The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and summarize content from a wide variety of untrusted scientific data files (e.g., CSV, JSON, FASTA, etc.). This content is processed by scripts/eda_analyzer.py and included in reports, creating an indirect prompt injection surface.
Ingestion points: Scientific data files are read by the eda_analyzer.py script and the agent is instructed to perform 'custom analysis' on these files in SKILL.md.
Boundary markers: The skill does not implement or suggest the use of boundary markers or instructions to ignore embedded commands within the processed data.
Capability inventory: The eda_analyzer.py script and the agent have the capability to read and write files on the local filesystem and use numerous external processing libraries.
Sanitization: Content extracted from analyzed files is included in markdown reports without sanitization or filtering.
[REMOTE_CODE_EXECUTION]: The provided reference documentation (references/chemistry_molecular_formats.md and references/proteomics_metabolomics_formats.md) identifies the pickle library as a standard tool for handling .pkl and .pickle files. Since the agent is encouraged to use these references to implement custom analysis on user-provided data, this creates a risk of arbitrary code execution via unsafe deserialization of untrusted files.

exploratory-data-analysis