The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents the use of pickle.load() and nx.read_gpickle() in references/io.md for graph serialization. Deserializing data using the pickle library from untrusted or external sources is a critical security vulnerability that allows for arbitrary code execution on the host system.
[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by ingesting data from multiple external file formats, including GraphML, GML, CSV, and JSON, as documented in references/io.md. The documentation lacks guidance on input sanitization or the implementation of boundary markers, meaning malicious instructions could be embedded in node or edge attributes and processed by the agent. This is particularly concerning given the skill's capabilities for file writing and command execution.
[COMMAND_EXECUTION]: In references/visualization.md, the skill utilizes external visualization tools such as Graphviz via the pydot library. These operations involve executing system commands through external binaries. If the graph data passed to these tools contains malicious payloads or shell metacharacters, it could lead to unauthorized command execution outside of the intended scope.

alterlab-networkx