The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (LOW): The script automatically downloads the 'stopwords' dataset from NLTK.
Evidence: nltk.download('stopwords') in scripts/topic_modeler.py.
Mitigation: Downgraded to LOW/INFO as NLTK is a trusted organization for linguistic data.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes external text data which could contain malicious instructions.
Ingestion points: Reads CSV files via pd.read_csv in scripts/topic_modeler.py.
Boundary markers: None present.
Capability inventory: Only local file system writes (JSON output); no network exfiltration or command execution capabilities.
Sanitization: Basic tokenization and punctuation stripping are performed, but do not filter for instructions.
Assessment: Tier LOW (External content + internal influence only). Results are used for data visualization and keyword extraction.

topic-modeler