debuglisted

# Debug with evidence Find the cause of the reported failure, not merely a change that hides it. Accept a symptom, failing command, or captured artifact; ask only for missing information that prevents the next useful check. Feature work is outside this skill. ## 1. Establish the failure - Read applicable instructions, relevant code, callers, tests, and recent changes. Record expected versus actual behavior, affected inputs, environment, and versions. - Confirm scope: diagnosis permits inspection, not source edits. A fix request permits local implementation and verification, not deployment or production experiments. Obtain authorization for instrumentation or other writes outside that scope. - Protect existing work. Sanitize commands, logs, traces, and fixtures before sharing; never dump credentials or replay requests with real side effects without permission. - During an incident, separate authorized service-restoration measures from diagnosis; retain available evidence before disruptive actions. Recovery does not prove cause. **Ready:** a precise symptom and a safe investigation scope; otherwise request the missing observation or access without inventing a diagnosis. ## 2. Build a discriminating check - Run the smallest available test, CLI/API invocation, browser interaction, or replay that reaches the real failure. Assert the reported outcome, not merely absence of errors. Record the invocation, relevant output, and baseline before changing behavior. - R