A portable Claude Code bundle: one skill + five subagents that coordinate a multi-step change as a DAG of parallel PRs. Repo-agnostic.
multi-agent-orchestration/
├── .claude-plugin/plugin.json
├── agents/ # subagents (pinned model/effort/isolation)
│ ├── mao-planner.md
│ ├── mao-builder.md
│ ├── mao-designer.md
│ ├── mao-reviewer.md
│ └── mao-audit.md
├── skills/multi-agent-orchestration/ # the skill
│ ├── SKILL.md
│ └── references/
│ ├── planning-protocol.md
│ ├── design-protocol.md
│ ├── environment.md
│ ├── deviations-and-progress.md
│ └── templates.md
├── install.sh # symlinks contents into ~/.claude/{agents,skills}/
└── README.md
./install.shThe script creates symlinks:
~/.claude/agents/mao-*.md→<bundle>/agents/mao-*.md~/.claude/skills/multi-agent-orchestration→<bundle>/skills/multi-agent-orchestration
After installation Claude Code discovers both the skill and the agents on next session start.
- Clone or copy this folder to wherever you keep tools (e.g.
~/.claude/plugins/local/multi-agent-orchestration/). - From inside the folder, run
./install.sh. - Done. The agents and the skill are now available in any Claude Code session on that machine.
./install.sh --uninstallRemoves the symlinks. The bundle files are untouched.
The skill is documented in skills/multi-agent-orchestration/SKILL.md. Two modes:
- Planning:
create multi-agent-orchestration-plan for <description>— fires themao-plannersubagent in an isolated worktree; it ships a PR with aplan.md. - Execute:
execute multi-agent-orchestration for <path>— the coordinator (your chat session) reads the merged plan and dispatches builders / designers / reviewers / audit, all as isolated-worktree subagents.
All five subagents pin model: opus, effort: xhigh, and isolation: worktree. Edit the agent files directly to override.
Not currently licensed. Personal use.