| name | mathflow |
| description | Use when mathematical research work needs stage-aware guidance before proceeding. |
Use when
Use this skill when a task involves mathematical research, derivations, mathematical modeling, simulation, numerical experiments, theorem exploration, or conclusion-building across multiple stages of analysis.
Inputs
- The problem statement, known constraints, and current evidence.
- Any existing model, equations, assumptions, datasets, or intermediate results.
- The current stage if the user already knows it, otherwise enough context to assess the stage.
Outputs
- A brief stage assessment that identifies the current stage and why it fits.
- One required next stage by default.
- A short ranked candidate set only when stages genuinely overlap and there is real ambiguity.
- A clear handoff describing what to do in the selected next stage.
mathflow selects the next needed stage skill on demand, one stage at a time.
mathflow is not an eager loader for the whole skill family; load only the stage skill needed for the next step.
Hard rules
- Always assess the current stage before choosing the next step.
- Route mathematical research tasks into the correct stage instead of jumping directly to calculations or conclusions.
- Choose exactly one required next stage by default.
- Return a short ranked candidate set only when stages genuinely overlap.
- Start open-ended work in
problem-analysis.
- Require
mathematical-modeling before introducing new equations or assumptions.
- Require
derivation-and-proof-checking when analytical claims depend on a derivation, proof attempt, or justification audit.
- Require
research-planning before numerical work.
- Require
numerical-experimentation for simulation, solver, sweep, or computational execution work.
- Require
result-validation before making strong claims.
- Require
result-validation before self-audit-loop; result-validation prepares work for audit.
- Require
self-audit-loop as the final skepticism gate before report-writing.
- Route validated final writeups to
report-writing only after self-audit-loop completes.
Behavior
- Read the task and identify whether the work is open-ended, model-building, numerical, interpretive, or near-conclusion.
- Produce a short stage assessment before naming any next action.
- If the task is open-ended or underspecified, route it to
problem-analysis.
- If new equations, abstractions, or assumptions are needed, require
mathematical-modeling first.
- If the work depends on checking or extending a derivation, proof, or mathematical argument, route it to
derivation-and-proof-checking.
- If the user wants computation, simulation, estimation, or parameter sweeps, require
research-planning first unless that stage is already complete.
- After planning is complete, route execution work to
numerical-experimentation.
- Before endorsing a result, require
result-validation.
- In second-wave mode, require
result-validation to prepare work for audit through special-case, limit-case, sensitivity, and consistency checks before any final takeaway.
- Require
self-audit-loop as the explicit final skepticism gate before report-writing.
- After validation and self-audit support the claims strongly enough for communication, route final writeup work to
report-writing.
Stages
problem-analysis: clarify the question, constraints, success criteria, and missing information.
mathematical-modeling: define variables, assumptions, structures, and candidate mathematical representations.
derivation-and-proof-checking: review derivations, proofs, and mathematical arguments so justified results stay separated from heuristics, conjecture, or unsupported steps.
research-planning: choose experiment objectives, baselines, metrics, stop rules, and fallback branches before numerical execution.
numerical-experimentation: run reproducible simulations, sweeps, solvers, or computational studies while retaining anomalies and report-ready evidence.
result-validation: test sensitivity, consistency, edge cases, special cases, limit cases, and alternative explanations so the evidence is prepared for audit.
self-audit-loop: run the explicit final skepticism gate before report-writing by searching for failure regions, overreach, and unsupported stretches.
report-writing: turn staged outputs into a final report that preserves claim strength, caveats, and evidence boundaries.
Routing and handoffs
mathflow is a single-step router: it selects the next needed stage skill on demand instead of eager-loading every child skill.
- A typical second-wave path is
problem-analysis -> mathematical-modeling -> derivation-and-proof-checking or research-planning -> numerical-experimentation -> result-validation -> self-audit-loop -> report-writing, but only the stages required by the actual task should be used.
derivation-and-proof-checking may follow mathematical-modeling directly when the next need is analytical justification rather than numerical execution.
research-planning feeds numerical-experimentation; do not collapse planning and execution into one stage.
result-validation prepares work for audit; self-audit-loop is the final skepticism gate before report-writing.
- If
self-audit-loop finds insufficient support, route back to mathematical-modeling, derivation-and-proof-checking, research-planning, or numerical-experimentation based on the source of the problem.
- Use
report-writing only after the necessary validation and self-audit work is complete and the evidence is ready to be communicated honestly.