MLView

MLView

Ask the assistant in your editor to explain an ML workflow, then explore its
answer as an interactive diagram linked to the source. MLView is a skill for
GitHub Copilot, Codex and Claude Code, a document format, and a VS Code
viewer. The assistant reads the code and configuration and writes the diagram:
steps, connections and findings, with quotes from the files it read.

Preview. The format and the viewer can still change between releases,
and fixes go into the next release.

The diagram of samples/configured_training, with a step selected.

How it works

- MLView needs no model API key of its own. Your assistant supplies the model,
  its tools and its permissions.
- The skill's helper checks a draft's structure, cited paths and exact quotes,
  records file hashes, and publishes a *.mlview.json file.
- Neither the helper nor the viewer imports or runs your code.
- Opening, filtering and navigating a diagram never call a model.

Install

Each release attaches the extension (a .vsix file) and the two skill ZIPs
to its page under Releases.

- Download the .vsix file. Install it with VS Code's
  Extensions: Install from VSIX... command.
- Download the skill ZIP for your assistant and extract it at the root of
  your project.
  The skill, per assistant
  says which ZIP is for which assistant.
- For Claude Code, the skill is also packaged as a
  plugin.

To work from a clone of this repository instead, build the extension as
docs/development.md describes. You need
Python 3.10 or newer with the packages of requirements-dev.txt in a
virtual environment, and Node 20.18.1 or newer. Then copy the skill into
your project with the installer:

The installer copies the skill into the project's .agents/skills/mlview/
folder. For Claude Code, add --destination…