snowmead/questionably-ultrathink
Overview
This skill orchestrates advanced reasoning by combining Atom of Thoughts (AoT) decomposition with Chain of Verification (CoVe) self-checks. It factors problems into isolated atomic questions, solves each in a fresh, constrained context, and contracts answers back into dependent questions to produce a verified synthesis. Use it for complex, high-stakes, or multi-part technical decisions where auditability and rigorous verification matter.
How this skill works
First, it generates a DAG of atomic questions without solving them. Each atom is solved by a fresh isolated solver that sees only that single question; the solver returns an answer plus a self-verification trace and confidence. A maintainer then rewrites dependent atoms with contracted "Given..." contexts. Levels are solved in parallel and iterated until the final synthesis atom is solved.
When to use it
- Complex multi-part engineering or architecture decisions
- Security design reviews and high-stakes technical choices
- Cases requiring rigorous factual verification and audit trails
- Debugging problems that benefit from isolated hypothesis checks
- When the user explicitly requests thoroughness or verification
Best practices
- Always clarify intent and success criteria before building the graph
- Select an appropriate rigor level: Standard, Thorough, or High-Stakes
- Pass only the atom question text to each atomic solver to avoid contamination
- Run atoms in parallel per level and preserve full verification traces
- Use the judge agent for an extra QA pass on High-Stakes workflows
Example use cases
- Evaluate trade-offs between microservice deployment strategies with verification of assumptions
- Audit a security design by decomposing threats and verifying mitigations independently
- Diagnose an intermittent production bug by isolating and verifying root-cause hypotheses
- Produce a provably vetted synthesis for a compliance or architecture memo
- Verify factual claims in multi-source research before presenting conclusions
FAQ
Each atomic solver is spawned fresh and receives only the single question text; it has no visibility into other atoms, IDs, or session data to prevent cross-contamination.
When should I choose High-Stakes rigor?
Choose High-Stakes for production or security decisions where any atom below 0.7 confidence must be re-solved and an optional judge evaluates coherence across atoms.
2 skills
This skill orchestrates UltraThink reasoning with isolated atomic solvers to rigorously verify complex problems end-to-end.
This skill orchestrates isolated fact-checked reasoning for complex problems, ensuring rigorous verification and unbiased final answers.
