The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external user datasets for statistical visualization, which creates a potential surface for indirect prompt injection if those datasets contain malicious natural language instructions.
Ingestion points: User-provided dataframes (e.g., df, df_paired, df_nested in examples/statanno_phd.R and SKILL.md) are ingested by visualization functions such as ggboxplot in R and sns.boxplot in Python.
Boundary markers: The skill instructions do not specify or implement boundary markers or instructions to ignore potential commands embedded within data fields.
Capability inventory: The skill exhibits file-writing capabilities via ggsave in R and utilizes standard package management tools like pip and install.packages for environment setup.
Sanitization: The provided templates do not include data validation or sanitization steps to filter for prompt injection strings within the processed categorical or numerical data.

bio-data-visualization-statistical-annotation