Models Setup
To balance cost, performance, and reasoning capabilities, this workflow uses three Claude model tiers.
Model Tiers
| Tier | Model | Characteristics | Used by |
|---|---|---|---|
| High-Reasoning | Claude Opus | Deep architecture analysis, complex planning, orchestration | Conductor |
| Production | Claude Sonnet | Code review, structured analysis, standard triage | Code Reviewer, Project Manager, OSS Triage Auditor |
| Utility | Claude Haiku | Lightweight monitoring, simple fixes, fast responses | Delivery Manager |
Recommended Plans
1. The "Eco" Plan (Budget-First)
Best for: Side projects, small utilities, or early-stage experiments.
- Planning & Research: Production (Sonnet)
- Implementation: Production (Sonnet)
- QA/Review: Production (Sonnet)
- Estimated Cost: ~$20/month (Claude Pro subscription)
2. The "Balanced" Plan (Standard OSS)
Best for: Production-grade libraries with regular contributions.
- Planning & Research: High-Reasoning (Opus)
- Implementation: Production (Sonnet)
- QA/Review: Production (Sonnet)
- Estimated Cost: ~$100/month (Claude Max subscription)
3. The "Power" Plan (High-Velocity)
Best for: Complex systems, security-critical code, or high-traffic projects.
- Planning & Research: High-Reasoning (Opus)
- Implementation: High-Reasoning (Opus) + Human verification
- QA/Review: High-Reasoning (Opus)
- Estimated Cost: $200+/month (heavy API usage)
Configuration
Settings are controlled via:
CLAUDE.md— Project-wide instructions, agent roster, hook-first rule.mcp.json— MCP server configuration, default model.claude/agents/*.md— Per-agent model tier (set in frontmatter).claude/project-config.md— Runtime constants (limits, labels, repo)