student-mentor
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No security issues detected. The skill demonstrates best practices in AI agent safety, including:
- Human-in-the-Loop: All outputs are explicitly defined as drafts and include non-removable warnings requiring the professor to verify accuracy before use.
- Principle of Least Privilege: The skill requests no dangerous tools (no network access, no shell execution) and limits its scope to text drafting.
- Evidence-Bound Execution: Internal rules prevent the agent from fabricating anecdotes or data, requiring all evaluative claims to be traced back to professor-provided inputs.
- Privacy Protection: The skill mandates 'Privacy Minimalism' by suggesting pseudonyms and prohibiting the storage of student-identifying information.
- [SAFE]: Indirect Prompt Injection Surface Evaluation:
- Ingestion points: Untrusted student work (essays, submissions) enters the context during 'feedback' and 'integrity-case' modes.
- Boundary markers: While explicit delimiters are not shown in templates, the instructions enforce logical boundaries through 'evidence-bound' drafting rules.
- Capability inventory: Impact is restricted to text generation; the skill has no capability for network exfiltration, file system modification, or command execution.
- Sanitization: All outputs are processed through templates that append mandatory verification blocks.
Audit Metadata