Help us improve
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
Share bugs, ideas, or general feedback.
By runtorque
Manage Torque agents and tasks directly from Claude: monitor agent status, view task boards, create and complete tasks, rename agents, and integrate with git checkpoint commits and diffs to streamline work-in-progress and task delegation.
npx claudepluginhub runtorque/torqueShare bugs, ideas, or general feedback.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge.
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Display Claude agent status in tmux window names via workmux
Lanes skills and slash commands for Claude Code
Context hub for agentic workflows - manage projects, tasks, sessions, and multi-agent coordination
Subagents and hooks for TTAL multi-agent orchestration
Marketplace for the Agency CLI - an AI agent orchestrator
Multi-agent orchestration for Claude Code. 12 specialized agents working in parallel — planning, building, reviewing, debugging. Plus a Hub for always-alive multi-project sessions controllable from Telegram or Slack.
Torque is a local agent-orchestration workspace for terminal-native developers. Manage AI coding agents, run them in isolated git worktrees, dispatch work from a kanban board, and let an embedded engineer coordinate the wave — from a native desktop window, your browser, or your iTerm2 Toolbelt.

Coding agents are powerful but messy in practice. Each one wants its own session, its own branch, its own context window, and its own terminal history. Spinning up three at once can quickly turn into three terminals, three worktrees, three half-remembered prompts, and a lot of "where was I?" friction.
Torque puts a thin orchestration layer in front of that workflow. You define groups, drop agents and terminals into them, dispatch work through reusable actions, and watch tasks move across a built-in kanban board. Every agent runs in its own isolated git worktree by default, so branch boundaries are enforced instead of merely hoped for.
One step further: each group can have an engineer. The engineer is Torque's orchestrator agent — it reads the board, dispatches workers, watches digests, merges finished branches, and coordinates the next wave. It is the same idea as a designated build engineer on a small team, but for a single-user OSS workspace.
What you get:
torque CLI for scripting from the command line.Two pieces of plumbing make multi-agent orchestration safe:
Scoped MCP tools via env-var injection. When Torque spawns an agent it
writes a per-agent .mcp.json (or .codex/config.toml) with the
TORQUE_CELL_ID env var baked in. Every MCP request the agent makes carries
that cell id as an X-Torque-Cell-Id header, and the daemon uses it to
filter the tool list before it leaves the server. A worker only sees the
torque_* reporting tools. An engineer additionally sees engineer_* tools
scoped to its own group — it physically cannot enumerate another group's
journal. An architect gets architect_* tools further scoped per actor.
There is no override flag; scope is the contract. → MCP scoping
Hooks for live work tracking. For Claude Code and Codex workers, Torque
installs SessionStart, PreToolUse, PostToolUse, and Stop hooks that
POST to a local /events endpoint. Every tool call, every progress report,
every session boundary streams back into the daemon in real time. That feed
drives the activity badges on agent cells, the engineer's event digest, the
worklog, and the auto-checkpoint triggers on the worktree. You see what each
agent is doing without attaching to its terminal.
The desktop app gives you the full Torque workspace in a dedicated native window — no Toolbelt sidebar, no browser tab — and it is the easiest way to get started.
git clone git@github.com:runtorque/torque.git
cd torque
make deps
make deploy
make desktop-deps
make run
make deploy installs the primary standalone/desktop app files under
~/.torque/app and refreshes the CLI symlink. make run starts a native
desktop window on its own profile and port (defaults: desktop profile, port
18933), so it does not collide with any Toolbelt instance you might also
run.
From the cloned repo, after make deploy has been run once:
make standalone
make open
Standalone mode launches the daemon and opens Torque in your default browser. It is useful when you want a wider workspace or are running on a remote / shared machine.
For macOS users who live in iTerm2, Torque can also embed directly in the Toolbelt sidebar so it sits next to your terminal sessions.
make deps
make deploy-toolbelt
make cli
Then in iTerm2: