fp.

  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.