Model Selection Matrix + Context Budgeting

[kanfil]

Summary

Epic issue for Tier 3 Model Selection Strategy - Complete Overview

Sub-items

Tier 3.1 - Core & Admin Command Model Selection Matrix

• ❌ Core Workflow Commands: Define defaults and options for /specify, /clarify, /plan, /implement, /analyze, /trace, /levelup
• ❌ Admin Commands: Define defaults for /architect, /constitution, /mode, /checklist
• ❌ Integration: Add --model parameter to all commands above

Tier 3.2 - Model-Aware Context Budgeting

• ❌ Context Window Limits: Document effective context windows per model
• ❌ Token Cost Calculator: Implement utility to estimate tokens for spec.md + plan.md + tasks.md
• ❌ Recommendation Engine: Show context utilization during execution
• ❌ Model Downgrade Suggestions: Auto-suggest Haiku when context > 70%
• ❌ Transparency in Prompts: Log context size in delegation prompts

Tier 3.3 - Two-Model Review Strategy

• ❌ Review Model Configuration: Store preferred review model in config.json
• ❌ Quality Gate Integration: Invoke review model for ASYNC task validation
• ❌ SYNC vs ASYNC Review: Apply two-model review only to critical SYNC tasks
• ❌ User Override: Support --review-model parameter to force specific reviewer
• ❌ Review Quality Thresholds: Document when review model is invoked

Tier 3.4 - Cost Optimization Guidance

• ❌ Usage Documentation: Create guide on when to use each tier (Haiku, Sonnet, Opus)
• ❌ Cost Calculator: Implement simple calculator with examples
• ❌ Budget Tracking: Optional logging of model usage per project/team
• ❌ Org Policy Templates: Provide templates for enforcing model choices per team

Tier 3.5 - Model Preference in Team Directives

• ❌ Directive Format: Define structure in team-ai-directives for model preferences per command
• ❌ Fallback Hierarchy: Implement precedence: user CLI override → project config → team directive → hardcoded default
• ❌ Org-Wide Defaults: Enable shared model policies across projects via team directives

Tier 3.6 - Command-Level --model Parameter Implementation

• ❌ CLI Parameter Addition: Add --model to all command templates
• ❌ Config Validation: Validate model choice against available models
• ❌ Preference Persistence: Store user's recent model choice per command
• ❌ Documentation: Update quickstart and command help text

Priority

Medium - Optimize cost & performance

Labels

epic, model-selection, cost

---

This issue consolidates all Tier 3 Model Selection Strategy tasks. Original part 2 task (#30) has been merged here.