The Agent Skills Directory

[COMMAND_EXECUTION]: Requests full Bash access to perform file operations, search for calibration samples, and manage the execution of background tasks.
[SAFE]: Incorporates sophisticated safety measures such as 'Leak Guard' (using grep) to ensure sensitive numerical data like play counts are not leaked into the scoring environment. It also enforces strict process isolation through the use of blind sub-agents to maintain data integrity during rubric updates.
[SAFE]: The skill uses the mcp__llm-chat__chat tool to conduct a cross-model audit of proposed changes. This external communication is an integral part of the skill's validation protocol and is used to verify the explanatory power of new scoring formulas against real-world data.

cheat-bump