Squadron

Manage Claude Code, Codex, and Grok sessions in your IDE. Let agents delegate tasks, review each other’s work, and collaborate across models.

Works in VS Code and editors built on it. macOS is supported; Windows support is available with some features still in progress. See Platforms.

Requirements

• One of these, installed and signed in:
  • Claude Code: curl -fsSL https://claude.ai/install.sh | bash
  • Codex: npm i -g @openai/codex
  • Grok: curl -fsSL https://x.ai/cli/install.sh | bash
• VS Code ≥ 1.95 (the extension also works in Antigravity via Open VSX)
• macOS (Apple Silicon or Intel), or Windows x64 with .NET Framework 4.8.1 or later. See Platforms for current limitations.

What you can do

• Run agents in your IDE with your Claude Code, Codex or Grok subscription.
• Follow agent cards with live status and sound cues when agents need you or finish.
• Attach files and images in chat, and browse past attachments and agent-created artifacts.
• Answer questions and permission prompts in a unified modal UI across models.
• Build multi-step workflows with agents across models, and sequence workflows through step instructions.
• Answer your sessions from your phone through Slack, with Remote.
• Keep an idle session's prompt cache warm, so coming back to it doesn't resend the whole conversation, with Warm (experimental).
• Keep your Mac awake while sessions work, Remote is on, or a session has Warm on. Keep-awake on the bar at the bottom of the board shows what is holding it and for how long.
• Sign in to each CLI from Settings (the gear), see each account's plan usage beside it, and pick the model and effort new sessions start on.
• Turn on the mascot, a small pixel chick that sits on the bar and flies to the card where something happens. Experimental and off by default: Settings → General → Mascot.
• And more…

Each window shows only the sessions it started. Sessions started in a terminal are not shown.

Agents that work together

A session can start other sessions, send them messages and wait for their answers, even when they run a different CLI. Ask Claude to have Codex build a feature and Grok review it: Claude starts both, hands each its task, and reads what they send back.

To let agents start and message other sessions, turn on Orchestration in Agents on the bottom hub (squadron.orchestrators.enabled). It's off by default, and while it's off every session command that acts (message, start, end, rename, warm, wait, step, Remote files) refuses. Listing sessions and models still works.

Enabling it also installs three skills for your CLIs: squadron for session commands, squadron-workflow-builder for creating reusable workflows, and squadron-heavy-runs, which has sessions take turns on test suites and builds instead of running them all at once.

The extension supplies you with 2 example workflows to play with, modify it per your work preferences as you go.

Browse your Linear backlog

Click Tickets to browse your Linear backlog by status while keeping your conversation open. Filter by assignee, status or priority, and click a ticket identifier to open it in Linear.

Remote: answer sessions from your phone

Remote connects this machine's sessions to Slack. When a session asks a question or wants a permission, finishes, fails or stops early, you're mentioned in that session's own Slack thread. Reply in the thread to answer it, or press Deny or End. /squadron new starts a job in a folder, model and effort you pick, and /squadron list shows what's running. @mention the bot in a channel thread to start a job there, with the thread's earlier messages handed to the agent.

Click Remote on the bar at the bottom of the board to set it up. The guide walks you through making the Slack app from its manifest. Paste the app's two tokens, add your Slack member id under Allowed accounts, press Check, then turn on Enable Remote.

Only the accounts you allow are heard, and a permission can be denied from your phone but never allowed. Remote runs in one window at a time, comes back after a reload, and keeps your Mac awake while it's on, unless Keep-awake or its Remote hold is off, or the battery is below 20% while unplugged. Windows has no keep-awake support yet, so Remote cannot prevent that machine from sleeping.

Warm: keep a session's cache warm

Experimental. Warm's default cache lifetimes are an hour for Claude, 35 minutes for Codex and 10 for Grok; actual cache retention can vary. Warm sends an idle session one short ping just before that runs out, for up to two hours after its last real turn, so picking the session back up doesn't pay to resend the whole context. A warmed card has a copper edge, and an ember runs along its footer while a ping is out.

Click Warm on the bar to switch it per session, and to set each CLI's cache lifetime, how early to ping and for how long. Turn on squadron.warmCache.enabled to have conversations without a saved Warm choice start warmed. A session's choice survives resumes. Pings spend tokens, and the cache isn't guaranteed to be reused. On macOS, Warm also keeps the machine awake unless Keep-awake or its Warm hold is off, or the battery is below 20% while unplugged. On Windows, pings require the machine to remain awake.

Getting started

1. Install Squadron from the VS Code Marketplace, or from Open VSX if you use Antigravity, Cursor, VSCodium or another VS Code fork.
2. Sign in to Claude Code, Codex, or Grok. You need at least one. Squadron uses the CLI you already have and never installs one for you.
3. Run Squadron: Show Sessions Overview from the Command Palette, or click the Squadron icon in the editor's title bar.
4. If Squadron can't find your CLI, click Log in at the top of the panel and use Set path. This happens a lot when the editor was opened from the Dock, because it doesn't pick up your shell's PATH.

Settings

Everything is under squadron.* in Settings. The ones worth knowing:

Platforms

• macOS: supported on Apple Silicon and Intel, including keep-awake.
• Windows: session support is available, including Squadron commands, Remote file transfers, MCP sign-in and Warm cache pings. Feature parity and verification are still in progress. Helper features require Windows x64 and .NET Framework 4.8.1 or later. Keep-awake is not available, including sleep protection for Remote and Warm.
• Linux: untested; keep-awake is not available.

Roadmap

Ideas we're working toward, in no particular order:

• Workflow Editor UI Refine
• Workflow v3: dynamic workflows (model-agnostic), continuous/routine workflows.
• Gemini support: when I have money to get access to Gemini 4...
• Implementation of Decision layers: Jev or other decision models, for workflow routing or automate choices, idk, still early in ideation
• Ticket writes and inapp ticketing: Create and update tickets from the extension. Potentially add a local ticket system for people who don't use Linear, Jira, or similar tools.
• Board UI rework: TBD. Better grouping UI for agent sessions on the board.