ReviewDistill

Distill the comments you write in a LaTeX paper into a reusable agent skill to save your proofreading time.

Documentation · GitHub

What is ReviewDistill?

You already write comments in the .tex file. Those remarks stay in that paper. ReviewDistill groups them into labels and exports a skill you reuse: an agent applies the same checks on the next paper.

You can add an API key if you want an LLM to suggest labels. Comments and keys stay on your computer.

Start using it

Clone this repository. You need Python ≥ 3.11. Then:

pip install -e .

That puts the reviewdistill command on your PATH and compiles the workbench. You do not need Node to run the UI. Compiling it from this repository needs Node ≥ 22 and pnpm once.

In a paper repository:

1. reviewdistill init --name paper-01 --command myremark
2. Proofread with \myremark{...} (or the commands you passed to init). Define \newcommand{\myremark}[1]{#1} in the preamble if needed.
3. reviewdistill extract (add --watch to keep extracting).
4. reviewdistill ui → http://127.0.0.1:8765. Pick a leaf label, or Label with AI if a key is set. Verify stamps the observation; Delete removes a bad extract.
5. Repeat on the next paper; the taxonomy is reused.
6. Export in the workbench, or reviewdistill export (writes SKILL.md)
7. Put that file in .cursor/skills/scholarly-review/ (or the equivalent for another coding agent).

The full guide: Introduction, Install, Your first paper, Label comments, Export.

To see the data folder: reviewdistill paths, or header Settings → Data. To copy it somewhere else: reviewdistill paths move ~/Documents/reviewdistill.

Run the documentation site locally: pnpm --dir website start.

For developers

Client source: client/. Docs site: website/. Product spec: docs/spec/README.md, docs/spec/data-schema.md, docs/spec/comment-identity.md.

A GitHub Release whose tag matches pyproject.toml (v0.1.0 or 0.1.0) publishes to PyPI via Trusted Publishing.

License

ReviewDistill is released under the MIT License.