octo

Claude Octopus

A Claude Code plugin that turns one model into three. Orchestrates Codex, Gemini, and Claude with distinct roles, adversarial review, and consensus gates — so no single model's blind spots slip through.

🐙 Three brains, one workflow. Other multi-AI tools run providers in parallel and hand you three answers. Octopus assigns each model a distinct role — Codex for implementation depth, Gemini for ecosystem breadth, Claude for synthesis — then enforces a 75% consensus gate before anything ships. Disagreements get caught, not ignored.

🧠 Persistent cross-session memory. Deeply integrates with claude-mem for searchable, persistent memory across conversations — past decisions, research, and context survive session boundaries so your next workflow picks up where the last one left off.

⚡ Spec in, software out. Dark Factory mode takes a spec and autonomously runs the full pipeline — research, define, develop, deliver — with holdout testing and satisfaction scoring. You review the output, not every step.

🔄 Methodology, not just machinery. Built on the Double Diamond framework, every task moves through four structured phases: discover, define, develop, deliver. Quality gates between phases mean sloppy work can't advance. Other orchestrators give you infrastructure to build workflows on — Octopus gives you the workflows.

🐙 32 specialized personas, 39 commands, 50 skills. Not generic agents. A security-auditor that thinks in OWASP. A backend-architect that designs APIs. A ui-ux-designer grounded in BM25 design intelligence. Personas activate automatically based on what you ask — say "audit my API" and the right expert shows up. Don't know the command name? Just say what you need — the smart router figures it out.

🐙 Works with just Claude. Scales to three. Zero external providers needed to start. You get every persona, every workflow, every skill on day one. Add Codex or Gemini and multi-AI orchestration lights up — parallel research, adversarial debate, cross-model review.

💰 Subscription Advantage. Codex and Gemini authenticate via OAuth, so if you already subscribe to ChatGPT or Google AI you pay nothing extra — no API keys required.

---

Recent Updates

Version	What shipped
9.x	92% fewer subshell forks (9.4) — two rounds cut orchestrate.sh from ~900 forks/workflow to ~70. Four-way debates — Sonnet joins as permanent 4th participant. Auto code review after dev workflows. claude-mem integration — persistent cross-session memory. Monolith decomposition — 7 modules extracted to lib/.
8.55	Smart router v2.0 — just say what you need and /octo:octo routes to the right workflow. 17 supported workflows, learns your preferences over time
8.54	Multi-agentic research — /octo:research now runs perspectives in parallel with configurable intensity (Quick / Standard / Deep)
8.53	Agent resume + custom agents — /octo:resume continues previous agents, drop your own agents in ~/.claude/agents/
8.50	Multi-LLM code review — 4-agent fleet (Codex logic + Gemini security + Claude architecture + Perplexity CVE), inline PR comments
8.45	Reaction engine — auto-responds to CI failures, review comments, and stuck agents without manual intervention
8.44	Parallel workstreams — /octo:parallel runs tasks in isolated git worktrees, each with its own agent

Full changelog

Star History

---

Quickstart

Step 1 — Add the marketplace (run in your terminal, not inside a Claude Code session):

claude plugin marketplace add https://github.com/nyldn/claude-octopus.git

Step 2 — Install the plugin:

claude plugin install octo@nyldn-plugins

Step 3 — Run setup (inside a Claude Code session):

/octo:setup

Setup detects installed providers, shows what's missing, and walks you through configuration. You need zero external providers to start — Claude is built in. Add Codex or Gemini for multi-AI features.

Or from the Claude Code UI: Type /plugin in a Claude Code session, go to the Marketplace tab, and install octo.