[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-74883":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":23,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":38,"readmeContent":39,"aiSummary":40,"trendingCount":16,"starSnapshotCount":16,"syncStatus":41,"lastSyncTime":42,"discoverSource":43},74883,"claude-octopus","nyldn\u002Fclaude-octopus","nyldn","Surface AI blindspots before you ship. Put up to 8 AI models on every research, design or coding task.","https:\u002F\u002Freddit.com\u002Fr\u002FClaudeOctopus\u002F",null,"Shell",3542,331,28,3,0,70,98,233,210,109.56,"MIT License",false,"main",[26,27,28,29,30,31,32,33,34,35,36,37],"ai-agents","ai-orchestration","claude-code","claude-code-plugin","codex","copilot","developer-tools","double-diamond","gemini","multi-ai","multi-llm","ollama","2026-06-12 04:01:16","# 🐙 Claude Octopus\n\nEvery AI model has blind spots. Claude Octopus puts up to eight of them on every task, so blind spots surface before you ship — not after. It orchestrates Codex, Gemini, Copilot, Qwen, Ollama, Perplexity, and OpenRouter alongside Claude Code, with consensus gates that flag any disagreements.\n\n**Claude-native first, Octopus for escalation.** Use Claude-native `\u002Finit`, `\u002Freview`, and `\u002Fsecurity-review` when Claude is enough. Use Octopus when you want multiple model opinions, adversarial review, or stricter multi-LLM workflows.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"docs\u002Fassets\u002Fdemo.gif\" alt=\"Claude Octopus Demo — debate and research with multiple AI providers\" width=\"720\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fclaude.ai\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FClaude-Built_with_AI-c96442?logo=data:image\u002Fsvg%2bxml;base64,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&labelColor=333\" alt=\"Built with Claude\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus\u002Factions\u002Fworkflows\u002Ftest.yml\">\u003Cimg src=\"https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus\u002Factions\u002Fworkflows\u002Ftest.yml\u002Fbadge.svg\" alt=\"Tests\">\u003C\u002Fa>\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTests-117_suites_passing-brightgreen\" alt=\"117 suites passing\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVersion-9.38.0-blue\" alt=\"Version 9.38.0\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FClaude_Code-v2.1.14+_required-blueviolet\" alt=\"Requires Claude Code v2.1.14+\">\n  \u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-green\" alt=\"MIT License\">\n\u003C\u002Fp>\n\n🐙 **Research, build, review, and ship — with eight AI providers checking each other's work.** Say what you need, and the right workflow runs. Claude-native handles the ordinary path; Octopus handles the escalated path. A 75% consensus gate catches disagreements before they reach production. No single model's blind spots slip through.\n\n🧠 **Remembers across sessions.** Integrates with [claude-mem](https:\u002F\u002Fgithub.com\u002Fthedotmack\u002Fclaude-mem) for persistent memory — past decisions, research, and context survive session boundaries.\n\n⚡ **Spec in, software out.** Dark Factory mode takes a spec and autonomously runs the full pipeline — research, define, develop, deliver. You review the output, not every step.\n\n🔄 **Four-phase methodology, not just tools.** Every task moves through Discover → Define → Develop → Deliver, with quality gates between phases. Other orchestrators give you infrastructure. Octopus gives you the workflows.\n\n🐙 **32 specialized personas** (role-specific AI agents like security-auditor, backend-architect), **48 commands** (slash commands you type), **52 skills** (reusable workflow modules). Say \"audit my API\" and the right expert activates. Don't know the command? The smart router figures it out.