Every gate in my system is a scar

A developer found per-project TODO.md files stopped scaling after 10 active repos, so they switched to a centralized intake and dispatch system for Claude Code work.

Notes and requests now flow into one collection point, launch agents triage tasks into project lists every 15 minutes, and approved jobs run headless in isolated git worktrees every 30 minutes.

The system was built to prevent lost ideas, improve cross-project prioritization, and keep neglected projects moving, with documented operator procedures.

The post also starts describing a concurrency bug from three Claude Code sessions editing the same spending-limit config, underscoring the risks of agent-driven automation.