keyword-extractor
Warn
Audited by Gen Agent Trust Hub on Feb 15, 2026
Risk Level: MEDIUMPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [Indirect Prompt Injection] (MEDIUM): The skill is designed to ingest and process untrusted external data which can influence agent reasoning.
- Ingestion points: The
extract,extract_from_filemethods, and--text/--inputCLI arguments ingest external content. - Boundary markers: Absent. The skill does not appear to use delimiters to separate data from instructions.
- Capability inventory: The skill can read local files, generate visualizations, and export data to JSON/CSV.
- Sanitization: None identified. Maliciously crafted input text could bias the keyword extraction to influence downstream agent actions.
- [Data Exposure] (LOW): The skill provides functionality to read local files via
extract_from_fileand the--inputCLI argument. While no active network exfiltration was detected, an agent could be tricked into reading sensitive files (e.g., config files), and the resulting 'keywords' or exported summaries could leak fragments of that sensitive data.
Audit Metadata