math-olympiad

Pass

Audited by Gen Agent Trust Hub on May 7, 2026

Risk Level: SAFE
Full Analysis
  • Local Tool Execution: The skill includes shell scripts (check_latex.sh, compile_pdf.sh) designed to verify the presence of LaTeX compilers and automate the generation of PDF documents. These scripts perform their stated functions using standard tools like pdflatex and xelatex without any suspicious side effects.
  • Computational Constraints (Deep Mode): The "Deep mode" instructions allow the agent to use local Bash or Python environments for specific mathematical tasks, such as symbolic identity checks or modular arithmetic. These instructions include robust security guidance, explicitly prohibiting the use of network-enabled tools or web searches to maintain process integrity.
  • Multi-Agent Orchestration: The workflow manages multiple specialized sub-agents (solvers, verifiers, and presenters) via the agent() tool. This orchestration is used to implement sophisticated reasoning patterns, such as context isolation and adversarial verification, which are standard for high-level mathematical problem-solving.
  • Data Ingestion and Processing: User-provided math problems are ingested and processed through various reasoning stages. While the skill interpolates this external content into prompts, the risk is mitigated by the domain-specific focus and the emphasis on pure reasoning and local computation.
Audit Metadata
Risk Level
SAFE
Analyzed
May 7, 2026, 05:20 PM