terminal-bench-agent-evaluation

【设计与创意】Evaluate agents on 89 challenging terminal-based tasks where frontier models score below 65%, providing realistic assessment of command-line interface automation capability. Use when benchmarking agents designed for system administration, automation, or DevOps tasks.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准