More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Conductor

    1

    Parallel AI coding agents in isolated git worktrees, merged back cleanly.

    Free

    1 useSecurity scanned
    Agent Conductor

    Agent Conductor

    Example session with this skill installed

    Our shop-utils library has no tests at all. Use agent-conductor to write unit
    tests for every module in src/, in parallel, one worker per module. Use Node's
    built-in test runner and don't touch the source code.

    • Read your context and instructions
    • Compiled the agent conductor

    Summary

    4 workers wrote tests for all 4 modules in parallel. They finished in 28–39 s each,
    and all 62 new tests pass. Everything is merged into main as 4 linear commits.
    Nothing has been pushed yet.

    Setup (asked once, saved for next time): worker agent agy · test command npm test · 4 workers at once

    TaskWorkerNew testsDiffTimeStatus
    Add unit tests for price moduleagy17+8228 s✅ merged
    Add unit tests for cart moduleagy12+8229 s✅ merged
    Add unit tests for slug moduleagy15+7431 s✅ merged
    Add unit tests for delivery moduleagy18+12539 s✅ merged (reviewed: flagged LARGE)

    Each worker could touch only its own file (test/<module>.test.js), so there were
    no conflicts. I checked the one flagged diff (test/delivery.test.js, 125 lines) by hand.
    Its weekend/Friday dates are correct and it changes no source code.

    $ git log --oneline
    78c4b20 Add unit tests for delivery module
    06e867e Add unit tests for slug module
    fd5f306 Add unit tests for cart module
    05fb21f Add unit tests for price module
    6aff247 shop-utils 0.3.0
    
    $ npm test
    ℹ tests 62
    ℹ pass 62
    ℹ fail 0
    

    Next steps

    • Review the 4 commits (git log -p -4). Each one is a single test file.
    • Say "push" when you're happy. I won't push without your OK.
    • Run cleanup.sh --all to remove the worker worktrees and the tmux session (I'll do it on request).

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Some jobs are big but repetitive: tests for every module, a page for every product, docs for every package. A single AI assistant does them one after another, and you wait.

    Agent Conductor lets your AI assistant act as a team lead. It breaks the job into small, independent pieces and hands each piece to its own AI worker. All the workers run at the same time. Each one works in its own separate copy of your project, so they never get in each other's way or overwrite each other's work.

    When the workers finish, the lead checks a short report from each one, runs the tests, and brings only the finished, passing work back into your project, one clear change per task. Anything unusually large is flagged for you to look at first. Nothing is ever published or pushed without your OK.

    You stay in control of cost: you choose which AI does the work (Google Antigravity, Claude Code, Codex, or another tool), and how many workers run at once.

    • Tasks must be independent. Two workers editing the same lines cause a merge conflict; the merge stops and asks you.
    • Workers use auto-approve mode inside their worktree. They are isolated by git, not sandboxed by the OS.
    • Each worker is a full agent session and counts against your plan or API quota.
    • Antigravity CLI (agy) needs one interactive sign-in first; headless mode has no API-key login yet.
    • The login check confirms you are signed in, not that your plan has quota or billing left.
    • Fresh worktrees have no installed dependencies. Tests usually need a setup command (e.g. npm ci).
    • Requires git and tmux. Tested on macOS; Linux untested; Windows only via WSL.
    • Built-in presets: agy, claude, codex. Other CLIs work through a custom command.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    Anthropic
    OpenAI
    Antigravity
    macOS
    Git

    Permissions

    Terminal / Shell
    Write Files
    Read Files
    Environment Variables
    Network Access

    Allowed Hosts

    antigravity.google
    anthropic.com
    openai.com

    File Scopes

    .conductor/**
    .git/info/exclude
    **
    (workers
    only
    inside
    their
    own
    worktree
    in
    .conductor/worktrees/<id>/)

    Workers run unattended in each agent's auto-approve mode (agy/claude --dangerously-skip-permissions, codex --dangerously-bypass-approvals-and-sandbox), because nobody is there to click "allow". Each worker is confined to its own git worktree and branch; nothing reaches your branch until the manager merges it, and nothing is pushed without your explicit OK. If your policy forbids auto-approve modes, plug in your own command (AGENT=custom) with stricter flags. Network access is used only by the worker agents to reach their model provider.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    The one you choose. On first use the skill checks which agents are installed and logged in (Antigravity, Claude Code, Codex) and asks you to pick one, plus a model, a test command and how many workers to run. It remembers your answers per project.

    No. Each worker gets its own git worktree and branch, so they never share files. Nothing lands on your branch until the manager merges it, and only work whose tests passed gets merged.

    Never. Merging happens locally, and pushing always needs your explicit OK.

    The skill is just scripts. You pay only for the agent sessions the workers use. You set the number of parallel workers, and every agent run has a time limit (20 minutes by default).

    The merge stops at the first conflict and the manager asks you how to proceed. The skill's splitting guide makes the manager give every file to a single task, so this is rare.

    No. They run in a tmux session in the background. If you're curious, `tmux attach -t conductor-<repo>` shows every worker live.

    Many small, independent pieces of work: tests per module, pages per product, docs per package, fixtures, migrations across separate folders. One tightly coupled change is better done by a single agent.

    No. Workers are marked with an environment flag and the skill refuses to run inside them, so you can't get runaway recursion.

    New in Agents & Orchestration