Stroq — AI for General (Professional Services)

As AI coding agents become more autonomous, they frequently process external text from issues, pull requests, and documentation that can contain indirect prompt injections or malicious instructions. Stroq addresses this vulnerability by acting as a security gate for agent actions. Instead of relying on the AI model itself to judge safety, Stroq uses a deterministic policy engine that runs locally with zero telemetry or network calls in the hot path. It normalizes text inputs, decodes hidden encodings, and traces the provenance of commands back to their source tool outputs. If an agent attempts to execute a command carrying a known secret variable, Stroq blocks the outbound call, identifying the specific variable and file without exposing the underlying value. Additionally, you can replay 36 known attack scenarios against your custom policies to test your security posture on every commit. While it provides critical visibility and protection against agentjacking and credential theft, it requires manual setup of local CLI tools and configuration of YAML policies to fit your specific development workflows.