The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions heavily utilize the gbrain CLI environment to perform benchmarking and optimization tasks, including commands like gbrain skillopt for rollouts and evaluations.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest external benchmark data (e.g., skillopt-benchmark.jsonl) and existing skill instructions to generate optimized content. While this creates a surface for indirect instructions to influence the agent's optimization logic, the skill incorporates several safety protocols, including mandatory validation gating (median-of-3 scoring) and required human review of bootstrapped judges to prevent the adoption of malicious or low-quality instructions.
[SAFE]: The skill references external documentation located on the author's GitHub repository (github.com/garrytan/gbrain). This is identified as a legitimate vendor resource used for providing tutorials and technical guidance.

skill-optimizer