The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (MEDIUM): Unsafe deserialization using the Python pickle module.
Evidence: In references/io.md, code snippets demonstrate loading graphs using pickle.load(f) and nx.read_gpickle('graph.gpickle').
Detail: The pickle module is inherently insecure. If an agent follows these examples to process a file provided by an untrusted user, it could lead to arbitrary code execution (RCE) on the host environment. Modern security practices discourage the use of pickle for data exchange.
[DATA_EXFILTRATION] (LOW): Broad data ingestion surface from external files and databases.
Evidence: references/io.md provides detailed instructions for reading data from CSV, SQL databases, GraphML, GML, and other formats.
Detail: While these are standard features of the library, they define a wide attack surface for Indirect Prompt Injection. If an agent processes maliciously crafted graph files (e.g., node attributes containing instructions), it could be manipulated into exfiltrating data or performing unintended actions.

networkx