Praison AI

Doom Loop Detector • AI Agent SDK

DoomLoopDetector: Detects and prevents doom loops in agent execution.

DoomLoopDetector

Defined in the Doom Loop module.

AI Agent

Detects and prevents doom loops in agent execution.

Monitors action history for patterns that indicate stuck states:

  • Repeated identical actions
  • Repeated similar actions
  • Consecutive failures
  • No meaningful progress

Constructor

ParamField Optional[DoomLoopConfig]

No description available.

Methods

start_session()

Start a new detection session.

 record_action()

Record an action for loop detection.

 record_response()

Record model response text for content streaming loop detection.

 mark_progress()

Mark meaningful progress.

 is_doom_loop()

Check if current state indicates a doom loop.

 get_loop_type()

Get the type of doom loop detected.

 get_loop_event()

Get the current loop event if one exists.

 get_recovery_action()

Get recommended recovery action.

 apply_backoff()

Apply backoff delay and return the delay used.

 reset_backoff()

Reset backoff to initial value.

 increment_recovery()

Increment recovery attempt counter.

 get_stats()

Get detection statistics.

Usage

python
detector = DoomLoopDetector()

    # Record actions
    detector.record_action("read_file", {"path": "foo.py"}, "content", True)
    detector.record_action("read_file", {"path": "foo.py"}, "content", True)

    # Check for loops
    if detector.is_doom_loop():
        event = detector.get_loop_event()
        recovery = detector.get_recovery_action()

Source

View on GitHub

praisonaiagents/escalation/doom_loop.py at line 90