Securabbit

• 

  Securrabit opener

• 

  Securrabit dashboard (Streamlit)

• 

  Live terminal analysis of swarm attack

• 

  Results dashboard along with sentry tracing

Inspiration:

Securing AI agents against adversarial exploits like prompt injections and PII leaks.

What it does:

SecuRabbit Swarm performs parallel red-teaming of AI agents using a multi-agent "hacker swarm" in isolated environments.

How we built it:

We used Daytona for sandbox orchestration, Sentry for real-time security monitoring, and CodeRabbit to harden code based on detected vulnerabilities.

Challenges:

Synchronizing parallel sandbox states and parsing non-deterministic LLM evaluation verdicts.

Accomplishments:

An automated, end-to-end security pipeline that identifies and proposes fixes for vulnerabilities in minutes.

What we learned:

Isolated sandboxing is critical for safely executing adversarial tests without risking production data.

What's next:

Expanding our "Attack Zone" categories and building a deeper CI/CD loop with CodeRabbit to auto-remediate identified security flaws.

Built With

• coderabbit
• daytona
• google-adk
• python
• sentry
• streamlit