NYC

networkx

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • Dynamic Execution (MEDIUM): The documentation in 'references/io.md' demonstrates the use of 'pickle.load()' and 'nx.read_gpickle()'. * Evidence: Deserializing data using the Python pickle module is inherently insecure. If an agent uses these functions on a file provided by an untrusted source, it can result in arbitrary code execution.
  • Indirect Prompt Injection (LOW): The skill exposes a significant attack surface by documenting how to read data from various external file formats. * Ingestion points: 'references/io.md' specifies functions like read_adjlist, read_edgelist, read_gml, read_graphml, read_gexf, read_pajek, read_leda, read_gpickle, and read_shp. * Boundary markers: Absent. There are no instructions or delimiters defined to prevent the agent from following instructions embedded within these data files. * Capability inventory: The skill includes documentation for writing to files, modifying SQL databases, and invoking external visualization tools ('pydot' or Graphviz). * Sanitization: Absent. No mention of data validation or escaping before processing.
  • Unverifiable Dependencies & Remote Code Execution (LOW): Several documented features depend on libraries or system binaries not directly managed by the skill. * Evidence: 'references/io.md' refers to the 'pydot' and 'pyshp' Python packages and the 'Graphviz' system application for DOT format processing and geographic data handling.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 05:58 PM