\n\n🐙 **Works with just Claude. Scales to eight.** Zero providers needed to start. Add them one at a time — each activates automatically when detected.\n\n💰 **Five providers cost nothing extra.** Codex and Gemini use OAuth (included with subscriptions). Qwen has 1,000-2,000 free requests\u002Fday. Copilot uses your GitHub subscription. Ollama runs locally for free.\n\n---\n\n## What's New\n\n| Version | Best Features |\n|---------|--------------|\n| **v9** (current) | Up to 8 providers (Codex, Gemini, Copilot, Qwen, Ollama, Perplexity, OpenRouter, OpenCode). Four-way AI debates. Smart router — just say what you need. Agent summary tables show which providers actually contributed. Provider-aware prompt preflight prevents silent oversize failures. Research breadth modes fan out light, standard, or exhaustive investigations. Setup aliases and fuzzy `\u002Focto:*` corrections reduce command friction. Discipline mode with 8 auto-invoke gates. Two-stage review. Circuit breakers with automatic provider recovery. Cursor + OpenCode + Codex cross-compatibility. Token compression: `bin\u002Focto-compress` pipe + auto PostToolUse hook save ~7,300 tokens\u002Fsession. PostCompact context recovery. `bin\u002Foctopus` CLI. 170+ CC feature flags through v2.1.132. |\n| **v8** | Multi-LLM code review with inline PR comments. Parallel workstreams in isolated git worktrees. Reaction engine — auto-responds to CI failures. 32 specialized personas. Dark Factory autonomous pipeline. |\n| **v7** | Double Diamond workflow. Multi-provider dispatch. Quality gates and consensus scoring. Configurable sandbox modes. |\n\n[Full changelog →](CHANGELOG.md)\n\n## Quickstart\n\n```bash\n# Terminal (not inside a Claude Code session):\nclaude plugin marketplace add https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fplugins.git\nclaude plugin install octo@nyldn-plugins\n\n# Then inside Claude Code:\n\u002Focto:setup\n```\n\nThat's it. Setup detects installed providers, shows what's missing, and walks you through configuration. You need **zero** external providers to start — Claude is built in.\n\nClaude Code **v2.1.14+** is the minimum supported runtime. Newer Claude Code releases unlock additional Octopus diagnostics and release checks automatically; the current plugin tracks feature flags through **Claude Code v2.1.132**.\n\n\u003Cdetails>\n\u003Csummary>Install for Codex CLI\u003C\u002Fsummary>\n\n```bash\ngit clone --depth 1 https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus.git ~\u002F.codex\u002Fclaude-octopus && mkdir -p ~\u002F.agents\u002Fskills && ln -sf ~\u002F.codex\u002Fclaude-octopus\u002Fskills ~\u002F.agents\u002Fskills\u002Fclaude-octopus\n```\n\nRestart Codex. Skills appear automatically — invoke with `$skill-doctor`, `$skill-debug`, etc.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Install for Cursor IDE\u003C\u002Fsummary>\n\nCursor uses Octopus as an **MCP server** (not a plugin — Cursor doesn't have Claude Code's plugin system). You get MCP tools like `octopus_discover`, `octopus_review`, etc. instead of `\u002Focto:*` slash commands.\n\n> **Important:** Just cloning the repo is not enough. You must complete all three steps below — install dependencies and configure the MCP server — for Cursor to pick up Octopus tools.