You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
discovery(orchestration): WorkflowExecutor god class (1,178 LOC) — decompose into StepOrchestrator/ParallelCoordinator/CheckpointResumer #6827
Delegates to `error_handler.py` but maintains its own retry state
Sub-workflow delegation
Recursive nesting depth limit
Checkpoint resume
Resume-from-checkpoint logic interleaved with normal execution
Per AutoBot CLAUDE.md "≤30-line functions, no god-classes" engineering standard, a 1,178-LOC class is a refactor candidate.
This issue is related to #6823 (executor fragmentation) but distinct: even within `WorkflowExecutor` alone, the responsibilities can be split.
Why now
Decomposition was started in #5058 (god-class refactor that produced `WorkflowRunner` + `PerformanceTracker` + `AgentRouter` + `CollaborationCoordinator`). `WorkflowExecutor` was deliberately left as the next target — see orchestrator.py:534 docstring noting the legacy planner/executor path.
Activity
refactor(orchestration): unify AgentRegistry, decompose WorkflowExecu…
mrveiss commentedon May 17, 2026
Closed via PR #8082 (commit cc3006b on Dev_new_gui).
Acceptance criteria met:
orchestration/checkpoint_resumer.py—CheckpointResumerownssave/apply/clearcheckpoint logic (99 LOC)WorkflowExecutor._save_checkpoint,_apply_checkpoints, andclear()now delegate toCheckpointResumerWorkflowCheckpointManagerandStepCheckpointno longer imported directly inworkflow_executor.py