English | 中文

This is an agent skill suite (9 atomic /command skills) + CLI for Tongji Look (look.tongji.edu.cn):

• login via Tongji IAM SSO (Playwright),
• list courses (recent list or full search),
• transcribe a lecture to SRT + TXT,
• generate a lecture timeline outline (*_timeline.txt) from SRT (agent-generated, Simplified Chinese),
• download lecture slide snapshots (filename includes snapshot time),
• then let the current agent write a Markdown study note from transcript + slide images.

Commands

Command	Description
/setup	Configure credentials, check dependencies (Python, Node.js, ffmpeg, vision-support, TeX), set workspace
/list	List courses, search by keyword, interactive selection
/trans	Transcribe one lecture to SRT + TXT; optionally download slides in parallel
/note	Generate study notes + timeline outline from transcript + slides
/add	Import supplementary materials (PDF, PPTX, DOCX) into a lecture session
/wiki	Build and locally serve the static course knowledge base
/page	Deploy the built course wiki to GitHub Pages via gh CLI
/cheatsheet	Generate A4 cheat sheet (LaTeX or HTML) from course notes
/ralphtrans	Batch transcribe all lectures in a course with checkpoint/resume

Install

Recommended: npx skills

If your agent supports the skills protocol:

npx skills install https://github.com/walkerkiller/look-tongji-notes

Claude Code Marketplace

In Claude Code:

/plugin marketplace add https://github.com/walkerkiller/look-tongji-notes
/plugin install look-tongji-notes

Other Platforms

Platform	How
Claude Code	Marketplace (above) or point plugin root to this repo
Codex CLI	.codex-plugin/plugin.json → ./skills/
Cursor	.cursor-plugin/plugin.json → ./skills/
Gemini CLI / OpenClaw / OpenCode / Hermes Agent	plugin.json → ./skills/

Usage (CLI)

<SKILL_DIR> is the folder that contains SKILL.md.

Setup credentials (recommended):

python "<SKILL_DIR>/../../scripts/look_tongji.py" setup

List recent courses:

python "<SKILL_DIR>/../../scripts/look_tongji.py" list

Search courses by name (recommended for accuracy, calls get_all_courses internally):

python "<SKILL_DIR>/../../scripts/look_tongji.py" list --all --query "<COURSE_NAME_KEYWORD>"

Transcript only (transcribe, aliases transcript / trans):

python "<SKILL_DIR>/../../scripts/look_tongji.py" transcribe --lecture-url "<LECTURE_URL>"

Combined mode (note, runs transcript + slide in parallel by default):

python "<SKILL_DIR>/../../scripts/look_tongji.py" note --lecture-url "<LECTURE_URL>"

Note style (affects how the generated note is formatted):

python "<SKILL_DIR>/../../scripts/look_tongji.py" note --lecture-url "<LECTURE_URL>" --note-style dialogue

Supports standard (lecture notes, default) and dialogue (Q&A format).

[!TIP] The CLI detects lectures shorter than 1 hour and prints a non-blocking warning: [Warning] 课时不足1小时 — suggesting a retry, as very short lectures may indicate an incomplete recording or playback error.

Download slide snapshots for a lecture:

python "<SKILL_DIR>/../../scripts/look_tongji.py" slide --lecture-url "<LECTURE_URL>"

If throttling is suspected, reduce concurrency:

python "<SKILL_DIR>/../../scripts/look_tongji.py" slide --course-id "<COURSE_ID>" --sub-id "<SUB_ID>" --concurrency 2 --retries 5

In the /note workflow, the agent generates a timeline outline after the SRT subtitle file is produced:

• File: ./tongji-output/<course_id>_<sub_id>_timeline.txt
• One line per segment (Simplified Chinese), format: Start-Over：Stage Main Content
  • Example: 00:00-05:30：Course Orientation and Assessment Description
• Skip only if the user explicitly says: no outline / no timeline.

Artifacts are written to the configured course-wiki workspace by default.

Agent Note

When a user says /setup / /list / /trans / /note / /wiki / /add / /page / /cheatsheet / /ralphtrans, follow the corresponding skills/<name>/SKILL.md and run the matching CLI commands in scripts/look_tongji.py. For /note, default to running transcript + slide download in parallel; only skip slide download when the user explicitly asks not to download slides/PPT. When writing notes, use both transcript output and slide images by default. If the user provides a course name, prefer list --all --query ... to avoid missing courses that are not in the recent list.

After notes are generated or updated, rebuild and preview the site with:

python "<SKILL_DIR>/../../scripts/look_tongji.py" index
python "<SKILL_DIR>/../../scripts/look_tongji.py" build
python "<SKILL_DIR>/../../scripts/look_tongji.py" serve --port 8765

The generated workspace can also become the user's own GitHub Pages repository. It includes:

• llmwiki/
• index.py
• build.sh
• serve.sh
• .github/workflows/wiki-checks.yml
• .github/workflows/pages.yml

Notice / Compliance

[!CAUTION] It is strongly recommended to set your Tongji account/password via the CLI (setup) before asking the agent to generate notes. Do not paste passwords into chat.

[!NOTE]

• Inspiration and parts of the code are from: Fudan_iCourse_Subscriber
• This project is intended for personal learning and review only, and for technical communication. It does not save full video files by default.
• Users must comply with the relevant platform rules and school policies. Any misuse (including re-distribution of copyrighted course videos/audios) is the user's responsibility.
• When logging in outside the campus network (or without Tongji VPN), enhanced authentication may be triggered. Keep this in mind if you run the agent remotely.

ToDo

• Generate course knowledge base static site skeleton (courses, sessions, video, timeline, i18n controls).

• Support login flows with enhanced authentication.

• Add course-level LLM wiki + notes database foundation.

• Build a standalone TUI/GUI tool for manual transcription/notes/Q&A.

Best practices

• Generate subtitles + notes for the latest lecture:

  • /note help me generate subtitles and notes for the latest lecture

• Generate subtitles + notes for a named course:

  • /note help me generate subtitles and notes for today's Advanced Mathematics lecture

• List recent courses and let the user choose:

  • /trans list recent courses and let me choose one to process

• If the agent cannot find the course:

  Use list --all --query "<keyword>" to search the full course list, or open the platform and copy the lecture URL.

  

  then say: /trans here is the link，generate note

Contributing

Coming soon.