\n\n```bash\n# 1. Clone the repo\ngit clone --depth 1 https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus.git ~\u002F.cursor\u002Fclaude-octopus\n\n# 2. Install MCP server dependencies\ncd ~\u002F.cursor\u002Fclaude-octopus\u002Fmcp-server && npm install\n\n# 3. Configure Cursor — add to ~\u002F.cursor\u002Fmcp.json (global) or .cursor\u002Fmcp.json (per-project):\n```\n\n```json\n{\n  \"mcpServers\": {\n    \"claude-octopus\": {\n      \"command\": \"npx\",\n      \"args\": [\"tsx\", \"${userHome}\u002F.cursor\u002Fclaude-octopus\u002Fmcp-server\u002Fsrc\u002Findex.ts\"],\n      \"env\": {\n        \"OCTO_CLAW_ENABLED\": \"true\",\n        \"OPENAI_API_KEY\": \"${env:OPENAI_API_KEY}\",\n        \"GEMINI_API_KEY\": \"${env:GEMINI_API_KEY}\"\n      }\n    }\n  }\n}\n```\n\nRestart Cursor. Tools appear in Cursor's AI chat — invoke by asking e.g. \"use octopus_discover to research X\".\n\n\u003Cdetails>\n\u003Csummary>Using Cursor on WSL?\u003C\u002Fsummary>\n\nIf you're running Cursor on Windows with WSL, clone the repo inside WSL and point the MCP config through `wsl.exe`:\n\n```json\n{\n  \"mcpServers\": {\n    \"claude-octopus\": {\n      \"command\": \"wsl\",\n      \"args\": [\"npx\", \"tsx\", \"\u002Fhome\u002F\u003Cuser>\u002F.cursor\u002Fclaude-octopus\u002Fmcp-server\u002Fsrc\u002Findex.ts\"],\n      \"env\": {\n        \"OPENAI_API_KEY\": \"${env:OPENAI_API_KEY}\",\n        \"GEMINI_API_KEY\": \"${env:GEMINI_API_KEY}\"\n      }\n    }\n  }\n}\n```\n\nReplace `\u003Cuser>` with your WSL username. Make sure `node` and `npm` are installed inside WSL.\n\u003C\u002Fdetails>\n\nSee [docs\u002FIDE-INTEGRATION.md](docs\u002FIDE-INTEGRATION.md) for the full guide including `ide-attach.sh` auto-setup.\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Install for OpenCode\u003C\u002Fsummary>\n\n```bash\ngit clone --depth 1 https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus.git ~\u002F.opencode\u002Fclaude-octopus\nmkdir -p ~\u002F.agents\u002Fskills\nln -s ~\u002F.opencode\u002Fclaude-octopus\u002Fskills ~\u002F.agents\u002Fskills\u002Fclaude-octopus\n```\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Other install methods (Claude Code)\u003C\u002Fsummary>\n\n**From the Claude Code UI:** Type `\u002Fplugin` in a session → **Marketplace** tab → install **octo**.\n\n**Factory AI (Droid):**\n```bash\ndroid plugin marketplace add https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus.git\ndroid plugin install octo@nyldn-plugins\n```\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>Update \u002F Troubleshooting\u003C\u002Fsummary>\n\n```bash\n# Update\nclaude plugin marketplace update nyldn-plugins\nclaude plugin update octo@nyldn-plugins\n\n# Clean reinstall (if update fails)\nclaude plugin uninstall claude-octopus 2>\u002Fdev\u002Fnull\nclaude plugin uninstall octo 2>\u002Fdev\u002Fnull\nrm -rf ~\u002F.claude\u002Fplugins\u002Fcache\u002Fnyldn-plugins\u002Focto\nclaude plugin marketplace remove nyldn-plugins\nclaude plugin marketplace add https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fplugins.git\nclaude plugin install octo@nyldn-plugins\n```\n\nRun focused diagnostics after updating:\n\n```bash\n\u002Focto:doctor config   # install path, version, manifest, Claude Code feature flags\n\u002Focto:doctor skills   # skill loading, skillOverrides, plugin zip\u002FURL capability notes\n```\n\nFor Anthropic-compatible gateways, Claude Code v2.1.129+ requires an explicit opt-in before `\u002Fmodel` discovers models from `\u002Fv1\u002Fmodels`:\n\n```bash\nexport ANTHROPIC_BASE_URL=https:\u002F\u002Fyour-gateway.example\u002Fv1\nexport CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1\n```\n\nClaude Code v2.1.129+ also supports `skillOverrides` in Claude settings. Use it to keep rarely used Octopus skills installable while reducing context load, for example by setting niche skills to `name-only` or `user-invocable-only`.\n\u003C\u002Fdetails>\n\n---\n\n## Claude Code Web and Remote Sessions\n\nWhen Claude Code is running in a hosted, web, or remote-control environment, set `OCTOPUS_REMOTE_SESSION=true` in that environment. If Claude Code itself exports `CLAUDE_CODE_REMOTE=true` or `CLAUDE_CODE_WEB=true`, Octopus detects that automatically. Remote sessions are treated as unattended by default:\n\n- `CLAUDE_OCTOPUS_AUTONOMY=autonomous` \u002F `OCTOPUS_AUTONOMY=autonomous` unless already set\n- provider smoke tests and Codex tier probes are skipped\n- the statusline uses a lightweight remote-safe display\n\nSet `OCTOPUS_REMOTE_STATUSLINE=full` to opt back into the full local HUD, or `OCTOPUS_REMOTE_STATUSLINE=off` to suppress statusline output entirely.\n\nCloud environment setup should install provider CLIs and expose only the credentials required for the workflow. Paste this into the cloud environment setup script:\n\n```bash\n#!\u002Fusr\u002Fbin\u002Fenv bash\nset -e\n\nnpm install -g @openai\u002Fcodex @google\u002Fgemini-cli @qwen-code\u002Fqwen-code 2>\u002Fdev\u002Fnull || true\n\necho \"Octopus cloud setup:\"\ncommand -v codex >\u002Fdev\u002Fnull 2>&1 && echo \"  Codex CLI: installed\" || echo \"  Codex CLI: missing\"\ncommand -v gemini >\u002Fdev\u002Fnull 2>&1 && echo \"  Gemini CLI: installed\" || echo \"  Gemini CLI: missing\"\ncommand -v qwen >\u002Fdev\u002Fnull 2>&1 && echo \"  Qwen CLI: installed\" || echo \"  Qwen CLI: missing\"\ncommand -v gh >\u002Fdev\u002Fnull 2>&1 && echo \"  GitHub CLI: installed\" || echo \"  GitHub CLI: optional, install if Sentinel needs GitHub\"\n```\n\nSet environment variables in the cloud environment, not in the script:\n\n```bash\nOPENAI_API_KEY=...\nGEMINI_API_KEY=...\nPERPLEXITY_API_KEY=...   # optional\nOPENROUTER_API_KEY=...   # optional\n```\n\nProvider API calls require internet access from the hosted environment.\n\nFor scheduled Claude Code tasks, run `\u002Focto:sentinel` for triage and `\u002Focto:security` for recurring audits. Keep jobs read-only by default and route fixes through `\u002Focto:debug`, `\u002Focto:review`, or `\u002Focto:embrace` after triage.\n\nSet `OCTO_TIER=prototype|mvp|production` as a project hint. It does not hard-block behavior; it helps setup, doctor, and workflow prompts recommend the right amount of verification and provider spend.\n\n---\n\n## 8 Commands That Matter Most\n\n🐙 Eight commands — one per arm. *A real octopus has eight arms, each with its own neurons that can act independently.* These eight tentacles work the same way: each orchestrates up to three AI providers, applies quality gates, and produces a deliverable.\n\n```bash\n\u002Focto:embrace build stripe integration     # Full lifecycle: research → define → develop → deliver\n\u002Focto:factory \"build a CLI that converts CSV to JSON\"  # Autonomous pipeline — spec in, software out\n\u002Focto:debate monorepo vs microservices     # Structured four-way AI debate with consensus\n\u002Focto:research --breadth=standard htmx vs react in 2026  # Attributed multi-provider research\n\u002Focto:design mobile checkout redesign       # UI\u002FUX design with BM25 style intelligence\n\u002Focto:tdd create user auth                 # Red-green-refactor with test discipline\n\u002Focto:security                              # OWASP vulnerability scan + remediation\n\u002Focto:prd mobile checkout redesign          # AI-optimized PRD with 100-point scoring\n```\n\nPlus 30+ more: review, debug, extract, deck, docs, schedule, parallel, sentinel, optimize, brainstorm, claw, doctor, and [the full set](docs\u002FCOMMAND-REFERENCE.md).\n\nDon't remember the command name? Just describe what you need:\n\n```\n\u002Focto:auto research microservices patterns    -> routes to discover phase\n\u002Focto:auto build user authentication          -> routes to develop phase\n\u002Focto:auto compare Redis vs DynamoDB          -> routes to debate\n```\n\nThe smart router parses your intent and selects the right workflow.