Run a Tool-Outage Drill for Your Workflow Twin

ROLE You are a reliability engineer for tool-using AI assistants. OBJECTIVE Design a controlled outage drill for my workflow twin. CONTEXT The twin performs [WORKFLOW]. Its tools are [CONNECTED TOOLS]. Actions with external consequences are [CONSEQUENTIAL ACTIONS]. A safe test environment is [TEST ENVIRONMENT]. INSTRUCTIONS Create scenarios for a timeout, expired permission, malformed tool result, partial completion, and an uncertain action outcome. For each, define the observable failure signal, safe assistant response, information shown to the user, and conditions for retrying. Include a case where retrying could duplicate an external action. Specify how to check completion through an authoritative system rather than the assistant's wording. Design a human recovery checklist and a record of attempted versus confirmed actions. Make the drill executable with mocked tool responses if live fault injection is unavailable. CONSTRAINTS Do not run destructive tests or contact real customers. Never treat a generated confirmation as proof of execution. Avoid fabricated statistics and flag assumptions. Identify platform-dependent behavior explicitly. OUTPUT FORMAT Fault matrix, mocked responses, expected behaviors, recovery checklist, and pass-or-fail criteria.