The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill focuses on workflows that handle messy and untrusted data files from external sources.
Ingestion points: External datasets are loaded using pandas (pd.read_csv, pd.read_excel) and Stata commands across multiple guides, such as csv-data-analyzer/SKILL.md and data-cog-guide/SKILL.md.
Boundary markers: Not applicable, as the data is parsed as standard tabular structures rather than being directly re-interpolated into live LLM prompts.
Capability inventory: Core actions are constrained to statistical summaries, data transformations, visualization, and local machine learning models; no malicious capabilities like shell access or network data transmission exist.
Sanitization: Pipelines implement thorough sanitization practices including type casting, regex-based special character removal, and missing value treatment.
[SAFE]: The documentation contains a standard reference and installation instructions for an open-source repository (https://github.com/Wilson-ZheLin/Streamline-Analyst.git) in streamline-analyst-guide/SKILL.md, which is benign and poses no security threat.

wrangling-skills