Most observability tools focus on models or cloud telemetry; this project flips the problem to the developer machine and asks: what can you learn from your local AI sessions? AI Engineer Coach analyzes on-disk AI assistant session logs and converts them into actionable developer feedback—without sending data off your machine.
What Sets It Apart
- Local, read-only analysis: all parsing and analytics run on the user's machine; the extension does not modify session files or phone home. This makes privacy and team adoption straightforward for sensitive codebases.
- Practice-focused signals: built-in dashboards surface practice scores, week-over-week trends, a Gantt-style timeline, screenshot-based coding moments, and a 7×24 activity heatmap to reveal habits and anti-patterns (45 editable detection rules).
- Output & skill telemetry: aggregates generated code volume by language/model/harness, discovers repeated prompt patterns as candidate reusable skills, and provides a rule editor + playground to tune detections against your own data.
- Multi-harness support and flexible surface: supports VS Code, GitHub Copilot app canvas, Xcode Copilot Chat, terminal CLIs (Copilot CLI), Claude/Codex/OpenCode sessions and more—so you get a unified view across tools used by your team.
Who it's for & tradeoffs
Great fit if you want private, evidence-driven coaching on how teams and individuals use AI assistants—engineers, team leads, or developer productivity teams who want reproducible practice signals rather than raw telemetry. It’s also useful for power users who want to extract prompt patterns and enforce prompt/context hygiene.
Look elsewhere if you need a packaged marketplace extension (this repo ships source and a VSIX you must build), require cloud-centralized analytics or long-term hosted telemetry, or expect production-grade model token accounting out of the box (some token features are marked disabled/experimental).
