The Agent Skills Directory

[SAFE]: The skill implements a benchmark automation flow including schema validation, isolated workspace environment setup under /tmp/opencode/, and subagent coordination via the Task tool. All operations utilize deterministic local tasks (just and uv scripts) without dangerous remote code execution or data exfiltration vectors. Hardcoded local path templates (e.g., /home/romuald/...) are used as placeholders for the evaluation environment and do not pose a credential or exposure risk.

skill-evaluator