Skip to content
FunCoding

problem-classifier

Classify each parsed mathematical-modeling subquestion by required output and structure, surface ambiguous framing trade-offs for human choice, and record primary/secondary task types without selecting algorithms.

Star
1,057
Category
Other
License
MIT
Last commit
Sep 24, 2026

plugins/mathmodeling-skills/skills/problem-classifier/SKILL.md

View on GitHub

Preconditions

  • planning/parse/problem_parse.json exists and maps every Qx to an output.
  • Material framing ambiguities are visible.

Read legacy parse paths only during migration.

Task Types

  • evaluation/ranking;
  • prediction/estimation;
  • optimization/decision;
  • mechanism/dynamics;
  • classification/clustering;
  • graph/routing/network;
  • simulation/scenario;
  • descriptive/inference;
  • mixed.

Detailed cues are in references/task-type-guide.md.

Workflow

  1. Classify from the required output, decision structure, constraints, and relationships—not keywords alone.
  2. Assign:
    • primary type;
    • optional secondary type;
    • confidence;
    • evidence from the parse;
    • consequences for validation and deliverables.
  3. Identify mixed or ambiguous framings that would change what the team can claim.
  4. For a load-bearing ambiguity, invoke one choice card explaining consequences. Do not silently settle it.
  5. Save planning/classification/problem_classification.json.
  6. Record the human framing decision in methods/Qx/qx_decisions.jsonl, or in planning/framing_decisions.jsonl when the Qx method directory does not yet exist.

Output Contract

{
  "schema_version": 1,
  "subquestions": [
    {
      "id": "Q1",
      "primary_type": "evaluation",
      "secondary_type": null,
      "confidence": "high",
      "evidence": [],
      "required_validation": [],
      "framing_decision_id": null,
      "risks": []
    }
  ]
}

Rules

  • Do not propose or choose methods.
  • Do not classify only from nouns such as “forecast” or “optimal”; verify the required output.
  • A subquestion may be mixed, but avoid listing many types without prioritization.
  • Human framing is required when alternative classifications lead to materially different outputs or claims.
  • Do not create a long taxonomy report when the JSON record is sufficient.

Verification

  • Every Qx has one primary type.
  • Mixed/secondary types are justified.
  • Classification evidence resolves to the parse.
  • Ambiguous framing is human-confirmed or remains a blocker.
  • No algorithm selection leaked into classification.

Reference

  • references/task-type-guide.md

Similar Skills

claude-handoff

MAmattpocock/skills288k10,178 new stars in the last 7 days

agent-payment-x402

AFaffaan-m/ECC278k4,046 new stars in the last 7 days

agent-self-evaluation

AFaffaan-m/ECC278k4,046 new stars in the last 7 days

bengali-nlp

AFaffaan-m/ECC278k4,046 new stars in the last 7 days

ponytail-gain

DIDietrichGebert/ponytail162k5,405 new stars in the last 7 days

ponytail-debt

DIDietrichGebert/ponytail162k5,405 new stars in the last 7 days