symbolic-equation
Fail
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: HIGHDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill's workflow in SKILL.md and logic in references/llmsr-patterns.md involve an LLM proposing Python code versions that are dynamically compiled and executed by evaluator.py using _sample_to_program and a sandbox (self._sandbox.run).
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted dataset descriptions and physical context through the $0 input parameter, which is then interpolated into the prompt used for equation evolution. 1. Ingestion points: $0 dataset description in SKILL.md. 2. Boundary markers: Absent in prompt construction. 3. Capability inventory: Python code execution with numpy and scipy. 4. Sanitization: Absent.
- [COMMAND_EXECUTION]: The execution of dynamically generated Python programs represents a code execution vector, as the fitness evaluation step relies on code proposed by the LLM.
Recommendations
- AI detected serious security threats
Audit Metadata