Summary
When prepare_workspace fails inside implement_task, the node returns last_error but does not increment retry_count. The routing function _route_implementation checks retry_count >= 3 to escalate, but since it's never incremented, the workflow loops forever.
Observed in production: AISOS-2208 hit this path and logged 5,600+ retries in under 5 minutes with no backoff, burning CPU on a tight loop of git fetch fork failures.
Root Cause
In src/forge/workflow/nodes/implementation.py, the prepare_workspace exception handler (lines 63-69) omits retry_count:
except Exception as exc:
logger.error("Unable to prepare implementation workspace for %s: %s", ticket_key, exc)
return {
**state,
"last_error": str(exc),
"current_node": implementation_node,
# ← missing: "retry_count": state.get("retry_count", 0) + 1,
}The other two error paths in the same function (lines 150-152 and 267-272) correctly increment retry_count.
The Loop
implement_task → prepare_workspace() raises (e.g. missing fork_owner/fork_repo in state) → returns with last_error but retry_count stays 0_route_implementation sees last_error with retry_count=0 < 3 → routes back to implement_task- Goto 1 — forever
Fix
Add "retry_count": state.get("retry_count", 0) + 1 to the return dict at line 65.
Log Evidence
2026-07-17 12:00:46,030 - forge.workflow.nodes.implementation - ERROR - Unable to prepare implementation workspace for AISOS-2208: Cannot recreate workspace for AISOS-2208: missing branch_name, current_repo, fork_owner, or fork_repo in state
2026-07-17 12:00:46,034 - forge.workspace.git_ops - INFO - Syncing with fork/forge/aisos-2208 before implementing changes
2026-07-17 12:00:46,075 - forge.workflow.nodes.workspace_setup - WARNING - Workspace sync failed for AISOS-2208; recreating workspace from fork: Git command failed: git fetch fork
This three-line pattern repeated ~5,600 times in under 5 minutes with no backoff.
Summary
When
prepare_workspacefails insideimplement_task, the node returnslast_errorbut does not incrementretry_count. The routing function_route_implementationchecksretry_count >= 3to escalate, but since it's never incremented, the workflow loops forever.Observed in production: AISOS-2208 hit this path and logged 5,600+ retries in under 5 minutes with no backoff, burning CPU on a tight loop of
git fetch forkfailures.Root Cause
In
src/forge/workflow/nodes/implementation.py, theprepare_workspaceexception handler (lines 63-69) omitsretry_count:The other two error paths in the same function (lines 150-152 and 267-272) correctly increment
retry_count.The Loop
implement_task→prepare_workspace()raises (e.g. missingfork_owner/fork_repoin state) → returns withlast_errorbutretry_countstays 0_route_implementationseeslast_errorwithretry_count=0 < 3→ routes back toimplement_taskFix
Add
"retry_count": state.get("retry_count", 0) + 1to the return dict at line 65.Log Evidence
This three-line pattern repeated ~5,600 times in under 5 minutes with no backoff.