- Home
- Skills
- Agents & Orchestration
- agent coordinator
Works with the AI Agents you already use
agent coordinator
Deterministic multi-agent orchestration for parallel coding tasks, GPU resource gating, and isolated worktree management.
$9
agent coordinator
Example session with this skill installed
Run a model evaluation benchmark using agy on gemini-2.5-pro. Wait for at least 4GB free GPU VRAM before starting.
- Read your context and instructions
- Compiled the agent coordinator
I have dispatched the evaluation benchmark to the Antigravity worker harness. It is currently waiting for the specified GPU resources to become available in an isolated Git worktree.
Multi-Agent Dispatch Status
| Task ID | Harness | Model | Status | Worktree |
|---|---|---|---|---|
agy-7721 | Antigravity | gemini-2.5-pro | WAITING_FOR_GPU | .coordinator/worktrees/agy-7721 |
worker-1 | Claude | claude-3-7-sonnet | IDLE | N/A |
Execution Details:
- Target Branch:
agent/agy/agy-7721 - GPU Constraint:
4000MBminimum free VRAM - Current VRAM:
1240MBfree (Waiting for release...)
The worker will automatically begin the benchmark once the resource gate clears. It will modify evals/benchmark.py and produce a TASK_REPORT.json upon completion.
Next steps
- Run
coord statusin 5 minutes to check if the GPU resource gate has opened. - Use
coord logs agy-7721 --tail 20to view the benchmark output once it starts. - Review the final diff via
git diff main..agent/agy/agy-7721after the task completes.
Connects securely to your tools. The creator never sees your data.
About this skill
Stop waiting for long-running builds or deep test suites to finish. When you're managing complex features, your agent shouldn't be locked into a single process or context. This skill implements a deterministic multi-agent architecture that offloads implementation, testing, and heavy compute tasks to autonomous worker agents.
What it does
- Parallel Task Execution runs multiple independent coding tasks concurrently using isolated Git worktrees to prevent race conditions.
- Heterogeneous Orchestration coordinates between Claude Code, Google Antigravity, OpenCode, and OpenAI Codex harnesses.
- GPU Resource Gating monitors and reserves VRAM for compute-heavy tasks like model training or 3D rendering before dispatching workers.
- Zero-Token Monitoring tracks background agent progress via structured
TASK_REPORT.jsonfiles instead of ingesting massive conversation transcripts. - Automated Worktree Management handles the lifecycle of dedicated branches and isolated workspaces for every task.
How it works
- Initialize the workspace with
coord initto establish role-based rules and token-efficient communication protocols. - Dispatch a task to a specific worker agent (e.g.,
agyfor GPU/large context,opencodefor free boilerplate) with defined file scopes and success criteria. - Monitor progress using
coord statusandcoord logsto view execution state without consuming LLM context tokens. - Review the structured report and git diff once the worker finishes its implementation and internal test loop.
- Merge the validated task branch and run
coord cleanto prune metadata and isolated workspaces.
Frameworks & tools
This skill integrates with Claude Code, Antigravity (agy), OpenCode (opencode), and OpenAI Codex (codex). It requires Python 3.9+ and Git.
Why this beats prompting it yourself
Manual multi-agent prompting often leads to context leakage and working-tree collisions. This skill provides a hard-coded orchestration layer that enforces Git isolation and structured reporting, ensuring agents don't overwrite each other's work.
Use cases
- Parallel Refactoring where one agent updates the API while another scaffolds the frontend and a third writes integration tests.
- GPU-Bound Workflows offloading PyTorch or CUDA training jobs to an agent while you continue architectural planning.
- Boilerplate Offloading using zero-cost models for unit test generation and documentation to save your primary model's quota.
Known limitations
Requires Git for full worktree isolation. Workers can only modify files explicitly scoped in the dispatch command. Supports Linux and macOS environments only. Need to mention the LLM models for the agents explicitly to reduce more token usage.
Requires Git for full worktree isolation. Workers can only modify files explicitly scoped in the dispatch command. Supports Linux and macOS environments only. Need to mention the LLM models for the agents explicitly to reduce more token usage.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 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
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 7 days ago
Requires
Permissions
Allowed Hosts