The 60-second timeout was too aggressive and could incorrectly mark legitimate long-running tool calls as stale. Relying on Redis TTL (1 hour) for cleanup is sufficient and more reliable.
autogpt_platform/CLAUDE.md
Getting Started (Released)