jeffersonengcivill/claude-octopus — reverse-engineered prompt

Build me an installable Claude Code plugin called Claude Octopus. I want it to help people use several AI models on the same coding task so one model’s blind spots get caught before code is shipped.

The user should be able to ask for research, planning, building, debugging, reviewing, or shipping, and the plugin should route the request to the right workflow. It should support Claude first, then let people add Codex, Gemini, Copilot, Qwen, Ollama, Perplexity, OpenRouter, and OpenCode when available. Add a simple consensus check so disagreements between models are clearly flagged.

Include a setup flow that detects what providers are installed, explains what is missing, and works even if the user only has Claude. I also want slash commands, reusable skills, and role based helpers like security auditor or backend architect. Add a four phase flow, Discover, Define, Develop, Deliver, with review gates between phases.

Please include docs, tests, and basic Cursor MCP support too. Look up current Claude Code plugin docs online if you need to.