Security

Developer documentation is a valuable resource for coding agents, but it also introduces a potential attack surface. Agents may use retrieved documentation to generate code, execute commands, and make changes to projects.

POSH uses multiple layers of protection to reduce the risk of prompt injection and other malicious content reaching coding agents.

Our security measures

Clear trust boundaries. The POSH MCP server explicitly identifies retrieved documentation as data, not instructions for the agent to follow.
Isolated builds. Documentation is processed in isolated environments to limit the impact of potentially malicious dependencies or build steps.
Static content analysis. We check for suspicious patterns, including invisible text and delimiter spoofing, that could be used to manipulate an LLM.
Prompt injection detection. Specialized models analyze documentation for potentially malicious instructions.
Supply-chain monitoring. When trusted third-party security sources flag a dependency as problematic, we remove affected assets.

Limitations

No prompt injection detection system is perfect. Our protections are designed to reduce risk, not eliminate it. Coding agents should continue to treat external documentation as untrusted data and apply appropriate restrictions to tool execution and access to sensitive resources.

We continuously review and improve our security measures.

To report a security concern, contact security@poshtui.com.