nyldn/claude-octopus

Claude Octopus – Multi‑LLM Council for Claude Code

What it is

• An open‑source plugin for Claude Code (Anthropic’s IDE‑style coding assistant) that lets you bring in up to twelve external LLM providers (e.g., Codex, Copilot, Ollama, Qwen, Perplexity, OpenRouter, etc.) alongside the built‑in Claude model.
• The plugin stays dormant until you explicitly invoke a slash command (/octo:*). Ordinary Claude prompts run as usual; Octopus only runs when you need a second opinion or a structured multi‑model review.

Key capabilities

Feature	What it does
Multi‑model council	/octo:council runs a deliberation of 3‑7 personas (Claude Opus, GPT‑5.6, Claude Sonnet, etc.) with configurable goal modes (advice, decision, plan, implement, review) and styles (balanced, adversarial, red‑team, executive). A 75 % consensus gate flags disagreements before code is shipped.
Persistent memory	Integrates with claude‑mem, agentmemory, and deja‑vu so decisions, research, and context survive across sessions.
Dark‑Factory mode	Takes a high‑level spec and autonomously runs the full pipeline – research, design, development, and delivery – presenting only the final output for your review.
Specialized personas & skills	31 built‑in roles (security auditor, backend architect, etc.), 54 slash commands, and 63 reusable workflow modules that you can mix‑and‑match.
Provider‑by‑provider activation	You can add providers one at a time; each becomes active only inside an explicit Octopus workflow. Zero external providers are required to start – Claude alone is enough.
Safety & cost guards	Consensus gates, budget caps, token‑compression, and optional peer‑check routing (OCTOPUS_PREMIUM_PEER_CHECK).
CLI & IDE support	Works as Claude Code slash commands, as a Codex plugin ($skill‑doctor, $skill‑debug), and as an MCP server for Cursor IDE (exposes tools like octopus_discover).
Health & diagnostics	Commands such as octopus doctor, octopus repair, octopus security-audit let you inspect installation state, provider readiness, and audit the plugin files.

Typical workflow

1. Setup – claude plugin marketplace add … && claude plugin install octo@nyldn-plugins then run /octo:setup inside Claude Code.
2. Normal work – Use Claude’s native /init, /review, /security-review commands.
3. Escalate – When you need a second opinion, run a command like /octo:council --goal decision "Should we split this monolith?" or /octo:embrace for a full Discover→Define→Develop→Deliver cycle.
4. Review – Octopus aggregates the responses, applies consensus gates, and presents a unified plan or code diff. You can still invoke individual provider tools (/octo:debug, /octo:research, etc.).

Installation quick‑start

# From a terminal (outside Claude Code)
claude plugin marketplace add https://github.com/nyldn/plugins.git
claude plugin install octo@nyldn-plugins

# Inside Claude Code
/octo:setup   # detects providers, writes config, and finishes installation

• Supports Linux/macOS natively; Windows via WSL or SSH.
• For Codex CLI: codex plugin marketplace add … && codex plugin add claude‑octopus@nyldn-plugins.
• For Cursor IDE: clone the repo, install the MCP server (npm install), and add it to ~/.cursor/mcp.json.

Maturity

• Actively maintained (latest release v11.14.2, CI badge shows tests passing).
• Extensive changelog with versioned feature additions, migration guides, and bug‑fixes.
• Requires Claude Code v2.1.14+ (tracks up to 185 capability flags, compatible with v2.1.284).

License

• MIT License – free to use, modify, and redistribute.

Who might benefit?

• Teams that rely on Claude Code but want independent validation from other LLMs before merging code.
• Developers building safety‑critical or security‑sensitive software who need adversarial review.
• Anyone looking for an “auto‑research‑to‑code” pipeline that can be turned on only when needed, without paying for every provider on every prompt.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.