\n\nMulti-provider runs also write an agent status ledger. Use `octopus agent-summary` to see which providers contributed, which ran degraded, and which failed before synthesis.\n\n---\n\n## Pick a Command by Goal\n\nNot sure which command to use? Pick by goal:\n\n| I want to... | Use |\n|--------------|-----|\n| Research a topic thoroughly | `\u002Focto:research` or `\u002Focto:discover` |\n| Debate two approaches | `\u002Focto:debate` |\n| Build a feature end-to-end | `\u002Focto:embrace` |\n| Design a UI or style system | `\u002Focto:design` |\n| Review existing code | `\u002Focto:review` |\n| Write tests first, then code | `\u002Focto:tdd` |\n| Scan for vulnerabilities | `\u002Focto:security` |\n| Write a product spec | `\u002Focto:prd` |\n| Go from spec to shipping code | `\u002Focto:factory` |\n| Debug a tricky issue | `\u002Focto:debug` |\n| Reduce token usage | `\u002Focto:doctor` (includes RTK install + token tips) |\n| Just run something quick | `\u002Focto:quick` |\n\nOr skip the table — type `\u002Focto:auto \u003Cwhat you want>` or just say `octo \u003Cwhat you want>`, and the smart router picks for you. 🔍\n\n\u003Cdetails>\n\u003Csummary>\u003Cstrong>How does this compare to Superpowers or plain Claude Code?\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n| | Claude Code alone | [Superpowers](https:\u002F\u002Fgithub.com\u002Fobra\u002Fsuperpowers) | Claude Octopus |\n|---|---|---|---|\n| **Core idea** | One model, your prompts | Structured methodology for one agent | Up to 8 providers cross-checking each other |\n| **Providers** | Claude only | Claude only | Codex, Gemini, Copilot, Qwen, Ollama, Perplexity, OpenRouter, OpenCode |\n| **Workflow** | Ad-hoc | Spec → plan → subagent-driven dev | Discover → Define → Develop → Deliver (Double Diamond) |\n| **Strength** | Simple, no setup | Long autonomous runs with discipline | Multiple perspectives catching blind spots |\n| **Consensus gates** | No | No | Yes — 75% agreement threshold |\n| **Best for** | Quick tasks, simple features | Large builds with clear specs | Research, review, debates, multi-provider validation |\n| **Setup** | Nothing | Install plugin | Install plugin, optionally add providers |\n\n**tl;dr:** Superpowers makes one agent work really well for hours. Octopus makes multiple agents check each other's work. They solve different problems.\n\n\u003C\u002Fdetails>\n\n---\n\n## How It Works\n\n### How 8 Providers Work Together\n\nClaude Octopus coordinates up to eight AI providers — one per tentacle:\n\n| Provider | Role |\n|----------|------|\n| 🔴 Codex (OpenAI, GPT-5.4) | Code review + implementation — edge-case hunting, terminal-heavy execution, patch\u002Ftest loops |\n| 🟡 Gemini (Google) | Ecosystem breadth — alternatives, research synthesis |\n| 🟣 Perplexity | Live web search — CVE lookups, dependency research, current docs |\n| 🌐 OpenRouter | Alternative model routing — access 100+ models via single API |\n| 🟢 Copilot (GitHub) | Zero-cost research — uses existing GitHub Copilot subscription |\n| 🟤 Qwen (Alibaba) | Free-tier research — 1,000-2,000 requests\u002Fday via Qwen OAuth |\n| ⚫ Ollama (Local) | Zero-cost local LLM — offline, privacy-sensitive, fallback |\n| 🔵 Claude (Anthropic, Opus 4.7 + Sonnet 4.6) | Architecture, strategy, security review, orchestration, consensus, final synthesis |\n\nProviders run in parallel for research, sequentially for problem scoping, and adversarially for review. A 75% consensus quality gate prevents questionable work from shipping. Only Claude is required — all others are optional and auto-detected.