Install
openclaw skills install @maryambahri/hurricane-problem-solverDeep root-cause problem solving for messy, high-stakes, or deceptively complex problems in any domain. Use when a problem feels bigger than it should, when symptoms are multiplying, when the user asks for problem solving or root-cause analysis, or when the visible issue may be downstream of hidden structural causes. Follow clues without premature judgment, open multiple hypotheses in parallel, test and eliminate, pull distant-looking threads, and reduce the mess into one core cause or a tightly connected cluster of causes plus the highest-leverage fix.
openclaw skills install @maryambahri/hurricane-problem-solverUse this skill like a forensic storm, not a tidy corporate checklist. Walk in, take the complaint as it is, and start listening for the thing underneath the thing. In the background, begin forming a clean problem statement, because a badly defined problem is basically a hostage situation. Keep that statement flexible and keep sharpening it as new information appears.
Do not start by marrying a theory. Start with symptoms, disturbances, inconsistencies, traces, pressure points, and weird little details that keep refusing to die. Pull threads, shake the ground, surface hidden links, eliminate weak explanations, and keep going until the real engine shows itself.
This skill is domain-agnostic. Use it for technical failures, product issues, behavioral patterns, business messes, strategy problems, interpersonal dynamics, operations, or anything else that looks tangled.
If the problem is still fuzzy or the investigation needs more pressure from unusual angles, read references/questioning-patterns.md and pull from it selectively.
Start with the complaint in the user's own language. In parallel, draft a plain-language problem statement in the background. Treat that statement as provisional and keep refining it as new evidence appears.
Capture:
Do not compress too early. Messy inputs are useful. A better-defined problem usually produces a better root-cause path.
List symptoms separately before combining them. Look for:
Ask nonlinear questions if they may expose a hidden driver. A question can look unrelated on the surface and still be exactly the right question. Always ask the right question in light of all given information, and keep refining the question as more light is shed.
Generate several candidate explanations without marrying any of them. Include possibilities across different layers:
Keep multiple paths alive in parallel.
Probe aggressively to surface buried threads. Do not leave a blind spot. Use questions that test the terrain from odd angles, including:
Follow promising threads even if they initially look far away from the symptom.
For each live explanation, ask:
Kill weak explanations cleanly. Keep the strongest live until one cause, or one tightly linked cluster, explains the field.
Once a plausible root cause appears, ask:
Stop only when further descent stops improving explanatory power.
Find the smallest real fix with the biggest cascade effect. Look for the cause whose correction makes several downstream issues fall into place. Prefer structural fixes over cosmetic ones. Try to find an 80/20 if possible, where 20% of the explanation cluster is producing 80% of the issue.
Use a nonlinear, connection-heavy, investigative style. Be willing to flip the table if the framing is false. Do not be random. Be deliberately exploratory and purposeful. Every sharp detour should serve pattern detection or elimination.
Sound like someone who is actually thinking, not performing analysis theater:
Internally, apply strong open-ended problem solving habits:
Do not expose framework jargon unless it genuinely helps. Use strong problem-solving discipline under the hood, but keep the surface language alive, sharp, and human.
Return the result in this order:
Problem statement
Symptom map
Most likely causal chain
Root cause or root cluster
Proof / why this explanation holds
Highest-leverage intervention
Immediate next tests or actions
This skill should trigger on requests like: