A sophisticated deep-research skill that utilizes Codex CLI to perform multi-source synthesis and high-context analysis for complex queries.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install codex-deep-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 codex-deep-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 Codex Deep Search?
The Codex Deep Search skill is a specialized tool within the Openclaw Skills ecosystem designed for tasks where standard web search APIs provide insufficient depth. It excels at handling complex, niche, or multi-faceted queries that require synthesizing information from various sources. By leveraging Codex CLI, this skill allows AI agents to cross-reference multiple links and generate comprehensive reports that go far beyond simple factual snippets.
Built for reliability and depth, this skill is the primary choice for users who need thorough research, detailed analysis, or background intelligence gathering. It integrates seamlessly into professional workflows by providing both synchronous execution for quick synthesis and a background dispatch mode for intensive research projects, making it a versatile asset for any Openclaw Skills deployment.
Codex Deep Search Use Cases
- Conducting exhaustive research on niche technical topics that require multi-source synthesis.
- Generating comprehensive reports when standard search engine results are too shallow.
- Performing deep-dive analysis on complex subjects using specific triggers like deep search or detailed research.
- Running long-duration research tasks in the background with automated notifications via Telegram.
How Codex Deep Search Works
- The AI agent identifies a request requiring deep research and invokes the Codex Deep Search skill.
- The skill chooses between Synchronous Mode (for short queries) or Dispatch Mode (for background execution).
- In Dispatch Mode, the process runs as a background bash script, allowing the agent to remain responsive while research continues.
- The system performs incremental writes, saving data after each search round to ensure persistence against memory issues or timeouts.
- Upon completion, the synthesized report is saved to a markdown file, and metadata is updated in a JSON file for easy tracking.
- For background tasks, a callback is sent to a designated Telegram group to notify the user of the completed research.
Codex Deep Search Setup
To deploy this skill within your environment, ensure you have the Codex CLI installed. The skill is executed via a primary bash script. For basic synchronous usage, use the following command:
bash /home/ubuntu/clawd/skills/codex-deep-search/scripts/search.sh \
--prompt "Your research query" \
--output "/tmp/search-result.md" \
--timeout 60
For advanced research using the dispatch pattern, run:
nohup bash /home/ubuntu/clawd/skills/codex-deep-search/scripts/search.sh \
--prompt "Complex multi-source query" \
--task-name "deep-research-01" \
--telegram-group "-5006066016" \
--timeout 120 > /tmp/codex-search.log 2>&1 &
Codex Deep Search Data Schema & Taxonomy
The skill maintains a structured data hierarchy within the Openclaw Skills framework to ensure research transparency and persistence:
| File Name | Description |
|---|---|
data/codex-search-results/<task>.md |
The final, incrementally-written research report in Markdown format. |
data/codex-search-results/latest-meta.json |
A JSON object containing task metadata, status, and execution identifiers. |
data/codex-search-results/task-output.txt |
The raw, unformatted output generated by the Codex CLI process. |
This schema allows for easy auditing of research steps and recovery of data even if a process is terminated prematurely.
Codex Deep Search Advanced Features
- Background Dispatch Pattern: Executes intensive research tasks as detached processes, preventing agent blocking and improving UX.
- Incremental Data Persistence: Saves search results round-by-round to survive Out-Of-Memory (OOM) errors or script timeouts.
- Low Reasoning Effort Mode: Specifically optimized to minimize memory footprint and prevent SIGKILL events during high-intensity data gathering.
- Telegram Callback Integration: Automatically pushes completion notifications and results to a specified chat ID for real-time updates.
- Timeout Protection: Includes configurable auto-stop mechanisms to manage resource consumption for runaway search tasks.
SKILL.md
Loading
Related Openclaw Skills
Aviation Healthcheck
A comprehensive monitoring tool for tracking aviation regulatory updates and Openclaw Skills environment health.
Voice Reply
A specialized skill that enables AI agents to automatically convert text responses into spoken audio messages across popular messaging platforms.
Bilibili Update Viewer
A Bilibili monitoring utility to track creator updates, new video releases, and social dynamics in real-time.
Jina.ai Reader
A powerful utility to fetch clean, AI-optimized Markdown content from any website, including paywalled and JavaScript-heavy pages.
Deep Search
A sophisticated research skill that performs iterative search queries and LLM synthesis to provide exhaustive, multi-source analysis.
last30days
A high-speed research engine that scrapes Reddit, X, and the web to synthesize real-time insights and generate precision AI prompts.
Featured*