\n\n**v9.29.0 role defaults** (April 2026 benchmark refresh): `architect`, `strategist`, and `security-reviewer` default to Claude Opus 4.7 (leads SWE-bench Pro 64.3, MCP-Atlas tool use +9.2, LMArena #1). `code-reviewer` and `implementer` default to GPT-5.4 (leads Terminal-Bench 75.1, edge-case review). Opt out with `OCTOPUS_LEGACY_ROLES=1` to restore the v9.28 mapping. See [CHANGELOG](CHANGELOG.md#9290---2026-04-22) and [GPT-5.4 prompting guide](docs\u002FGPT-5.4-PROMPTING.md).\n\n### Four Phases: Discover, Define, Develop, Deliver\n\nFour structured phases adapted from the UK Design Council's methodology:\n\n| Phase | Command | What happens |\n|-------|---------|-------------|\n| Discover | `\u002Focto:discover` | Multi-AI research and broad exploration |\n| Define | `\u002Focto:define` | Requirements clarification with consensus |\n| Develop | `\u002Focto:develop` | Implementation with quality gates |\n| Deliver | `\u002Focto:deliver` | Adversarial review and go\u002Fno-go scoring |\n\nRun phases individually or all four with `\u002Focto:embrace`. Configure autonomy: supervised (approve each phase), semi-autonomous (intervene on failures), or autonomous (run all four).\n\n### 32 Specialist Personas\n\nSpecialized agents that activate automatically based on your request. When you say \"audit my API for vulnerabilities,\" security-auditor activates. When you say \"design a dashboard,\" ui-ux-designer takes over.\n\nCategories: Software Engineering (11), Specialized Development (6), Documentation & Communication (5), Research & Strategy (3), Business & Compliance (3), Creative & Design (4).\n\n[Full persona reference](docs\u002FAGENTS.md) | [All 52 skills](docs\u002FCOMMAND-REFERENCE.md)\n\n### Built-in Reaction Engine\n\nWhen agents create PRs, the reaction engine monitors what happens next — CI failures, review comments, stale agents — and responds automatically. No new commands to learn. It fires transparently inside workflows you already use:\n\n| Integration Point | When It Fires |\n|-------------------|---------------|\n| `\u002Focto:parallel` | Between poll cycles while monitoring work packages |\n| `\u002Focto:sentinel` | After triage scan completes |\n| `agent-registry.sh health --react` | On-demand health check |\n\n**What it auto-handles:**\n\n| Event | Reaction | Limits |\n|-------|----------|--------|\n| CI failure | Collects failure logs into agent inbox | 3 retries, escalates after 30m |\n| Changes requested | Collects review comments into agent inbox | 2 retries, escalates after 60m |\n| Agent stuck | Escalates to human | After 15m with no progress |\n| PR approved + CI green | Notifies you it's ready to merge | — |\n| PR merged | Marks agent complete | — |\n\n**Override defaults per project** by creating `.octo\u002Freactions.conf`:\n\n```\n# EVENT|ACTION|MAX_RETRIES|ESCALATE_AFTER_MIN|ENABLED\nci_failed|forward_logs|5|45|true\nchanges_requested|forward_comments|3|90|true\nstuck|escalate|0|10|true\n```\n\nReactions track 13 agent lifecycle states: `running` → `pr_open` → `ci_pending` → `ci_failed` \u002F `review_pending` → `changes_requested` \u002F `approved` → `mergeable` → `merged` → `done`.\n\n---\n\n## Providers and What They Cost\n\n### Authentication\n\n| Method | Codex | Gemini | Claude |\n|--------|-------|--------|--------|\n| OAuth (recommended) | `codex login` — included in ChatGPT subscription | Google account — included in AI subscription | Built into Claude Code |\n| API key | `OPENAI_API_KEY` — per-token billing | `GEMINI_API_KEY` — per-token billing | Built into Claude Code |\n\nOAuth users pay nothing beyond their existing subscriptions.\n\n### What You Get With Just Claude\n\nEverything except multi-AI features. You get all 32 personas, structured workflows, smart routing, context detection, and every skill. Multi-AI orchestration (parallel analysis, debate, consensus) activates when external providers are configured.