questionably-ultrathink-skill_skill

Overview

This skill orchestrates rigorous, high-assurance reasoning for complex problems by combining Atom-of-Thought decomposition with isolated fact verification. It structures problems into independent atoms, generates claims and verification questions, then runs verifiers in isolation before synthesizing a verified final answer.

How this skill works

The orchestrator builds a DAG of atomic questions, then a claim generator produces claims and pre-creates verifier files (questions only). Independent verifier agents research each verification question with zero context about the claims. A verification maintainer cross-checks verifier outputs against claims and synthesizes a final answer. Background execution lets you run verifiers in parallel to improve throughput.

When to use it

• Complex multi-part questions requiring systematic decomposition
• Architecture, security, or high-stakes technical decisions
• When you must verify facts independently and avoid confirmation bias
• Debugging issues that need isolated evidence and traceable verification
• Requests containing triggers like “be thorough”, “verify”, or “are you sure”

Best practices

• Always ask clarifying questions first if intent, scope, or success criteria are unclear
• Select the analysis rigor upfront (Standard, Thorough, High-Stakes) based on consequences
• Run claim generators in parallel but block until claims.md exists to get claim counts
• Spawn verifiers as background tasks so independent atoms can progress concurrently
• Keep task IDs and monitor verifier files (look for an “# Answer” section) rather than relying on shared context

Example use cases

• Evaluating trade-offs and verified facts for a cloud security architecture decision
• Breaking a complex debugging problem into atoms and independently verifying root-cause hypotheses
• Producing a traceable, evidence-backed final answer for compliance or policy analysis
• Running parallel verification for multiple independent design options to speed decision-making
• High-stakes code review where independent fact-checking prevents bias in conclusions