agentic-research-workflow
Fail
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: CRITICALCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The
research_agentfunction insrc/agents.pyuses the dangerouseval()function to parse tool arguments:arguments = eval(tool_call.function.arguments). Because these arguments are generated by the LLM—which can be manipulated by user input or malicious external data—this creates a direct path for Arbitrary Code Execution (ACE) on the host environment. - [EXTERNAL_DOWNLOADS]: The installation instructions require cloning a repository from
https://github.com/https-deeplearning-ai/agentic-ai-public.git. This repository is not from a verified organization or a recognized trusted source. Executing unverified code from third-party repositories presents a significant supply chain risk. - [REMOTE_CODE_EXECUTION]: The combination of arbitrary code execution via
eval()and the dependency on external, unvetted scripts significantly increases the risk of full system compromise by a remote attacker.
Recommendations
- AI detected serious security threats
Audit Metadata