chainreactors/cyber-harness

What it solves

Cyber-harness provides a unified runtime for cybersecurity agents. It allows developers to combine model reasoning, tool execution, knowledge, and collaboration into a single extension-based system, simplifying the creation of automated security testing and research tools.

How it works

The system uses an "everything-is-an-extension" architecture where capabilities like reasoning, tool execution, and observation are assembled via a common lifecycle. It offers two primary modes of operation: an agent mode where the LLM chooses tool calls, and a scan pipeline mode that follows rules and events with optional AI verification stages.

Who it’s for

It is designed for security researchers and developers building Go-based applications that require autonomous or semi-autonomous cybersecurity capabilities.

Highlights

• Modular Extension System: All core capabilities are treated as extensions, allowing for easy composition of custom distributions.
• Flexible Execution: Supports both LLM-driven agentic behavior and deterministic rule-based scanning.
• Full Distribution: Includes a Web workbench, browser automation, passive reconnaissance, and deep crawling via the aiscan-full package.
• Broad Model Support: Compatible with providers like OpenAI and Anthropic (and DeepSeek via OpenAI-compatible APIs).