Misatay – VS Code Extension for Stay-in-the-Loop AI Coding

TL;DR

TL;DR: Misatay is a VS Code extension for GitHub Copilot that keeps engineers in the loop — plan to a repo file, watch AI work commit-by-commit, review in context, and only hit merge when you actually believe the work is done.

Source and Accuracy Notes

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

• Project page: github.com/dshearer/misatay — README read in full
• VS Code Marketplace: open-vsx.org (not verified — marketplace page not accessed)
• License: MIT — verified from repository
• HN launch thread: news.ycombinator.com/item?id=46928648

What Is Misatay?

Misatay is an open-source VS Code extension that layers a structured human-in-the-loop workflow onto GitHub Copilot. Where agents like Cursor or Windsurf aim to offload as much coding as possible, Misatay’s philosophy is the opposite: you should always be involved and in charge.

The name comes from “me + satay” — a pun on being hands-on. The project’s manifesto frames this clearly:

To make the best use of AI, AIs and people need to work side-by-side. AI will boost productivity by empowering human workers, to understand more and to build more.

It is explicitly an experiment. The author is looking for ways to keep programming fun, find tight human-AI development loops, and help engineers grow rather than get sidelined.

Setup Workflow

Step 1: Install the Extension

Misatay is a VS Code extension distributed through the marketplace. Open the command palette (Cmd+Shift+P on macOS) and search for “Misatay: Install Agent.” This one-time initialization step adds an agent prompt and skill files to your repo.

The developer acknowledges this is a friction point and hopes to move these files into the extension itself in a future release, eliminating the repo-level install step.

Step 2: Configure Auto-Accept Delay

For the best review-and-approve workflow, Misatay recommends setting copilot.agent.autoAcceptDelay to a value greater than zero (the suggested range is 5–10 seconds).

This means: changes land in your editor automatically after the delay, giving you a window to spot-check before acceptance. Setting it to zero disables auto-accept entirely, requiring you to manually accept every change — which works but defeats the efficiency goal.

To configure in VS Code:

1. Open Settings (Cmd+,)
2. Search for copilot.agent.autoAcceptDelay
3. Set your preferred delay in seconds

How the Misatay Workflow Works

The Misatay workflow is structured around four practices:

Plan to the repo. You write your feature plan in a repo file. If the Copilot session resets or you switch contexts, the plan is preserved — Misatay won’t lose track of what you were building.

AI commits as it goes. Misatay commits its own changes to Git as it works through tasks. This keeps a clean record of which changes belong to which task — useful when reviewing a large feature branch before merge.

In-context code reviews. When Misatay asks you to review, it shows you what changed and why in the context of the original task. You are not rubber-stamping blind diffs — you are evaluating work against a stated goal.

AI asks for help. When Misatay hits a wall, it asks you rather than spinning tokens on dead ends. The README frames this as a cost-saving feature: wasted tokens on looping are avoided.

Manifesto and Project Goals

From the README’s Manifesto section:

• To make the best use of AI, AIs and people need to work side-by-side
• AI boosts productivity by empowering human workers to understand more and build more
• Keep programming fun
• Find ways to provide assurance of correctness for AI-generated code
• Find a tight loop of development involving people and AIs
• Find ways that programming with an AI can help the human learn and grow

The project status is marked as experimental. The author actively invites feedback via GitHub Issues.

Practical Evaluation Checklist

Based on a README-only review (not a live installation):

• [ ] VS Code extension, GitHub Copilot required (not Claude, not Cursor)
• [ ] Repo-level install adds prompt and skills (one-time per repo)
• [ ] Auto-accept delay configurable for review windows
• [ ] Changes commit to Git as the AI works
• [ ] Code review is task-contextual, not a raw diff
• [ ] MIT license
• [ ] Explicitly marked as experimental — not production-hardened

Security Notes

Misatay initializes a prompt file and skill set into your repo. Treat these like any other committed configuration:

• Review what gets added before the first commit
• The skill files execute as part of your Copilot session — keep them under the same review discipline as pre-commit hooks
• Repo access scope is whatever your Copilot session already has

FAQ

Q: Does Misatay work with Claude Code or other AI editors? A: No. The README explicitly states it only works with GitHub Copilot.

Q: How does this differ from just using Copilot Chat? A: Misatay adds a structured plan-and-review loop. Copilot Chat is ad-hoc by comparison. Misatay saves plans to repo files (surviving session resets), commits AI work in discrete units, and frames reviews against the original task.

Q: Is this production-ready? A: The project status is marked as “an experiment.” The author uses it for real work but acknowledges it is not hardened. Review the GitHub Issues before relying on it for anything critical.

Conclusion

Misatay is a small, opinionated VS Code extension that tries to solve the trust problem in AI coding: when the agent says “All done!” — do you actually believe it? By keeping plans in the repo, committing AI work incrementally, and framing reviews around tasks, Misatay aims to make the answer yes more often than it would be with raw Copilot.

The tradeoff is clear: this only works with GitHub Copilot, requires a one-time per-repo setup, and the project is experimental. If those constraints fit your stack and you want a tighter human-in-the-loop on AI coding, it is worth a look.