math-modeling-solver
Pass
Audited by Gen Agent Trust Hub on Jul 12, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its data ingestion and processing workflow.
- Ingestion points: In Stage 1 and Stage 1.5, the skill ingests untrusted data from two main sources: competition problem texts pasted by the user and literature search results retrieved from external academic databases (e.g., IEEE, arXiv, Google Scholar) using the agent's search capabilities.
- Boundary markers: The instructions in
SKILL.mddo not implement strict delimiters (such as XML tags or specific character sequences) or "ignore embedded instructions" warnings for the ingested problem text or search snippets before processing them. - Capability inventory: The skill has significant capabilities, including performing network searches and generating complex, executable Python and MATLAB code based on the content of the ingested data.
- Sanitization: There is no evidence of automated sanitization or filtering of the external content before it is interpolated into the agent's reasoning process and code generation stages.
Audit Metadata