instructor
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONINGEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents processes for ingesting untrusted data from LLM providers into structured models, creating a potential surface for indirect prompt injection.
- Ingestion points: LLM completions processed via the
instructorclient'screatemethod as documented in overview and usage files. - Boundary markers: Data is constrained by Pydantic schemas and
response_modeldefinitions. - Capability inventory: The skill contains documentation and a version-checking script with no dangerous shell or network execution capabilities.
- Sanitization: Employs Pydantic's automated type checking and field validation to sanitize incoming data.
- [METADATA_POISONING]: The skill's description field uses aggressive instructional language and mandatory triggers to influence the agent's selection process for structured extraction tasks.
- [EXTERNAL_DOWNLOADS]: The skill references the official Python Package Index (PyPI) for installation and version verification. These references are to a well-known service and are documented neutrally.
Audit Metadata