math-computational
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied mathematical problems which could potentially contain instructions to influence the agent's behavior.
- Ingestion points: Mathematical problems provided by the user (SKILL.md).
- Boundary markers: The process includes a dedicated step to 'Frame the problem' and restate input as a mathematical mapping, which helps maintain logical separation. The skill also defines an execution boundary of 'in-memory, no file writes'.
- Capability inventory: Execution of Python code using NumPy, SciPy, and SymPy libraries (SKILL.md).
- Sanitization: The skill mandates input validation (checking for NaN, infinity, and dimension mismatches) and sanity assertions on results.
- [DYNAMIC_EXECUTION]: The skill uses runtime code generation and execution to perform complex mathematical computations.
- Evidence: Step 4 ('Implement / compute') instructs the agent to run Python code containing the mathematical algorithms.
- Mitigation: The skill explicitly restricts these operations to in-memory execution and prohibits file writes, limiting the potential impact of executed code.
- [SAFE]: No evidence of malicious prompt injection, data exfiltration, obfuscation, or persistence mechanisms was found. The skill's behavior is consistent with its stated purpose of mathematical problem-solving.
Audit Metadata