ml-feature-engineering
Warn
Audited by Gen Agent Trust Hub on Jul 8, 2026
Risk Level: MEDIUMPROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: Eight of the fifteen files in this skill (comprising nearly 10,000 lines of text) consist of hundreds of sections of identical, redundant technical filler text about kernel system calls and distributed query engines. This noise-to-signal manipulation is a deceptive practice used to obscure actual logic and significantly inflate the skill package size to hinder manual or automated analysis.
- [INDIRECT_PROMPT_INJECTION]: The workflows in SKILL.md for processing machine learning data sources (CSV, relational tables, time-series) represent an indirect injection surface as they ingest untrusted external data without sufficient protective measures.
- Ingestion points: Machine learning data sources entered into the agent context via workflows defined in SKILL.md.
- Boundary markers: Absent from the response format and completion criteria instructions.
- Capability inventory: The skill generates code for file reading, data transformation, and automated feature synthesis with libraries like Featuretools and tsfresh across multiple files.
- Sanitization: No validation or filtering logic for external column headers or text content is implemented in the provided code snippets.
Audit Metadata