2026-10-12: tool_use error rate at 30.25% in duo-workflow-svc

tool_use error rate at 30.25% in duo-workflow-svc (Severity 2)

Problem: The tool_use SLI for duo-workflow-svc in the main stage is showing elevated errors, with about 25–40% of tool calls failing.

Impact: About 25–40% of tool calls have been failing since around 5:00am UTC, concentrated in the code_review flow. Code reviews still run, but review quality is degraded because they can’t search the wider codebase and rely mostly on the MR diff.

Causes: We found that a Duo CLI version bump introduced failures in find_files. The first revert only covered custom-image jobs, so default-image jobs were still affected until we opened an additional revert for the default workflow image.

Response strategy: We opened revert merge requests to roll back the Duo CLI changes for both custom-image jobs and the default workflow image. We decided not to roll back to a previous GitLab package because it would revert multiple security fixes.