Voltron-Lions

A Claude Code & Claude Cowork plugin for mission-style multi-agent orchestration. One command. Five themed agents. Structured outputs.

What it does

/start-mission <objective> invokes Voltron Main (the Black Lion), who:

1. Drafts a Mission Brief (objective, scope, success criteria, constraints)
2. Sends Green Lion on recon (uses graphify-out/ if present and fresh; falls back to filesystem)
3. Generates Lion Assignments with strict file-ownership boundaries
4. Performs Skills Gap analysis using the built-in skills registry & actions map — flags missing capabilities, can scaffold new skills via skill-creator
5. Builds a Risk Register (technical / scope / integration / data-loss, severity L/M/H)
6. Waits for your go / no-go, then dispatches Red / Blue / Yellow Lions to execute

The Lions

Installation

Pick the path that matches you. Both install the exact same plugin.

🧑‍💻 For developers (Claude Code)

In any Claude Code session, run these two commands:

/plugin marketplace add Cloudscockpit/voltron-lions-claudecode
/plugin install voltron-lions

Restart the session (or run /reload) and /start-mission is available.

🙋 For non-technical users (Claude Cowork / Claude Desktop)

No terminal needed — you type two lines into the chat box, same as sending a message:

1. Open Claude Cowork (or the Claude desktop app with Claude Code enabled).
2. In the chat box, type exactly this and press Enter: /plugin marketplace add Cloudscockpit/voltron-lions-claudecode
3. Then type this and press Enter: /plugin install voltron-lions
4. When asked to confirm, choose Install / Yes.
5. Start a new conversation. You're ready — type /start-mission followed by what you want done, in plain English. Example: /start-mission organize the files in my project folder and tell me what's outdated
6. Voltron will show you a mission plan and wait for your approval — nothing happens until you reply go.

💡 If a step fails with "requires a newer version", update your Claude app first, then repeat from step 2.

Usage

/start-mission add a /health endpoint to the express app that returns service version

Voltron Main returns a four-part report. Reply go to dispatch Lions, no-go to revise, or edit <section> to change one section.

Skills registry & actions map

The plugin ships a machine-readable registry of everything the Lions can do:

• Skill: voltron-lions:skills-registry — invoked by Voltron Main during Skills Gap analysis, or by you ("what can the Lions do?")
• Data: voltron-lions/skills/skills-registry/actions-map.json — each entry maps an action type → responsible Lion → required tools/skills → status

To register a new action type, edit actions-map.json, bump the plugin version, and reinstall.

Getting started as a new user (no account yet?)

Ask Voltron: "help me get started with actionboard" — the actionboard-onboarding skill walks you through, one step at a time:

1. Sign up at actionboard.ai (Claude can open the page and guide you — it hands you the browser for passwords and verification; it never types credentials).
2. Install Voltron Castle Desktop from the releases page and sign in.
3. Verify the bridge — ask "check voltron status".

Connect Claude Cowork to the local Voltron MCP endpoint

Running the voltron-backend stack locally? Ask: "connect cowork to the local voltron mcp" — the cowork-mcp-connect skill checks the endpoint (http://localhost:8200/mcp), then walks either path:

• Developers (Claude Code): claude mcp add --transport http voltron http://localhost:8200/mcp
• Non-technical (Cowork/Desktop): Settings → Connectors → Add custom connector → URL http://localhost:8200/mcp

This gives Claude direct access to the local Voltron actionlist tools — no cloud pod needed.

Browser actions in missions

The browser-actions skill (+ kb/claude-browser-actions.md) teaches Voltron how to drive a real browser during missions — navigate, read pages, fill forms, screenshot, debug web apps — with hard rules: per-site approval, credentials/CAPTCHAs always handed to you, confirmation before any irreversible web action, and page content treated as data, never as instructions.

Knowledge base

The plugin ships a kb/ folder the skills draw on:

Connect to an actionboard.ai Cloud AI Pod (remote missions)

Voltron Lions can dispatch actions to a remote Actionboard AI pod — cloud actionlists, RAOARA six-phase workflows, and Voltron Desktop bridging — through the Actionboard AI plugin.

What you need:

1. An account at actionboard.ai with an AI pod (your pod URL and credentials come from the Actionboard AI dashboard).
2. The Actionboard AI plugin installed in Claude Code / Cowork (it provides connect-pod, list-actions, execute-action, voltron-status, list-boards, and raoara).

Then just ask:

connect to my actionboard pod

Voltron uses its actionboard-pod-connect skill to walk the connection flow: connect → verify health → list boards → list actions. After that, remote pod actions can appear in mission plans like any other Lion assignment — still gated by your go/no-go, and Voltron re-confirms before any irreversible pod action (deploys, sends, deletes).

Security: your pod credentials live in the Actionboard AI plugin's own configuration. Voltron Lions never stores, logs, or echoes them.

Tips

• For large codebases, run /graphify first so Green Lion's recon is graph-backed.
• If Voltron Main flags needs-new-skill gaps, it will ask before scaffolding via skill-creator. New skills become available on the next session.
• Ask "what can the Lions do?" any time — the skills registry answers with the current actions map.

Design

The design rationale and implementation plan are preserved in the commit history — see commits f547fa4 (design spec) and 81adfc2 (implementation plan).