The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs environment discovery in SKILL.md to determine hardware resources for parallelization. It executes commands such as python3 -c "import os; print(os.cpu_count())", sysctl -n hw.memsize, and nvidia-smi to probe CPU, RAM, and GPU capabilities.
[COMMAND_EXECUTION]: The Mutator role in roles/Mutator.md executes evolved programs within isolated sandbox directories using a user-defined entrypoint command (<entrypoint>).
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface (Category 8) due to its iterative evolutionary design:
Ingestion points: roles/Mutator.md reads "inspirations" (source code from other programs) and "artifacts" (execution logs, stderr, and loss curves) from previous generations.
Boundary markers: No specific boundary markers or instructions to ignore embedded instructions within the ingested code or logs are present.
Capability inventory: The Mutator.md role has the capability to write to the file system and execute shell commands.
Sanitization: There is no evidence of sanitization or filtering applied to the ingested program code or execution artifacts before they are passed to the model for the next mutation step.

alpha-evolve