Description

Spec generation gets stuck in an infinite loop where the internal Stop hook repeatedly demands the agent call a "StructuredOutput" tool, but the agent never does.

Environment

• OS: Windows 11
• Node.js: v22.17.1
• Automaker: Latest (cloned fresh today, Jan 12 2026)
• Authentication: Claude Code CLI OAuth (Max subscription)

Steps to Reproduce

1. Fresh clone of automaker repo
2. npm install
3. npm run dev:electron
4. Create a new project
5. Attempt to generate a spec for any feature

Expected Behavior

Spec generation completes and produces a specification document.

Actual Behavior

The agent gets stuck in a loop. Server logs show repeated messages like:

[SpecRegeneration] User message (tool result): {
  "type": "user",
  "message": {
    "role": "user", 
    "content": [{
      "type": "text",
      "text": "Stop hook feedback:\nYou MUST call the StructuredOutput tool to complete this request. Call this tool now."
    }]
  },
  "isSynthetic": true
}

The isSynthetic: true indicates these are internal Automaker messages, not external hooks.

This loop continues indefinitely (tested 500+ iterations), accumulating text but never completing.

Investigation Notes

• Initially suspected external Claude Code hooks were interfering
• Completely disabled all external hooks (renamed ~/.claude/hooks folder)
• Problem persists identically with no external hooks
• The "Stop hook feedback" is internal to Automaker's spec regeneration flow
• Agent receives text responses but never calls the StructuredOutput tool

Logs

Stream message count reaches 500+ with pattern:

1. Assistant sends text response (~38 chars each iteration)
2. Synthetic user message demands StructuredOutput tool call
3. Repeat indefinitely

Total accumulated text grows (e.g., "total now: 8930 chars") but spec never completes.