fix: add maxTurns limit to prevent agent loop runaway

[NSEvent]

Summary

Fixes #30 - Adds maxTurns limit and improves tool response format to prevent agent loop runaway.

Changes

1. Add maxTurns: 10 to query options

File: container/agent-runner/src/index.ts

options: {
  maxTurns: 10, // Prevent runaway loops (see issue #30)
  // ... other options
}

10 turns is sufficient for complex tasks while preventing runaway behavior that caused the incident (21 duplicate messages in 32 seconds).

2. Improve send_message tool response

File: container/agent-runner/src/ipc-mcp.ts

Before (ambiguous):

Message queued for delivery (filename)

After (clear success signal):

✓ Message sent successfully. The user will receive: "preview of message..."

The clearer response format helps the SDK understand the task is complete, reducing retry behavior.

Test plan

• Rebuild container: ./container/build.sh
• Test with reminder request to verify no duplicate sends
• Verify agent completes within 10 turns for normal tasks

Acceptance Criteria from Issue

• maxTurns: 10 added to query options in container/agent-runner/src/index.ts
• Tool response format improved in ipc-mcp.ts

🤖 Generated with Claude Code

[gavrielc]

@NSEvent Thank you for the PR.

maxTurns is not an appropriate fix to the issue you reported. This would end every session after 10 turns (i.e. 10 messages between the user and the agent). With this kind of agent, people typically want the same session to continue indefinitely (auto-compaction clears the context window when needed). There is an explicit clear command for clearing the conversation and starting a new session.

Closing this I'm not sure it's an appropriate fix. Will look into the issue separately