The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for analyzing potentially untrusted datasets, which introduces a vulnerability surface if the input data contains malicious instructions intended to manipulate the agent.
Ingestion points: External datasets loaded for Exploratory Data Analysis (EDA), statistics, and visualizations as described in SKILL.md and workflows/analyze-dataset.md.
Boundary markers: There are no specific instructions or delimiters provided to alert the agent to ignore prompt injection attempts or system instructions embedded within cell values or document text of the datasets it processes.
Capability inventory: The skill uses Bash and Python (pandas, numpy, scipy) for analysis, providing a significant execution environment that could be misused if instructions in the data are followed.
Sanitization: No data validation, character escaping, or content-filtering steps are defined for the data read from external sources before it is processed by the analysis tools.

data-analysis