating a version-controlled knowledge base of deep-dive technical investigations.
How DeepthinkLite Works
- Phase 0 - Frame the ask: The agent restates the request and defines success criteria to ensure total alignment.
- Phase 1 - Generate questions.md: A high-leverage investigation map is created, outlining specific questions and intended sources like local docs or the web.
- Phase 2 - Research: The agent collects evidence using available tools, prioritizing primary sources while maintaining a security sandbox.
- Phase 3 - Write response.md: A structured final answer is generated, including reasoning, recommendations, and explicit references.
DeepthinkLite Setup
To get started with DeepthinkLite, use the following CLI commands to initialize a research query. You can choose between raw source snippets or a summary-only mode for tighter security through Openclaw Skills.
# Standard query with raw source snippets
deepthinklite query "<your deep research question>" --out ./deepthinklite --source-mode raw
# Secure mode with summaries only
deepthinklite query "<your deep research question>" --out ./deepthinklite --source-mode summary-only
DeepthinkLite Data Schema & Taxonomy
DeepthinkLite organizes every research run into a specific slug-based directory structure containing the following files:
| File | Description |
|---|---|
questions.md |
The investigation map containing high-leverage questions and the plan. |
response.md |
The final structured answer, reasoning summary, and references. |
meta.json |
Technical metadata regarding the specific run configuration. |
All files are generated in standard Markdown or JSON for easy version control and diffing.
DeepthinkLite Advanced Features
- Time budget enforcement (10 to 60 minutes) to ensure high-quality, non-shallow output.
- Prompt-injection resistant handling of untrusted data from web and local sources.
- Explicit permission prompts before the agent accesses sensitive local paths or external URLs.
- Dual source modes for flexible content retrieval (raw vs. summary-only).
- Seamless integration with other Openclaw Skills and local-first tools like RAGLite.
SKILL.md
Loading
Related Openclaw Skills
Anakin Web Data Extraction
A powerful CLI tool for converting websites into clean, structured data using AI-driven scraping and autonomous research agents.
Usage Visualizer
An advanced analytics engine that transforms raw session logs into professional visual reports and text summaries with 100% local processing.
Calorie Visualizer
A local-first nutrition tracking skill that logs meals and automatically generates visual progress reports.
Web2Labs Studio
An AI-powered video automation suite that converts raw footage and URLs into polished social media content, subtitles, and thumbnails.
raglite
A local-first RAG cache that distills documents into structured Markdown and indexes them for hybrid vector and keyword retrieval.
RAGLite
A local-first RAG cache that distills documents into structured Markdown and indexes them for hybrid vector and keyword search.
Featured*
