September 5, 2026Assessments
Prompt evidence and agent usage on reports
Candidate reports now quote the candidate's two lowest-scored prompts, each with its score, its weakest area, and a link to that moment in the replay.
What changed
- Candidate reports now quote the candidate's two lowest-scored prompts, each with its score, its weakest area, and a link to that moment in the replay.
- On questions that grant Claude Code or Codex, the report counts the prompts the candidate sent to each granted agent, including any agent they never used.
- The sandbox terminal lists servers the candidate has running, with buttons to preview or stop each one. Sandboxes now start near the candidate, so the terminal responds faster.