The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface because it processes untrusted external data (such as CSV or JSON datasets) to determine its execution parameters and generated analysis code. The absence of explicit boundary markers or sanitization logic increases the risk that malicious instructions embedded within the data could be interpreted by the agent.
Ingestion points: The skill reads external datasets through the scripts/data_loader.py utility and instructions in SKILL.md that guide the agent to load files like customer_data.csv.
Boundary markers: No markers (such as XML tags or clear delimiters) are specified in the prompts or scripts to separate dataset content from instructions.
Capability inventory: The agent has high-privilege access to the Bash(cmd:*), Write, and Edit tools, allowing for significant system interaction.
Sanitization: The provided scripts and instructions lack data validation or sanitization routines for the content extracted from external files.
[COMMAND_EXECUTION]: The skill requires the high-privilege Bash(cmd:*) tool to execute clustering algorithms and generated Python scripts. While this is necessary for the skill's primary function, it provides a powerful capability that could be misused if the agent's behavior is influenced via malicious input data.

running-clustering-algorithms