Commander

CC Commander — AI PM for Claude Code · 71 Skills · 22 Agents · Free for Now

> Ship faster with Claude Code and Cowork Desktop. One plugin install activates 71 skills, 22 specialist AI agents, and 23 lifecycle hooks across Claude Code Desktop, Claude Code CLI, Cursor, Windsurf, and every MCP-capable IDE.

Install in 30 seconds:

Settings → Plugin Marketplace → Add from GitHub: KevinZai/commander → Install

What you get: 71 click-first /ccc-* plugin skills · 22 specialist sub-agents (architect, reviewer, debugger, designer, and more) · 23 lifecycle hooks · 2 bundled MCP servers · 502+ ecosystem skills · Free for now.

> 🖥️ Primary surface: Claude Code Desktop (aka Cowork Desktop). Install once via Settings → Plugin Marketplace. All 71 plugin skills, 22 agents, 23 lifecycle hooks (39 handlers), and 2 credential-free bundled MCP servers (+16 opt-in via /ccc-connect) appear inside every session automatically — no terminal needed. > > Cowork Desktop and Claude Code Desktop are the same app, two UI modes. The plugin works identically in both. All screenshots in docs/screenshots/ were taken in Desktop.

[](https://github.com/KevinZai/commander/stargazers) [](https://opensource.org/licenses/MIT) [](https://twitter.com/kzic) [](./SKILLS-INDEX.md) [](./commander/cowork-plugin/README.md) [](#️-who-its-for) [](./CHANGELOG.md)

Powered by:

· affiliate disclosure

Kevin Zicherman · @kzic · Built from 200+ community sources · Aggregates 19 vendor packages

[Why CC Commander](#-why-cc-commander) · [The 70 skills](#the-70-plugin-skills) · [Browse Skills](SKILLS-INDEX.md) · [Agent Bible](docs/BIBLE-AGENT.md) · [Ecosystem](docs/ECOSYSTEM.md) · [BIBLE](BIBLE.md) · [Changelog](CHANGELOG.md)

🎯 Who it's for

| If you are... | CC Commander is... | |--------------|-------------------| | 👋 New to AI coding agents → using Claude Cowork Desktop | Your onboarding buddy — /ccc pops up a native chip picker: Build · Review · Ship · Design · Learn · More. Click. No typing. 28 guided workflows, zero config. | | 💻 A developer → using Claude Code Desktop or Claude Code CLI | Your project manager — routes complexity, compounds knowledge across sessions, Kevin Z Method built in. /ccc-plan → spec interview → plan file. | | 🔧 In Cursor / Windsurf / Cline / Continue / Codex | One hosted MCP endpoint unlocks all 459+ skills in your IDE of choice | | 📱 On Claude mobile or iPad | Same skills, same license, synced across devices (hosted MCP) |

---

⭐ 30-second install

> For Cowork Desktop or Claude Code Desktop users (99% of people). No terminal needed.

1. Open Settings → Plugin Marketplace
2. Click Add from GitHub → enter KevinZai/commander → Add
3. Find commander in the marketplace → click Install
4. Cmd+Q, reopen the app, type /ccc — you're live

✅ 71 plugin skills activate. ✅ Zero config. ✅ Zero API keys.

Power-user CLI install:

/plugin marketplace add KevinZai/commander
/plugin install commander

Or full dev environment + ccc binary:

curl -fsSL https://raw.githubusercontent.com/KevinZai/commander/main/install-remote.sh | bash

---

🌐 One plugin, every agent

> Commander is a PM layer for whichever AI coding tool you use. Same 71 skills, same 22 specialist agents, same lifecycle hooks — different install path per platform. The package name stays commander everywhere; the surrounding UI (Claude Code, Codex, Cursor, etc.) provides the platform context.

| Platform | Status | Install | |---------|--------|---------| | Claude Code Desktop / Cowork Desktop ⭐ | Shipping (primary) | Settings → Plugin Marketplace → Add from GitHub: KevinZai/commander | | Claude Code CLI | Shipping | /plugin marketplace add KevinZai/commander then /plugin install commander | | Cursor / Windsurf / Cline / Continue / Codex | Shipping (hosted MCP) | Point MCP client at mcp.commanderplugin.com | | Codex / Gemini native plugin runtimes | Roadmap | Same repo, platform-specific adapter (planned v4.2+) |

One brand: Commander. One tagline: The guided AI PM for every AI coding agent. One repo: KevinZai/commander. Everywhere.

