fix: cap consecutive empty LLM response retries to prevent infinite loop

[Kailigithub]

Summary

Added a consecutive empty response counter to prevent the agent from entering an infinite retry loop when the LLM consistently returns empty content.

Closes #201

Changes

• ga.py: Added _blank_streak counter to GenericAgentHandler.__init__
• ga.py: Modified do_no_tool() to track consecutive blank responses and abort after 3 failures

Details

Previously, when the LLM returned an empty response, do_no_tool would yield a warning and retry indefinitely (up to the max_turns=40 outer loop limit). This caused excessive API calls, cost waste, and log flooding with "LLM returned an empty response. Retrying..." messages.

The fix introduces a _blank_streak counter:

• Increments on each consecutive empty response
• Resets to 0 on any non-empty response
• After 3 consecutive empty responses, returns next_prompt=None which signals the agent loop to exit gracefully