Review AI-proposed changes to your course materials before students see them.
CourseLens is for instructors who use an AI coding agent (such as Claude Code or OpenAI Codex) to update a programming or data science course. It lays out the agent's suggested edits in the order your course is taught, shows each edit in the student-facing material, and shows which files the agent changed, checked, or never opened. It accompanies the paper "The Edit Is Not the End: CourseLens for Reviewing AI-Proposed Course Updates."
- Course-order timeline: files are grouped by the week they're taught.
- Side-by-side preview: see each proposed change highlighted in the rendered lecture or assignment.
- File review analysis: upload the agent's activity log to see files it suggested changes to, checked without changing, or never checked.
- Review and follow up: mark each change Needs review, Approved, or Removed, tag files to revisit, and see everything in a Summary.
CourseLens never edits your files and doesn't judge whether a change is correct; you do.
You need Python 3.11+, uv, Node.js 20.19+, Git, and your course materials in a folder on your computer. No AI service or API key is needed.
1. Download
git clone https://github.com/dstl-lab/CourseLens.git
cd CourseLens2. Install the backend
cd backend
cp .env.example .env
uv syncOpen backend/.env and set your course folder:
COURSE_REVIEW_ROOT=/Users/zara/courses/my-course3. Install the frontend (in a second terminal, from the CourseLens folder)
cd frontend
npm install4. Add your schedule (recommended; from the backend folder)
uv run python -m scripts.init_course_config --weeks 15This creates courselens.json in your course folder with guessed weeks. Open it and fix the weeks, for example { "week": 3, "files": ["lec05", "hw1"] }. Files without a week appear under Unscheduled. See docs/courselens-config.md for all settings.
5. Start CourseLens (keep both running)
uv run uvicorn app.main:appnpm run devRun the first in the backend folder and the second in the frontend folder, then open http://localhost:5173.
- Get suggestions from your AI agent. Run your agent in your course folder and add the instructions from docs/course-change-preview-agent-prompt.md to your request, so it returns its suggestions in CourseLens format.
- Paste them into Paste generated suggestions and click Review changes.
- Upload the activity log (optional) with Upload .jsonl:
- Claude Code:
~/.claude/projects/<course-folder-name>/<session-id>.jsonl - Codex:
~/.codex/sessions/<year>/<month>/<day>/rollout-….jsonl
- Claude Code:
- Review: click View on a file to see its changes in context, set each change's status, and ⚐ Tag files that need follow-up, especially those under Not checked.
- Finish with Summary.
Set COURSE_REVIEW_ROOT=../examples/demo-course in backend/.env, start CourseLens, paste examples/demo-course-suggestions.json, and upload examples/demo-course-agent-log.jsonl. Under Not checked, notice studio03: the agent missed it.
- Everything runs on your computer and nothing is uploaded. Don't run CourseLens on a public server: it serves your whole course folder.
- Review decisions are saved only in your browser.
- Activity logs from Claude Code and Codex are supported.
- Changes inside PDFs can't be highlighted.
| Problem | Fix |
|---|---|
| "No course folder is set" | Set COURSE_REVIEW_ROOT in backend/.env and restart the backend. |
| "Couldn't reach the CourseLens backend" | Start the backend (step 5). |
| Files in the wrong week or missing | Edit courselens.json (settings) and reload. |
Andrew Hudson Yang, Susana Haing, Thaarak Sriram, Chengyun Xu, Zhiqing Wang, and Sam Lau. 2027. The Edit Is Not the End: CourseLens for Reviewing AI-Proposed Course Updates. In Proceedings of the 58th ACM Technical Symposium on Computer Science Education V.1 (SIGCSE TS 2027).
CourseLens was created by researchers at the University of California, San Diego:
| Name | |
|---|---|
| Andrew Hudson Yang | any012@ucsd.edu |
| Susana Haing | shaing@ucsd.edu |
| Thaarak Sriram | tsriram@ucsd.edu |
| Chengyun Xu | chx023@ucsd.edu |
| Zhiqing Wang | zhw131@ucsd.edu |
| Sam Lau | lau@ucsd.edu |
MIT. Everything in examples/ is made up and covered by the same license.