v1.0.0 - Stable Production Release

🎉 Stable Release

House Agents moves to v1.0.0 after successful production validation. All core agents tested and performing with 95-99% token savings in real-world usage.

Production Metrics:

• 187,000 tokens quarantined across operations
• ~5,000 tokens returned to main context
• 95-99% token reduction across all agents

Added

• house-mcp agent - MCP tool configuration specialist (4th production agent)
  • Works with all user-configured MCP servers (inherits all MCP tools automatically)
  • Reads verbose documentation and returns minimal working configs
  • Supports any MCP server (context7, notion, puppeteer, etc.)
  • Uses Claude Haiku 4.5 for cost efficiency

Changed

• All sub-agents now use Claude Haiku 4.5 for optimal performance
  • house-research, house-bash, house-git, house-mcp now explicitly use claude-haiku-4-5-20251001
  • 67% cost reduction on sub-agent operations ($1/$5 vs $3/$15 per million tokens)
  • 2x faster response times compared to Sonnet 4
  • 90% of Sonnet 4.5's agentic coding performance
  • Users can override by setting model: inherit in agent files if desired
• Added "Model Architecture" section to README.md explaining model usage strategy
• house-mcp moved from "Future Agents" to production - MCP bug #7296 workaround found
  • Omitting tools: field allows sub-agents to inherit all MCP tools from main session
  • Updated to be MCP-agnostic (works with any configured MCP server)

Removed

• Benchmark suite (benchmark/ directory) - manual testing more practical than automated benchmarks
• test-agents.sh script - superseded by Claude Code's built-in agent validation
• examples/test-and-build.md - example superseded by updated documentation