---

🎯 Why CC Commander

Stock Claude Code is a blank terminal with amnesia. Every other plugin solves one slice. CC Commander is the first guided AI PM — click-first UX, an intelligence layer that thinks before you do, and the whole ecosystem pre-wired. One install, everything active.

🥇 What no competitor has

| Moat | CC Commander | Cursor / Windsurf / Copilot | OSS (Cline / Continue / Aider) | |------|--------------|----------------------------|-------------------------------| | Lifecycle hooks (23 × 39 handlers) | ✅ SessionStart, PreToolUse, PostToolUse, Stop, PreCompact, Notification, etc. | ❌ None | ❌ None | | 22 specialist agent personas | ✅ Architect, Security-Auditor, TypeScript-Reviewer, Designer, Debugger, etc. | ❌ None ship pre-built | ❌ None ship pre-built | | 18 MCP servers (2 bundled + 16 opt-in) | ✅ Zero-config — Notion, Slack, Supabase, Figma, Linear, GitHub, Vercel, Fly… | ⚠️ Manual MCP setup required | ⚠️ Manual MCP setup required | | Spec-first planning (/ccc-plan → interview → spec → multi-agent execution) | ✅ Native | ❌ No planning primitive | ❌ No planning primitive | | Cross-platform (CLI + Desktop + VS Code path + hosted MCP) | ✅ Same brain everywhere | ❌ IDE-locked | ⚠️ IDE-locked | | Pricing transparency | ✅ Free for now — no signup, no card, no gates | ⚠️ Cursor $20+ confusing tiers; Windsurf credit confusion | ✅ BYOK free |

The marketing angle: Set it once, the AI works smarter every session. Lifecycle hooks fire automatically. Specialist agents already know your codebase conventions. The MCP ecosystem is pre-wired. No other tool ships any of this out of the box.

• 🖱️ Click-first UX — every menu is a native AskUserQuestion chip picker. No typing. No numbered menus. No ASCII prompts. Works identically in Cowork Desktop, Claude Code Desktop, and the CLI.
• 🧠 /ccc-suggest intelligence layer — Opus-class real-time recommendation. Scans your project state, recommends one starred next step with reasoning + named 3rd-party plugins. Kills info-paralysis.
• 🧩 71 plugin skills including 13 /ccc-* specialist workflows — plain slash commands (no commander: prefix), skill-based architecture: ccc, ccc-start, ccc-browse, ccc-plan, ccc-build, ccc-review, ccc-ship, ccc-design, ccc-learn, ccc-xray, ccc-linear, ccc-fleet, ccc-connect + /ccc-e2e, /ccc-save-session, /ccc-resume-session, /ccc-changelog, /ccc-doctor, /ccc-tuneup, /ccc-upgrade, /ccc-yolo-setup, /ccc-ecc and 11 domain routers.
• 🧭 Orchestrator / Executor model — /ccc-orchestrate has Fable/Opus write a Skill.md-style goal file, GPT-5.5 via codex or a Sonnet subagent execute it, and the orchestrator verify acceptance criteria. Pair it with /ccc-handoff for frequent fresh-chat resets and /ccc-adopt to bring the doctrine into any existing project. Pay for Fable on the thinking, not the typing.
• 🎭 22 specialist agents with persona voices — architect, security-auditor, performance-engineer, content-strategist, data-analyst, designer, product-manager, technical-writer, devops-engineer, qa-engineer, reviewer, builder, researcher, debugger, fleet-worker, typescript-reviewer, python-reviewer, go-reviewer, rust-reviewer, java-reviewer, kotlin-reviewer, csharp-reviewer. Each with a distinct voice layer in commander/cowork-plugin/rules/personas/.
• 🔌 2 credential-free bundled MCP servers — context7 (library docs) + sequential-thinking. Opt-in via /ccc-connect: 16 more including Tavily, GitHub, Supabase, Figma, Playwright, Slack, Notion, claude-mem, Exa, Firecrawl, Zapier, Google Drive, Vercel, Neon, Fly.io, Upstash.
• 🪝 23 lifecycle hooks × 39 handlers — SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, Stop, Notification, PreCompact, SubagentStop, PermissionRequest. Schema-compliant, 100% pass on claude plugin validate.
• 🗺️ /ccc-cheatsheet live Mermaid map — reads the filesystem as single source of truth, renders the whole plugin as a flow diagram. Never drifts.
• 🔄 Weekly vendor auto-sync — 19 vendor submodules auto-update via GitHub Actions. The ecosystem stays current without you touching a thing.
• 🌐 Cross-client, one license — Cowork Desktop, Code Desktop, Code CLI, Cursor, Windsurf, Cline, Continue, Codex, mobile (hosted MCP).
• 🚁 Fleet orchestration — /ccc-fleet runs multiple Sonnet agents in parallel git worktrees. Fan-out, pipeline, FOR/AGAINST, background modes.
• 🆓 Free for now — no credit card, no signup. No feature gating, no paywalls.
• 🧬 Shared brain: plugin + CLI — same intelligence layer, same skill catalog, same personas. Install either, get both.
• 📖 The Kevin Z Method — BIBLE.md ships with the plugin. 7 rules, 200+ sources, 14 months of production methodology.
• 📄 Printable cheat sheets — [docs/cheatsheets/](docs/cheatsheets/) ships 3 print-ready PDFs: [CC Commander Essentials](docs/cheatsheets/ccc-essentials.pdf), [The Orchestrator/Executor Method](docs/cheatsheets/orchestrator-executor.pdf), and [Agents & Model Tiers](docs/cheatsheets/agents-model-tiers.pdf).
• 📚 459+ skills across 11 CCC domains — design, marketing, SaaS, DevOps, SEO, testing, security, data, research, mobile, makeover. Plus the vendor ecosystem.
• 🎯 Plugins-name-plugins — /ccc-suggest calls out specific 3rd-party plugins by name (claude-mem, superpowers, caveman, impeccable, graphify, and more) at the right moment. You learn the whole ecosystem through CC Commander.

🧠 The Fable Method

The operating doctrine distilled from Anthropic's frontier Fable 5 model, encoded as 12 enforceable gates — so ANY model running CC Commander produces Fable-shaped results, not just the flagship one. Arm it with /ccc-fable on, self-check with /ccc-fable audit. Full doctrine: [commander/cowork-plugin/rules/fable-method.md](commander/cowork-plugin/rules/fable-method.md).

• Never trust a single pass — the agent that does the work never grades it. Every finding gets adversarial verification from a fresh context.
• Loops need verifiers, state, and stops — repetition without a real verifier, a state file, and a deterministic stop condition is just an agent agreeing with itself on repeat.
• Prove it before you alarm — a scary number is a hypothesis until you run the exact check, open real samples, and hunt the disconfirming case.
• Context is disposable, state is durable — the durable record is handoff docs, memory, and the tracker, never the live chat.

The method IS the moat. A mid-tier model following all 12 gates reliably out-delivers an ungated frontier model on any task longer than one turn.

💡 How the intelligence works (the beginner headline)

New to Claude Code? Start here: /ccc-suggest.

Every time you run it, CC Commander reads your project state — detected stack, current branch, last 5 commits, open todos, recent errors — and returns one starred recommendation with plain-English reasoning. If another plugin would serve you better right now (e.g. claude-mem for persistent memory, superpowers for structured thinking, caveman for token-heavy iteration), it names it. Three reasoning tiers fire in order:

1. Strong signals — an open PR, a failing test, a dirty worktree, a mid-plan session
2. Stack signals — nextjs + Stripe detected → suggest ccc-saas; mobile repo → suggest ccc-mobile
3. User intent — pattern-match your recent prompts against the 502-skill catalog

One click → one next step. No info paralysis.

---

Built on Claude Agent SDK primitives

CC Commander is built on Anthropic's 2026 Claude Agent SDK sub-agent architecture — the same primitives that power agentic workflows across Claude Code Desktop, Cowork Desktop, and the CLI. The brain/hands pattern separates orchestration (the PM layer: planning, routing, decision-making) from execution (22 specialist sub-agent personas that each embody a distinct role, model, and voice). 23 lifecycle hook events with 39 handlers fire automatically throughout every session — from SessionStart initialization through SubagentStop result aggregation — so the right agent is always running at the right moment without you lifting a finger.

---

Using Cursor, Windsurf, Cline, Continue, or Codex?

[Hosted MCP setup](./mintlify-docs/features/browse-modes.mdx) — one URL + license key unlocks all 459+ skills in your editor.

---

🌟 What's new in v5.1.0

• 🧠 Opus 4.8 session default — claude-opus-4-8, $5/$25 per MTok, Fast mode 2.5× at $10/$50. New ultra/xhigh effort levels. thinking: {type: "adaptive"} replaces deprecated budget_tokens.
• ⚡ Dynamic workflows (research preview, requires Claude Code v2.1.154+) — 4 bundled JS workflow scripts spawn parallel subagents that cross-check findings: ccc-audit (backs /ccc-xray), ccc-deep-review (backs /ccc-review), ccc-migrate, ccc-fleet. Skills fall back to Agent() on older clients.
• 🚀 /ccc-ultracode — new skill for xhigh effort + automatic workflow orchestration. Or add workflow: to any prompt for a one-off run.
• 🪝 Hooks 9 → 23 events — 14 new lifecycle events: SessionEnd, PostCompact, SubagentStart, TaskCreated, TaskCompleted, and 9 more. Handlers: 24 → 38. Correctness fix: session-save moved Stop → SessionEnd.
• 🎯 71 plugin skills — recent channel/CI/ECC/setup skills (/ccc-brainstorm, /ccc-qa, /ccc-loop, /ccc-hermes, /ccc-nightwatch, /ccc-ci, /ccc-yolo-setup, /ccc-ecc) + 5 language reviewer agents (Go, Rust, Java, Kotlin, C#) + /ccc-ultracode.
• 🔐 4 security mediums resolved — telemetry scrubber hardened, error messages sanitized.
• 🤖 22 specialist sub-agent personas — 5 new language reviewers added (Go · Rust · Java · Kotlin · C#).
• 📦 15 vendor pins refreshed — ECC v1.7.0, oh-my-claudecode v4.14.4, repomix v1.14.1, and 12 more.
• 🧪 284 tests passing — +5 security regression tests added.
• 🖥️ Desktop-first positioning throughout — all docs now lead with "Claude Code Desktop is the primary surface."

🚀 What's new in v6.0.0

• 🧠 Selective Fable 5 deep-mode — Opus 4.8 stays session default (everyday), Fable 5 escalation-tier via /model claude-fable-5 with once-per-day deep-reasoning nudge. Motto: pay for Fable on the thinking, not the typing.
• 🎛️ Smart model routing — selectModelForComplexity(score) auto-assigns: 0-29 Haiku, 30-65 Sonnet, 66-85 Opus, 86-100 Fable. Dispatch tiers: power=Fable/Opus, assisted=Opus/Sonnet, guided=Sonnet/Haiku.
• 💰 Savings counter + ccc --savings — every dispatch logs estimated cost vs baseline. Footer shows daily estimate (sv$X.XX); report command prints full breakdown to ~/.claude/commander/savings.json.
• 🎭 4 deep personas on Fable — architect, debugger, security-auditor, product-manager run claude-fable-5 for complex reasoning.

⚡ Dynamic Workflows + Ultracode

> Research preview — requires Claude Code v2.1.154+. Skills fall back to standard Agent() dispatch on older clients.

Dynamic workflows are JS scripts the runtime executes in the background, spawning many subagents that cross-check findings and merge results before returning. Four bundled workflows ship in commander/cowork-plugin/workflows/:

| Workflow | Backs | What it does | |----------|-------|-------------| | ccc-audit | /ccc-xray | Repo-wide audit — security, performance, architecture, test coverage in parallel | | ccc-deep-review | /ccc-review | 4-dimension branch review (correctness · security · perf · maintainability) + reconciler | | ccc-migrate

…

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: KevinZai
• Source: KevinZai/commander
• License: MIT
• Homepage: https://commanderplugin.com

Install and usage instructions live in the source repository linked above.