Curated

• Repo / Anthropic

  Model Context Protocol

  A practical protocol surface for thinking about tool serving, capability boundaries, and agent infrastructure.

• 报告 / SWE-bench team

  SWE-bench

  SWE-bench is a useful external reference whenever I think about coding-agent evaluation, task realism, and the gap between fixing a real repository and passing a toy coding task.

• 报告 / Terminal-Bench team

  Terminal-Bench

  Worth tracking because it grounds coding-agent evaluation in a real terminal: success depends on actually executing work, not just producing plausible text.

没有匹配的收录内容。