matematic-expert-panel
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-defined 'cases' (e.g., 'Czy wdrazac voice clone...') to be analyzed by multiple agent personas. If the case description contains adversarial instructions, it could influence the agent's behavior during the persona simulation or scoring phase.
- Ingestion points: Case description provided by the user in 'Faza 1
- Setup + case framing'.
- Boundary markers: Absent. The input is processed directly into the context of the 7 personas.
- Capability inventory: 'Read' and 'Write' access to the local file system (used for generating reports).
- Sanitization: No explicit validation or delimiter wrapping is defined for the incoming case content.
- [SAFE]: The skill references the SuperClaude_Framework (MIT) and various internal 'anthropic-skills'. These are used as methodological guides and valid tool references rather than automated remote code execution triggers.
Audit Metadata