categorize-corpus

Pass

Audited by Gen Agent Trust Hub on Jul 4, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides purely instructional content without any associated scripts, configuration files, or executable payloads.
  • [SAFE]: The skill references well-known and trusted entities for model hosting and inference, including Hugging Face, Ollama, and official cloud API providers.
  • [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection as it involves processing untrusted document corpora.
  • Ingestion points: User-provided documents interpolated into the classification prompt (SKILL.md).
  • Boundary markers: The prompt uses clear labels ("Document:") and requests structured JSON responses to constrain model output.
  • Capability inventory: No direct execution capabilities are present in the skill itself; it describes external LLM usage.
  • Sanitization: Not explicitly implemented in the example, but the instructions recommend JSON mode for output validation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 4, 2026, 03:12 AM
Security Audit — agent-trust-hub — categorize-corpus