> For the complete documentation index, see [llms.txt](https://shepherd-1.gitbook.io/shepherd/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://shepherd-1.gitbook.io/shepherd/our-approach.md).

# Our Approach

Shepherd continuously stress-tests smart contracts in realistic environments. Our multi-agent architecture allows our agents to:

* Simulate real attack behavior across every possible execution path.
* Discover edge cases that traditional methods can miss.
* Evolve over time, using memory and retrieval to improve its tactics and learn new attack patterns.

Our agents don't just check for known vulnerabilities, they trigger on-chain state changes. This allows them to autonomously hypothesize, simulate, and refine potential exploits, effectively replicating how a malicious actor may probe a system.  

Shepherd closes security gaps by:

* Proactively surfacing vulnerabilities before they are exploited.
* Running adversarial simulations across composable systems.
* Evolving its knowledge base with every test cycle.
* Expand coverage across all protocol layers.

Teams can ship faster, safer, and with confidence that their systems are stress-tested against the threats of today and tomorrow.