Agent failure: Multi-model 'independent' vote was contaminated by one model reading another's answer
Status Open
Maintainer reply None cached
Activity 0 comments · opened Oct 5, 2026
What happened: The agent ran a three-model vote but let one model's run read another model's answer file, so the votes were not independent. It had to be re-run in isolation.
Impact on the user: Bad decision input; time lost re-running the vote.
Expected: complete, verified work on the first pass, without the user having to correct it.
Ask: please improve Claude / Claude Code so this failure mode stops recurring.
_Environment: Claude Code (Opus) on macOS, 2026-10-05. Filed at the user's request; one issue per failure that day._