#llm-orchestration

Claude as subagents in Codex — scoped workers, resumable sessions, git-ignored logs.

Route bulk implementation to cheap executor LLMs (opencode/antigravity/cursor) with Claude as architect + judge - committed failing tests are the merge gate

Local-first kanban board where Claude Code agents work your cards — dependency chains with merge gates, an AI manager that reviews real diffs and merges green PRs, group lanes, CI-gated. One runtime dependency.

Long-running delegation architecture for Claude Code: a delegator main session + full-power orchestrator subagents — nested agents that actually delegate, skill-driven spawning, gates, crash-safe resume. Probe-proven physics + cleanroom test harness.

An operating system for a 32-agent AI software org — charters, a file-based board, a gated lifecycle, and Claude Code orchestration.