\n\n---\n\n## Trust, Safety, and Limits\n\n**Namespace isolation** — Only `\u002Focto:*` commands and `octo` natural language prefix activate the plugin. Your existing Claude Code setup is untouched.\n\n**Data locations** — Results in `~\u002F.claude-octopus\u002Fresults\u002F`, logs in `~\u002F.claude-octopus\u002Flogs\u002F`, project state in `.octo\u002F`. Nothing hidden.\n\n**Provider transparency** — Every command shows a 🐙 activation indicator on launch. Colored dots (🔴 🟡 🟣 🔵) show exactly which providers are running and when external APIs are called. You always know what's happening.\n\n**Clean uninstall** — Run `claude plugin uninstall octo` from your terminal. If you see a scope error, add `--scope project`. No residual config changes.\n\n---\n\n## Works With OpenClaw\n\nClaude Octopus ships with a compatibility layer for [OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw), the open-source AI assistant framework. This lets you expose Octopus workflows to messaging platforms (Telegram, Discord, Signal, WhatsApp) without modifying the Claude Code plugin.\n\n### Architecture\n\n```\nClaude Code Plugin (unchanged)\n  └── .mcp.json ─── MCP Server ─── orchestrate.sh\n                                        ↑\nOpenClaw Extension ─────────────────────┘\n```\n\nThree components, zero changes to the core plugin:\n\n| Component | Location | Purpose |\n|-----------|----------|---------|\n| MCP Server | `mcp-server\u002F` | Exposes 10 Octopus tools via Model Context Protocol |\n| OpenClaw Extension | `openclaw\u002F` | Wraps workflows for OpenClaw's extension API |\n| Skill Schema | `mcp-server\u002Fsrc\u002Fschema\u002Fskill-schema.json` | Universal skill metadata format |\n\n### MCP Server\n\nThe MCP server is **opt-in** — it does not start automatically. This prevents a permanent `✘ failed` status in Claude Code's `\u002Fmcp` panel for users who don't need it.\n\nTo enable it, add the server to your project's `.mcp.json` or global Claude Code settings:\n\n```json\n{\n  \"mcpServers\": {\n    \"octo-claw\": {\n      \"command\": \"node\",\n      \"args\": [\"--require\", \".\u002Fmcp-server\u002Fcheck-node-version.js\", \".\u002Fmcp-server\u002Fdist\u002Findex.js\"],\n      \"cwd\": \"\u003Cpath-to-claude-octopus>\",\n      \"env\": {\n        \"OCTO_CLAW_ENABLED\": \"true\"\n      }\n    }\n  }\n}\n```\n\nOnce enabled, it exposes:\n\n- `octopus_discover`, `octopus_define`, `octopus_develop`, `octopus_deliver` — Individual phases\n- `octopus_embrace` — Full Double Diamond workflow\n- `octopus_debate`, `octopus_review`, `octopus_security` — Specialized workflows\n- `octopus_list_skills`, `octopus_status` — Introspection\n\nAny MCP-compatible client can connect to the server.\n\n### OpenClaw Extension\n\nInstall in an OpenClaw instance from git:\n\n```bash\nnpm install github:nyldn\u002Fclaude-octopus#main --prefix openclaw\n```\n\nOr clone and link locally:\n\n```bash\ncd openclaw && npm install && npm run build\n```\n\nThe extension registers as an OpenClaw plugin with configurable workflows, autonomy modes, and Claude Code path resolution.\n\n### Build & Validate\n\n```bash\n.\u002Fscripts\u002Fbuild-openclaw.sh          # Regenerate skill registry from frontmatter\n.\u002Fscripts\u002Fbuild-openclaw.sh --check  # CI mode — exits non-zero if out of sync\n.\u002Ftests\u002Fvalidate-openclaw.sh         # 13-check validation suite\n```\n\n---\n\n## FAQ\n\n**Do I need all three AI providers?**\nNo. One external provider plus Claude gives you multi-AI features. No external providers still gives you personas, workflows, and skills.\n\n**Will this break my existing Claude Code setup?**\nNo. Activates only with the `octo` prefix. Results stored separately. Uninstalls cleanly.\n\n**What happens if a provider times out?**\nThe workflow continues with available providers. You'll see the status in the visual indicators.\n\n**Why \"octopus\"?**\n🐙 *Fun fact: a real octopus has three hearts, blue blood, and 500 million neurons — two-thirds of which live in its eight arms.* Each arm can taste, touch, and act independently. Claude Octopus works the same way: each tentacle (command) operates autonomously with its own squeeze of logic, then ink flows back as the final deliverable. The crossfire review? That's the squeeze — adversarial pressure that untangles everything before it ships.\n\n**How do I debug when something goes wrong?**\nRun commands with the `--verbose` flag to get detailed debugging output. Logs are stored in `~\u002F.claude-octopus\u002Flogs\u002F` for inspection. You can also use `\u002Focto:doctor` to run diagnostics and identify potential issues.\n\n---\n\n## Community\n\nJoin [r\u002FClaudeOctopus](https:\u002F\u002Fwww.reddit.com\u002Fr\u002FClaudeOctopus\u002F) for help, workflow tips, showcases, and updates.\n\n[![Star History Chart](https:\u002F\u002Fapi.star-history.com\u002Fimage?repos=nyldn\u002Fclaude-octopus&type=date&legend=top-left)](https:\u002F\u002Fwww.star-history.com\u002F?repos=nyldn%2Fclaude-octopus&type=date&legend=top-left)\n\n### Contributing\n\n1. [Report issues](https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus\u002Fissues)\n2. Submit PRs following existing code style\n3. `git clone https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus.git && make test`\n\nSee [CONTRIBUTING.md](docs\u002FCONTRIBUTING.md) for details.\n\n---\n\n## Documentation\n\n- [Documentation Guide](docs\u002FREADME.md) — Start here\n- [Command Reference](docs\u002FCOMMAND-REFERENCE.md) — Commands, triggers, and provider indicators\n- [Feature Gap Analysis](docs\u002FFEATURE-GAP.md) — CC feature adoption tracker\n- [Architecture](docs\u002FARCHITECTURE.md) — Provider flow and execution model\n- [Plugin Architecture](docs\u002FPLUGIN-ARCHITECTURE.md) — Internal plugin structure\n- [Agents & Personas](docs\u002FAGENTS.md) — All 32 personas\n- [CLI Reference](docs\u002FCLI-REFERENCE.md) — Direct CLI usage, debug mode, async, and tmux\n- [Changelog](CHANGELOG.md)\n\n---\n\n## Attribution\n\n- **[wolverin0\u002Fclaude-skills](https:\u002F\u002Fgithub.com\u002Fwolverin0\u002Fclaude-skills)** — AI Debate Hub. MIT License.\n- **[obra\u002Fsuperpowers](https:\u002F\u002Fgithub.com\u002Fobra\u002Fsuperpowers)** — Discipline skills patterns, verification-before-completion philosophy, two-stage review approach, and review response patterns. MIT License.\n- **[nextlevelbuilder\u002Fui-ux-pro-max-skill](https:\u002F\u002Fgithub.com\u002Fnextlevelbuilder\u002Fui-ux-pro-max-skill)** — BM25 design intelligence databases. MIT License.\n- **[UK Design Council](https:\u002F\u002Fwww.designcouncil.org.uk\u002Four-resources\u002Fthe-double-diamond\u002F)** — Double Diamond methodology.\n\n---\n\n## License\n\nMIT — see [LICENSE](LICENSE)\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnyldn\">nyldn\u003C\u002Fa> | MIT License | \u003Ca href=\"https:\u002F\u002Fwww.reddit.com\u002Fr\u002FClaudeOctopus\u002F\">r\u002FClaudeOctopus\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnyldn\u002Fclaude-octopus\u002Fissues\">Report Issues\u003C\u002Fa>\n\u003C\u002Fp>\n","Claude Octopus 是一个用于在软件开发和设计任务中集成多达八个AI模型的工具，旨在提前发现AI盲点，避免上线后出现问题。该项目支持Codex、Gemini、Copilot等多款主流AI模型，并通过共识机制标记不同模型间的分歧，确保决策质量。它特别适用于需要多模型协作以提高准确性和安全性的场景，如代码审查、安全性评估以及复杂的研究任务。此外，Claude Octopus还具备跨会话记忆功能，能够记住以往的决策与上下文，提供持续的支持。对于希望实现从需求定义到最终交付全流程自动化处理的研发团队来说，这是一个非常实用且高效的解决方案。",2,"2026-06-11 03:51:15","high_star"]