The Agent Skills Directory

[SAFE]: The skill instructions emphasize engineering best practices, such as feature consistency, stale-feature detection, and model-input contract validation.\n- [INDIRECT_PROMPT_INJECTION]: The skill operates on machine learning model outputs and feature data which represent a potential attack surface; however, the instructions proactively mitigate this risk by requiring strict schema/contract validation and deterministic fallback behaviors.\n- [COMMAND_EXECUTION]: While the skill has 'workspace-write' permissions to modify code, all instructions focus on implementing local fixes and validation logic rather than executing arbitrary shell commands or external scripts.

ml-engineer