gam-agentic-memory
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content (PDFs, videos, text files, and agent trajectories) to build a memory database. This content is then retrieved and processed by an LLM during queries, creating a risk that malicious instructions embedded in the source files could be executed.
- Ingestion points: Content is ingested via Workflow.add(), TextGAM.add(), and the gam-add CLI utility, supporting various formats including research papers, video lectures, and agent traces.
- Boundary markers: The documentation lacks mention of explicit delimiters or sanitization logic to separate user-provided data from agent instructions.
- Capability inventory: The skill interacts with the local file system to store memory structures and exposes a REST API that can be accessed over the network.
- Sanitization: There are no documented mechanisms for filtering or escaping instructions found within the processed files.
- [COMMAND_EXECUTION]: The documentation provides instructions for executing shell commands to install the package (pip install), configure environment variables (export), and launch local API or web servers (uvicorn, python).
Audit Metadata