The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to discover equations by generating and executing Python code at runtime. The evaluator.py logic described in references/llmsr-patterns.md shows that LLM-proposed code samples are compiled and run within a sandbox environment to determine their fitness scores. This dynamic code generation and execution pose a inherent risk if the LLM produces malicious logic.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes the dataset description and physical context ($0) which are then used to guide the LLM's equation generation.
Ingestion points: Untrusted data enters the agent context through the $0 dataset description parameter in SKILL.md.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands in the LLM Instruction Prompt found in references/llmsr-patterns.md.
Capability inventory: The skill possesses the capability to execute generated Python scripts using the evaluator.py component.
Sanitization: The skill mitigates risks by executing programs within a sandbox and applying a 30-second timeout as specified in evaluator.py and config.py.

symbolic-equation