Devin AI 8105e74267 Fix agent loop not stopping after max iterations reached ...

Fixes #3847

The agent execution loop was not stopping after max_iter was reached.
After calling handle_max_iterations_exceeded(), the loop continued and
made additional LLM calls, overwriting the formatted answer.

Changes:
- Modified handle_max_iterations_exceeded() to always return AgentFinish
  instead of AgentAction | AgentFinish, ensuring a final answer is always
  produced when max iterations are exceeded
- Added early return in _invoke_loop() after handle_max_iterations_exceeded()
  to prevent additional LLM calls
- Updated test_agent_custom_max_iterations to expect 2 LLM calls instead of 3
  (1 initial + 1 from max_iter handler, no longer 3 due to extra call bug)
- Added test_agent_max_iterations_zero to verify no get_llm_response calls
  when max_iter=0
- Added test_agent_max_iterations_one_stops_after_two_calls to verify exactly
  2 LLM calls with max_iter=1
- Fixed type annotations to properly handle AgentAction | AgentFinish | None

Co-Authored-By: João <joao@crewai.com>

2025-11-06 15:16:12 +00:00

..

crewai

Fix agent loop not stopping after max iterations reached

2025-11-06 15:16:12 +00:00

crewai-tools

fix: allow adding RAG source content from valid URLs (#3831)

2025-11-04 07:58:40 -05:00

devtools

feat: bump versions to 1.3.0 (#3820)

2025-10-31 18:54:02 -07:00