GitHub Skill

task-classifier

Classify task complexity (1-10) to route to optimal model. Use when starting any non-trivial task to determine resources and approach. Triggers include: 'classify', 'complexity', 'what model', 'how complex'.

alfredolopez80☆ 74 仓库 Stars14 仓库 Forks

安装与使用

请根据来源仓库和你的 AI 工具说明安装此技能。先阅读技能文件和依赖要求,再添加到工作流。

查看来源 ↗使用指南

技能说明

Skill: Task Classifier

ultrathink - Take a deep breath. We're not here to write code. We're here to make a dent in the universe.

The Vision

Classification should make the path forward feel inevitable.

Your Work, Step by Step

  1. Assess scope: Files, modules, and integrations.
  2. Assess risk: Security, breakage, rollback.
  3. Assess ambiguity: Unknowns and trade-offs.
  4. Assign complexity: Recommend models and iterations.

Ultrathink Principles in Practice

  • Think Different: Avoid defaulting to mid complexity.
  • Obsess Over Details: Map true blast radius.
  • Plan Like Da Vinci: Evaluate before judging.
  • Craft, Don't Code: Keep reasoning concise.
  • Iterate Relentlessly: Reclassify as scope changes.
  • Simplify Ruthlessly: Choose the simplest valid path.

Purpose

Classify task complexity (1-10) to route to optimal model.

Complexity Matrix

Score Complexity Examples Recommended Model
1-2 Trivial Typo fix, comment update MiniMax-lightning
3-4 Simple Add field, simple function MiniMax M2.1
5-6 Medium New API endpoint, refactor Sonnet → Codex/Gemini
7-8 High Multi-file feature, migration Opus → Sonnet → CLIs
9-10 Exceptional Architecture redesign, security fix Opus (thinking)

Classification Criteria

Technical Factors
  • Files affected (1 = trivial, 10+ = exceptional)
  • Cross-module dependencies
  • Database schema changes
  • External API integrations
Risk Factors
  • Security implications (auth, payments, data)
  • Breaking change potential
  • Rollback difficulty
Cognitive Factors
  • Novel problem (no existing patterns)
  • Ambiguity in requirements
  • Trade-off decisions required

Output Format

{
  "complexity": 7,
  "reasoning": "Multi-file feature with database changes and auth integration",
  "recommended_model": "opus",
  "delegation": ["codex:security", "gemini:integration-tests"],
  "estimated_iterations": 12,
  "risk_level": "medium"
}

Model Routing

Complexity Primary Secondary Fallback
1-2 MiniMax-lightning - -
3-4 MiniMax M2.1 - -
5-6 Sonnet Codex/Gemini MiniMax
7-8 Opus Sonnet + CLIs MiniMax
9-10 Opus (thinking) Codex Gemini

Iteration Limits

Model Max Iterations
Claude (Sonnet/Opus) 15
MiniMax M2.1 30
MiniMax-lightning 60

chat Comments (0)

chat_bubble_outline

No comments yet. Be the first to share your thoughts!