Evidence-Guided Troubleshooter

Stop guessing at the cause

When something goes wrong, general AI often jumps to the most common explanation or gives a long checklist of generic fixes. That can waste time, repeat tests you already tried, or push you toward resets, replacements, and other costly actions before the problem is understood.

What this skill does

• Separates known facts, user assumptions, and important unknowns.
• Keeps 2–4 genuinely distinguishable candidate causes when the evidence is incomplete.
• Chooses one safe, reversible, low-cost test with the highest expected information gain.
• Updates the candidates from each result instead of defending its first guess.
• Clearly labels whether the issue is not localized, narrowed, strongly supported, acted on, or verified resolved.
• Stops when further troubleshooting costs more than the problem, requires unavailable access, or needs professional help.

Designed for real back-and-forth

The user normally performs only one small test per turn and reports the minimum result needed. If a test cannot be completed, the skill offers at most one lower-burden alternative. A fix is never called resolved until it works again in the original failure condition.

Good fits

• Low-risk technology and device problems.
• Workflows with an unexplained bottleneck or failure point.
• Everyday problems with observable outcomes.
• Business symptoms that can be decomposed into measurable causes.

Important boundaries

This is not a universal answer engine. It switches away from causal troubleshooting when the real task is data verification, a business choice, a subjective preference, or a value tradeoff. It stops and recommends appropriate help for urgent medical, fire, gas, electrical, fraud, legal-deadline, major security, permanent-data-loss, or dangerous-disassembly situations.

Free test version

After the issue is resolved or troubleshooting stops, the skill asks once whether the process felt easier, about the same, or more troublesome than asking AI normally. This feedback is used to test whether the method actually reduces guessing and unnecessary work.