# SoloTraveller

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Aman Kumar singh](https://cerebralvalley.ai/u/aman6389)
- **GitHub:** https://github.com/AmanSingh-tech/AgentHarness
- **Demo video:** https://www.youtube.com/watch?v=l5KS26WiRew
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2350

AgentHarness is an AI Agent Safety and Evaluation Platform designed to make autonomous AI agents more reliable, secure, and predictable.
As AI agents evolve from answering questions to executing actions through external tools, they introduce new risks such as prompt injection, unauthorized tool usage, sensitive data leakage, and unreliable task execution. Existing agent frameworks focus primarily on capabilities, but lack comprehensive mechanisms to evaluate and control how agents behave.
AgentHarness addresses this through two core components:
AgentShield acts as a security gateway between AI agents and their tools. It enforces schema validation, least-privilege policies, risk-based action approval, prompt-injection defenses, and provenance-aware tracking of sensitive information to prevent secret and PII leakage. It also performs post-action verification and maintains tamper-evident audit logs.
AgentGym is a deterministic evaluation environment that tests agents against 11 fault scenarios, including malicious inputs, tool failures, and unexpected execution conditions. It independently verifies execution outcomes instead of relying solely on the agent's own claims.
The platform supports tool-calling workflows through MCP and provides visibility into agent decisions, tool interactions, security violations, and evaluation results.
Our goal is to move beyond building more capable AI agents and establish the infrastructure needed to make them trustworthy enough for real-world deployment.

## More from Meta Global AI Developer Hackathon

- [ninemomo](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2347)
- [NodXAI Science](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2348)
- [Franck Melka](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2349)
- [Corgi_team](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2351)
- [FS-SC](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2352)
- [Team Sentinel](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2353)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/2350. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt