The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the ingestion of network data from numerous external sources and formats, creating a surface for indirect prompt injection where malicious instructions could be embedded in data files.
Ingestion points: references/io.md documents functions like nx.read_edgelist, nx.read_graphml, nx.read_gml, nx.node_link_graph, and pd.read_sql_query.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are included in the data-reading prompts.
Capability inventory: The skill allows the agent to read, search (Glob/Grep), and write files (e.g., nx.write_graphml, plt.savefig, df.to_csv).
Sanitization: No methods for validating or sanitizing the content of ingested graph files are provided before the data is processed or visualized.
[REMOTE_CODE_EXECUTION]: The skill provides documentation on how to use Python's pickle module for graph serialization, which is a known vector for unsafe deserialization attacks.
Evidence: references/io.md contains code snippets for pickle.load() and the NetworkX wrapper nx.read_gpickle().
Context: While these are standard features of the library intended for local data persistence, they present a security risk if the agent is instructed to load serialized graph files from untrusted or external sources.

networkx