Hey HN, I released v2.1.0 of Claude Code Orchestrator today.
Quick background: This tool lets you run multiple Claude Code sessions in parallel on isolated git worktrees. You give it a project description, it generates a PRD, spawns workers, monitors them, reviews the merged work, and delivers working code. Based on Boris Cherny's patterns (he created Claude Code).
What's new in v2.1: Quality agents now run at three checkpoints instead of just one.
Before: Agents only ran when a PR was ready for review.
Now:
1. Pre-PR (Worker): type-check, lint, test must pass BEFORE creating a PR
2. Post-CI (Orchestrator): QA Guardian, security scan (npm audit), code simplifier, DevOps review
3. Post-Merge (Planner): All agents run again on the combined code
This catches issues that only appear when parallel workers' code is merged together.
Other changes:
- Security scan now runs on ALL PRs (not just infra changes)
- Code simplifier threshold lowered from 100+ to 50+ lines
- Fixed shellcheck warnings (the irony of an automation tool failing lint)
The workflow:
/project "Add OAuth authentication"
# Walk away
# Come back to working code + macOS notification
GitHub: https://github.com/reshashi/claude-orchestrator
Requirements: macOS, iTerm2, Claude Code CLI, jq
Happy to answer questions about the architecture or agent coordination patterns.
Hey HN, I released v2.1.0 of Claude Code Orchestrator today.