model-selection

vamseeachanta

Mis à jour Today

69 vues

Autre

À propos

Cette compétence offre un guide automatisé de sélection de modèle pour Claude Code, basé sur la complexité de la tâche, le coût et les besoins en latence. Elle propose un arbre de décision et un tableau de référence rapide pour aider les développeurs à choisir efficacement entre les modèles Opus, Sonnet et Haiku. Utilisez-la au début de nouvelles tâches pour optimiser les performances et l'utilisation des ressources.

Installation rapide

Claude Code

Recommandé

Principal

npx skills add vamseeachanta/workspace-hub

Commande PluginAlternatif

/plugin add https://github.com/vamseeachanta/workspace-hub

Git CloneAlternatif

git clone https://github.com/vamseeachanta/workspace-hub.git ~/.claude/skills/model-selection

Copiez et collez cette commande dans Claude Code pour installer cette compétence

Documentation

Model Selection Skill

Version: 1.0.0 Category: Optimization Triggers: Starting tasks, choosing Claude model, usage optimization

Quick Reference

Model Selection Decision Tree

NEW TASK
    │
    ├── WORK REPO + COMPLEX → OPUS
    ├── WORK REPO + STANDARD → SONNET
    ├── PERSONAL + SIMPLE → HAIKU
    └── DEFAULT → SONNET

Quick Selection Guide

Model	Target %	Use For
OPUS	30%	Architecture, multi-file refactoring (>5 files), security review
SONNET	40%	Standard implementations, code review, documentation
HAIKU	30%	Quick queries, status checks, simple operations

Automated Model Suggestion

# Get model recommendation before each task
./scripts/monitoring/suggest_model.sh <repository> "<task description>"

# Examples:
./scripts/monitoring/suggest_model.sh digitalmodel "Design authentication architecture"
# → Recommends: OPUS (complexity score: 4)

./scripts/monitoring/suggest_model.sh digitalmodel "Implement user login"
# → Recommends: SONNET (complexity score: 1)

./scripts/monitoring/suggest_model.sh hobbies "Quick file check"
# → Recommends: HAIKU (complexity score: -3)

Complexity Scoring

Algorithm evaluates:

Keywords - architecture/refactor → +3, implement/feature → +1, check/status → -2
Repository Tier - Work Tier 1 → +1, Personal → -1
Task Length - >15 words → +1, <5 words → -1

Score Mapping:

Score ≥3: OPUS
Score 0-2: SONNET
Score <0: HAIKU

Repository Tiers

Work Repositories

Tier 1 (Production): 60% Opus, 30% Sonnet, 10% Haiku

digitalmodel, energy, frontierdeepwater

Tier 2 (Active): 30% Opus, 50% Sonnet, 20% Haiku

assetutilities, worldenergydata

Tier 3 (Maintenance): 10% Opus, 30% Sonnet, 60% Haiku

doris, saipem, OGManufacturing

Personal Repositories

Active: 20% Opus, 40% Sonnet, 40% Haiku Experimental: 5% Opus, 25% Sonnet, 70% Haiku Archive: 0% Opus, 20% Sonnet, 80% Haiku

Usage Monitoring

Check before starting work: https://claude.ai/settings/usage

Alert Thresholds:

Sonnet >70% → Switch to Opus/Haiku
Session >80% → Batch work or wait
Overall >80% → Defer non-critical

OPUS Use Cases

✅ Multi-file refactoring (>5 files) ✅ Architecture decisions ✅ Complex algorithm design ✅ Security-critical code review ✅ Cross-repository coordination ✅ Performance optimization strategies

SONNET Use Cases

✅ Standard feature implementation ✅ Code review (single PR) ✅ Documentation writing ✅ Test generation ✅ Bug fixing (standard complexity) ✅ Configuration updates

HAIKU Use Cases

✅ File existence checks ✅ Simple grep/search operations ✅ Quick status updates ✅ Log analysis (pattern matching) ✅ Template generation ✅ Format validation

Emergency Protocols

If Sonnet >80%

⛔ STOP using Sonnet immediately
✅ Switch to Opus for critical work
✅ Switch to Haiku for everything else
📅 Defer non-urgent work to Tuesday

If Session >80%

⏸️  Pause AI tasks
⏰ Wait for session reset (~3-4 hours)
📦 Batch work for next session

Full Reference

See: @docs/AI_MODEL_SELECTION_AUTOMATION.md See: @docs/CLAUDE_MODEL_SELECTION_QUICK_REFERENCE.md

Use this when starting tasks, selecting models, or optimizing AI usage.

Dépôt GitHub

vamseeachanta/workspace-hub

Chemin: .claude/skills/ai/optimization/model-selection

Compétences associées

data-science-expert

Autre

This skill provides expert-level data science assistance including statistical analysis, machine learning, and data visualization. It helps developers with tasks like data cleaning, model building, and creating plots using Python libraries like pandas and matplotlib. Use it when you need guidance on EDA, statistical modeling, or visualizing complex datasets.

Voir la compétence

usage-optimization

Autre

This Claude Skill helps developers optimize AI usage efficiency by promoting script-first patterns, batch operations, and prepared input files. It provides effectiveness ratings for different approaches, emphasizing execution over description to maximize productivity. Use it when you need to reduce AI interaction time and get more actionable, automated outputs from Claude.

Voir la compétence

agent-usage-optimizer

Autre

This skill optimizes AI model selection by reading quota states and recommending the best Claude/Codex/Gemini allocation for each task. It provides quota-aware routing and headroom displays, making it ideal for work sessions with multiple queued items or when approaching quota limits. Developers should use it before starting sessions with 3+ work items or when Claude quotas drop below 50% remaining.

Voir la compétence

agent-usage-optimizer-provider-capability-reference

Autre

This reference skill provides a quick comparison table of AI providers (Claude models, Codex, Gemini) with their strengths and ideal use cases. It helps developers optimize agent usage by selecting the most suitable provider for tasks like architecture, coding, or bulk data processing. Use it to make informed decisions on model selection based on task requirements and constraints like quota and cost.

Voir la compétence