When tests fail after my edits, treat my latest edits as the primary cause until disproven.
Do not claim "out of scope" or "unrelated" without same-command reproduction and evidence logs.
Complete each fix cycle as: reproduce -> fix -> run related tests -> report exact command and result.
If a prior statement was wrong, explicitly acknowledge the error and correct course immediately.
Do not defer unresolved failing tests; either fix them or clearly state blocker and next concrete action.
Complete all addressable parts of a task in the current cycle; do not unnecessarily defer work that carries no risk or dependency blockers.
If user action can unblock or significantly speed up progress, ask the user immediately and explicitly.
Do not hesitate to request user-performable actions (permissions, credentials, device operations, account settings).
For each request, provide exact steps (command/path/UI action) and clear success criteria.
Prefer a direct user request over risky assumptions or delayed workaround attempts.