# orchestrate

> One entry point, nine subagent-orchestration strategies, for any AI coding agent that reads a SKILL.md. Installs as an agent skill.

orchestrate turns "run this plan with subagents" into a controlled process instead of an improvised
one. Point it at a plan file or a task description and it picks (or you force) a strategy — staged
sequential cycles, parallel worktree fan-out, hierarchical sub-orchestrator fleets, agent teams,
deterministic workflows, goal/Ralph loops, advisor/executor cost splits, adversarial planning, or
external CLIs (Codex/Grok/Cursor/Antigravity/opencode/Hermes/Kimi/Pi) as workers — then coordinates,
dispatches, and gates the subagent work through it. `/orchestrate` auto-triages when no strategy is
forced: it measures task shape (plan present, task count, file-overlap independence, context size,
verifiable stop condition) and states why it picked what it picked. Before the first multi-agent
dispatch it prints a flight plan — the topology tree with model + reasoning effort per agent,
gates, and an honest budget range — and gates on the user's approval (confirm=off skips the gate,
never the print). Claude Code is the reference
host; the skill is plain Agent Skills format (agentskills.io) with a host-adapter layer
(references/shared-hosts.md), so it runs on Codex, Cursor, Antigravity, opencode, Grok Build,
Hermes, Kimi Code CLI, and Pi too — detecting its host at kickoff and degrading honestly where a
primitive is missing.

## Install

- npm/skills.sh: `npx skills add gabros20/orchestrate-skill -g`
- Clone + installer: `git clone https://github.com/gabros20/orchestrate-skill.git && cd orchestrate-skill && ./install.sh codex`
- Codex runtime destination: `${CODEX_HOME:-$HOME/.codex}/skills/orchestrate`

## Docs

- [README.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/README.md): overview, install, composition presets, saved aliases, repo map.
- [skills/orchestrate/SKILL.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/skills/orchestrate/SKILL.md): the router — command grammar, strategy catalog, universal rules.
- [docs/installation.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/docs/installation.md): step-by-step install and verification.
- [docs/usage.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/docs/usage.md): invocation grammar, dimensions, and resuming runs.
- [docs/strategies.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/docs/strategies.md): all nine strategies in depth — when to use each, their mechanics.
- [docs/recipes.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/docs/recipes.md): composed dimension recipes and saved aliases.
- [docs/research/foundations.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/docs/research/foundations.md): the evidence and sources behind the design choices.
- [docs/research/token-optimization.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/docs/research/token-optimization.md): research record on token/context/cost optimization for multi-agent orchestration.

## Optional

- [docs/designs/v1.0.0-initial-architecture.md](https://raw.githubusercontent.com/gabros20/orchestrate-skill/main/docs/designs/v1.0.0-initial-architecture.md): implementation design shipped in v1.0.0 — why the skill is shaped the way it is.
- [Visual guide](https://orchestrate-skill.vercel.app): every strategy, dimension, and alias on one page.