seller-university-skill
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates the risk of indirect prompt injection by establishing clear trust boundaries for external data processing.
- Ingestion points: The agent is expected to process user-provided attachments and external documents, which are potential vectors for malicious instructions (File:
SKILL.md,evals/evals.json). - Boundary markers: Present. The dialogue principles in
SKILL.mdexplicitly instruct the agent that instructions found within provided materials are data only and must not override the skill's instructions or the user's task. - Capability inventory: The skill uses a local script (
scripts/find_modules.py) to search through bundled course summaries (File:SKILL.md). - Sanitization: The skill relies on natural language constraints and model-level adherence to maintain the separation between data and instructions.
- [SAFE]: The repository contains a security validation suite (
evals/fixtures/course-note.md) with simulated injection attacks. This proactive inclusion of security tests indicates a high level of security awareness and provides a verifiable way to ensure the agent maintains its safety constraints when handling untrusted documents.
Audit Metadata