The Agent Skills Directory

[DYNAMIC_EXECUTION]: The reference documentation (references/io.md) includes instructions and code snippets for using Python's pickle module to serialize and deserialize graph objects. While the documentation provides a clear security warning to only load trusted files, the pickle.load() function is inherently capable of executing arbitrary code if used on a malicious binary file.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process graph data from various external formats including CSV, JSON, GraphML, and SQL databases (references/io.md). This creates a surface for indirect prompt injection if the agent uses attributes or labels from these data sources to drive further actions or responses.
Ingestion points: File reading operations in references/io.md such as read_edgelist, read_graphml, and pd.read_sql_query.
Boundary markers: The provided code examples do not include specific delimiters or instructions to ignore potential commands embedded within data attributes.
Capability inventory: The skill is configured with Bash(python:*), Bash(uv:*), and Write capabilities in SKILL.md.
Sanitization: Data is processed using standard NetworkX and Pandas functions without additional sanitization layers to detect or neutralize embedded instructions.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the networkx library and its associated dependencies (numpy, scipy, pandas, matplotlib) using uv pip install (SKILL.md). These are well-known, industry-standard packages from the official Python Package Index (PyPI).

alterlab-networkx