oh-my-claudecode

oh-my-claudecode (OMC) is an open-source orchestration layer for Claude Code that aims to remove the learning curve of multi-agent workflows. It is distributed both as a Claude Code plugin/marketplace entry and as an npm CLI published under the name oh-my-claude-sisyphus, which installs the oh-my-claudecode and omc commands. Installation follows two paths. Plugin users add the marketplace and install the plugin, then run /omc-setup inside a session. CLI users install the npm package globally and run omc setup from a terminal. The README notes a known npm warning about a deprecated prebuild-install dependency coming from better-sqlite3, and states it does not by itself indicate a failed install. OMC distinguishes terminal CLI commands (omc ...) from in-session skills (/...). Setup, provider questions, and team orchestration exist on both surfaces but behave differently: omc team launches tmux CLI worker panes, while /team runs an in-session native team workflow. Autopilot, Ralph, Execute, and Deep Interview are described as in-session skills with no corresponding CLI subcommands. The canonical orchestration surface is Team, which runs a staged pipeline: team-plan, team-prd, team-exec, team-verify, and team-fix in a loop. Native teams require enabling an experimental agent-teams environment flag in Claude Code settings; if disabled, OMC warns and falls back to non-team execution where possible. The CLI team runtime can spawn workers for Codex, Gemini, Antigravity, Grok, Cursor, and Claude CLIs, with workers created on demand and terminated when their task completes. Cursor workers can be preferred for autopilot execution through project configuration, and models can be pinned via environment variables or role routing. Other documented modes include Autopilot for autonomous single-lead execution, Execute for persistent verify/fix loops, Verify for evidence-based completion checks, and an artifact-only Ultragoal mode that produces durable goal and evidence artifacts without starting a loop. The README advises using only one primary loop authority per session and cautions against claiming that Claude Code's /goal evaluator independently runs commands or reads files. Additional features include a set of specialized agents with tier variants, model routing between cheaper and more capable models, a HUD statusline for orchestration metrics, analytics and cost tracking, and a skill system. Skills can be project-scoped under .omc/skills or user-scoped under ~/.omc/skills, with project skills taking priority and intended to be committed for team sharing. A /skillify command extracts reusable patterns, and matching skills auto-inject into context. Runtime state, plans, logs, and handoffs are written under .omc/ by default, with .omc/skills/** as the intentional committable exception; OMC_STATE_DIR can relocate state so it survives worktree deletion. The README also documents named autopilot stage profiles configured in omc.jsonc, limited to specific admitted stage sequences and currently requiring Linux with flock. It explicitly states that OMC does not ship a VS Code extension, that the npm package exports TypeScript helpers for local Node programs using the Claude Agent SDK, and that CI/CD automation should prefer deterministic terminal commands rather than interactive slash commands.