Problem

Auto Run sessions can hang indefinitely with no visible indication to the user that something has gone wrong. Multiple users have reported this:

• Agent process remains alive but produces no output for hours
• Progress percentage never updates, leaving users unsure if it's working or stuck
• No timeout or stall detection at the playbook execution level
• No thought stream / activity log to help diagnose why a hang occurred

User Reports

"Some of the playbooks would run for hours with no indication, the percentage never seemed to change" — @waynez

"Sometimes my auto runs will get stuck. Recently it was because the node path was not available to the agent - just kept updating the playbook with that note" — @calebpeffer

"The claude process was still there. Was stuck for like 6 hours doing nothing. I'm not sure what the hell happened." — @pedramamini

Root Causes (Known/Suspected)

1. Agent environment issues — Missing node in PATH causes the agent to loop writing notes instead of executing tasks
2. Agent process stall — Process alive but not producing output (unknown root cause, needs investigation)
3. No stall detection at Auto Run level — The existing stall detection (stall-count in batch processor) only tracks task-level progress, not agent-level activity

Proposed Improvements

1. Enhanced Auto Run Logging / Thought Stream

Add a visible activity log or "thought stream" to Auto Run that shows:

• Agent spawn/exit events with timestamps
• Last output received from agent (truncated)
• Current task being processed
• Time elapsed on current task

This gives users visibility into what's happening and helps diagnose hangs.

2. Agent-Level Heartbeat / Activity Timeout

Detect when a spawned agent process has produced no output for an extended period:

• Track last output timestamp per agent during Auto Run
• After configurable threshold (e.g., 10 minutes of silence), surface a warning
• After extended threshold (e.g., 30 minutes), offer to kill and retry or skip

3. Better Progress Indication

• Show per-task elapsed time in the Auto Run UI
• Show agent activity indicator (last output timestamp, "active" vs "idle")
• Surface the current task description being worked on

Related

• Existing stall detection: src/renderer/hooks/batch/useBatchProcessor.ts (tracks document-level stalling, not agent-level)
• Process management: src/main/process-manager.ts
• Batch processor: src/renderer/hooks/batch/useBatchProcessor.ts, useDocumentProcessor.ts