An attacker can bypass intended directory restrictions by providing parent-directory traversal sequences (e.g., ../../../) within the file_name parameter. This can be exploited through direct invocation of these tools or via prompt injection attacks where an agent processes malicious input containing the traversal payloads. Successful exploitation permits an attacker to escape the base_dir boundary, enabling arbitrary file read access, overwriting sensitive system files, or executing arbitrary Python code with the privileges of the application process. This vulnerability presents a high risk for environments where Agno agents are configured to interface with local filesystems or have broad execution authority.

Impact

The vulnerability allows unauthorized actors to compromise the integrity and confidentiality of the host environment. By leveraging the save_to_file or run_python_file actions, an attacker can achieve remote code execution (RCE) in the context of the agent process, potentially leading to full system compromise, exfiltration of environment variables or credentials, and persistence within the affected host or container environment.

Recommendation

• Update the Agno library to the patched version that implements strict path normalization and boundary validation for the file_name argument.
• Implement restrictive filesystem permissions for the service user account running Agno agents to limit the impact of potential traversal attempts.
• Monitor application logs for suspicious tool input patterns containing repeated directory traversal sequences like "../".
• Restrict agent access to unnecessary filesystem paths via environment isolation, such as containerization or chroot jails, where feasible.