OpenClaw OSINT Skill (Unified Architecture)

A decoupled, scalable OSINT framework for OpenClaw using an intelligent Director-Worker architecture to separate massive data scraping from high-level reasoning.

Architecture: One Command, Two Agents

This skill abstracts complex multi-model pipelines into a single command.

osint (The Unified Entry Point)

• Trigger: /osint <target> [optional question]
• The Flow:
  1. The Director (Pro Model): OpenClaw spawns a high-reasoning "OSINT Director" agent. It ingests your question and checks if a raw data lake exists for the target.
  2. The Worker (Flash Model): If data is missing, the Director automatically spawns its own "Harvester" subagent using a fast, low-cost model (gemini-3-flash-preview). The Harvester runs deep web scrapes, pivoting through emails and aliases, dumping everything into a massive Markdown file (reports/<target>_raw_data.md).
  3. The Synthesis: The Director wakes back up, reads the massive raw data lake, cross-references timelines and bios, and synthesizes a precise, cited intelligence brief for the user.

Installation

Link the skill into your OpenClaw workspace:

openclaw skills install path/to/osint-stalker-repo/osint

Example Workflow

1. User: "What is the last known location of lidorshimoni?"
2. Orchestrator (osint-director): Checks if reports/lidorshimoni_raw_data.md exists. It doesn't.
3. Orchestrator: Spawns osint-harvester subagent (Gemini Flash).
4. Harvester: Enumerates accounts, scrapes GitHub bios, LinkedIn posts, and personal websites. Dumps raw markdown into the Data Lake.
5. Orchestrator: Re-awakens, ingests the massive markdown file, and synthesizes the answer.
6. Orchestrator: Replies to the user: "Based on a scraped GitHub bio (github.com/lidorshimoni) and a Hackaday project post (hackaday.io/...), the last known location is Kiryat Gat, Israel."