Misatay - AI Agent That Keeps You in the Loop

TL;DR

TL;DR: Misatay is a VS Code extension for GitHub Copilot users that keeps you involved in every AI coding session — plan tasks, review AI-generated commits, and stay in charge without micromanaging.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: marketplace.visualstudio.com/items?itemName=dshearer.misatay
• Source repository: github.com/dshearer/misatay — README read in full before writing
• License: MIT (verified via GitHub API license.spdx_id)
• HN launch thread: news.ycombinator.com/item?id=46928648

What Is Misatay?

Misatay is an experimental VS Code extension that wraps GitHub Copilot’s AI agent mode with a structured human-in-the-loop workflow. Instead of letting the AI run loose across your codebase, Misatay enforces a loop: plan a feature, let the AI work on tasks, review the diffs, then merge.

The core idea comes from the developer’s manifesto:

To make the best use of AI, AIs and people need to work side-by-side.

Misatay commits AI changes to Git as it goes, so every task has a traceable commit history. When the AI hits a blocker, it asks for help instead of spinning — the goal is to reduce token waste and keep the engineer driving.

Setup Workflow

Step 1: Install the Extension

Open VS Code and install from the Visual Studio Marketplace:

https://marketplace.visualstudio.com/items?itemName=dshearer.misatay

Or search “Misatay” in the VS Code Extensions view.

Step 2: Configure Copilot auto-accept delay

Misatay recommends setting copilot.agent.autoAcceptDelay to a value greater than 0 (for example, 5-10 seconds). This gives you time to review AI changes before they are accepted automatically.

To configure:

1. Open VS Code Settings (Cmd+, or Ctrl+,)
2. Search for copilot.agent.autoAcceptDelay
3. Set the value to your preferred delay in seconds

Step 3: Initialize Misatay in a Repo

For repos where you have not used Misatay before, open the command palette (Cmd+Shift+P) and run:

Misatay: Install Agent

This adds an agent prompt and some skills to your repository. Work is underway to move these files into the extension itself to avoid repo pollution.

Step 4: Plan and Execute

The recommended workflow is:

1. Plan — Open Misatay and describe the feature you want to build. Save the plan to your repo so it persists across AI sessions.
2. Work — Let Misatay tackle tasks from the plan. It commits changes to a feature branch as it goes.
3. Review — Misatay walks you through what changed and why, in the context of the task.
4. Ship — When you are satisfied, merge the feature branch.

Deeper Analysis

What makes Misatay different from plain Copilot Chat?

Copilot Chat answers questions and suggests code, but the AI operates without awareness of your task breakdown or commit history. Misatay introduces a task-driven layer: the AI knows what it is working toward, logs its progress to Git, and surfaces blockers to you rather than silently struggling.

The token-saving argument

The author makes a practical case: when an AI asks for help instead of looping endlessly on a problem, you spend fewer tokens. Misatay is designed to surface those moments of confusion rather than waiting for you to notice the AI is stuck.

Experimental status

Misatay is explicitly marked as an experiment. The author is interested in finding workflows that keep the engineer in control while still making AI productive. Contributions and ideas via GitHub Issues are welcome.

Practical Evaluation Checklist

• VS Code extension install — works in any VS Code environment with Copilot
• Works only with GitHub Copilot (no Claude, no other providers)
• Requires copilot.agent.autoAcceptDelay configuration for best experience
• Repo-level initialization step required per project (file-based, not extension-isolated yet)
• Changes committed to Git as the AI works — full traceable history
• AI asks for help on blockers — no silent spinning
• MIT license — open to contributions and forking

Security Notes

Misatay initializes by writing agent prompt files into your repository. Review these files before committing them if your project has strict security policies. The extension only affects the local Copilot session — it does not send data to external services beyond normal Copilot traffic.

FAQ

Q: Does Misatay work with Claude or other AI providers? A: No. Misatay only integrates with GitHub Copilot, as stated in the project README.

Q: Can I use Misatay on existing projects without initializing it? A: No. Each new repository requires running Misatay: Install Agent from the command palette before the workflow begins.

Q: Does Misatay work with Copilot Chat only, or also with Copilot Edits? A: The extension is designed for the agent mode of Copilot. Full compatibility details with specific Copilot modes are not yet documented — check the GitHub Issues for updates.

Conclusion

Misatay is a thoughtful experiment in human-AI collaborative coding. By enforcing a plan → review → commit loop on top of GitHub Copilot, it gives engineers a way to use AI agents without surrendering visibility or control. If you already pay for Copilot and want a more structured workflow, it is worth a try — just watch the experimental flag and read the GitHub Issues before relying on it for production work.