The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted scientific data files (CSV, FASTA, JSON, HDF5, etc.) and processes them to generate a markdown report. If these data files contain malicious natural language instructions, they could potentially influence the agent's behavior when it reviews the resulting EDA report.
Ingestion points: The scripts/eda_analyzer.py script reads external files using various parsers including pandas, numpy, biopython, and h5py.
Boundary markers: The generated report does not include specific delimiters or instructions for the agent to ignore content from the data files, which may lead to the agent following instructions accidentally extracted from the data.
Capability inventory: The skill has access to shell tools (Bash(uv:*), Bash(python:*)) and file system tools (Read, Write, Edit).
Sanitization: The script performs standard data extraction but does not explicitly sanitize the content of the scientific data for injection patterns before including them in the report metadata or summary sections.
[COMMAND_EXECUTION]: The skill executes a local Python script scripts/eda_analyzer.py using uv run python. This script utilizes several third-party libraries for parsing scientific data formats, which is an expected part of the skill's functionality.

alterlab-eda