Microsoft's AI Engineer Coach Turns Your Coding Logs Into a Private Skills Dashboard
Microsoft open-sources a VS Code extension that reads your AI coding session logs and coaches you on how to use agents better — all locally, zero telemetry.
PRO- Microsoft engineers open-sourced AI Engineer Coach, a VS Code extension that reads local AI session logs and scores your coding habits.
- Supports GitHub Copilot, Claude Code, Codex CLI, OpenCode, Copilot for Xcode, and Copilot CLI — all in one dashboard.
- 45 editable anti-pattern detection rules catch bad habits like context-free prompts, mega-sessions, and unreviewed generated code.
- Fully local and read-only: all parsing runs on-device with zero telemetry and no extra token costs for core analytics.
- Includes a Skill Finder that detects repeated prompt patterns and converts them into reusable skills, plus a Context Health score for your repos.
- Not on the VS Code Marketplace — you build the
.vsixyourself vianpm ci && npm run package, MIT licensed.
We obsess over which AI model is fastest, cheapest, or best at coding. What almost nobody measures is whether we are actually getting better at working with these tools. AI Engineer Coach is Microsoft's answer to that blind spot: an open-source VS Code extension that reads the session logs your AI coding assistants already write to disk, and turns them into a private analytics dashboard that coaches you on your own habits.
AI Engineering Coach is an open-source VS Code extension built by a group of Microsoft engineers, with the tagline "better agentic engineering" , analyze your AI coding assistant usage across any harness, from one dashboard. It's a community effort by Microsoft employees, explicitly not an official Microsoft product or part of any support offering. It's MIT-licensed and free.
The Problem It Solves
Its main value is helping developers measure and improve their AI-assisted coding process instead of treating agent use as an untracked black box. Think about what you actually know about your own AI workflow: which prompts you repeat, how often you review generated code before shipping it, whether your repos even give the agent enough context to succeed. The answer for most people is: not much.
In real work, the human and the harness matter just as much as the model. The same model can produce very different results depending on the quality of the prompt, the available context, the instructions in the repo, the tool loop, and the developer's review habits. AI Engineer Coach focuses on exactly that layer.
One Dashboard, Every Harness
The extension is designed to help developers improve how they work with tools such as VS Code, Claude, Codex, OpenCode, GitHub Copilot for Xcode, and Copilot CLI. Each tool gets a dedicated parser that normalizes its logs into a single shared data model, so you get unified analytics regardless of which assistant you use day-to-day.
Here's what it reads today:
- GitHub Copilot (VS Code / Insiders)
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
