Look-Tongji-Notes

This repository is frozen. Development continues in Coursive; do not start new work, pull requests or releases here unless the maintainer asks for a fix.

Purpose

Root of the Look Tongji Notes skill repository. Provides a multi-agent skill + CLI for processing lectures from Tongji Look (look.tongji.edu.cn): IAM SSO login, course listing, lecture transcription (SRT/TXT via BcutASR), slide snapshot download, supplementary material import, timeline outline generation, Markdown study-note writing, and static course knowledge-base site generation via the vendored llm-wiki library. Packaged as both a standalone CLI and an installable skill for Claude Code, Codex CLI, Cursor, Gemini CLI, OpenClaw, OpenCode, Hermes Agent, and GitHub Copilot.

Key Files

Subdirectories

For AI Agents

Working In This Directory

• This is the source repository root. The 9 installable command skills live under skills/<name>/SKILL.md in flat layout.
• SKILL.md is the primary compatibility index: it maps /command trigger phrases (/setup, /list, /trans, /note, /wiki, /add, /cheatsheet, /page, /ralphtrans) to CLI commands.
• Credentials are stored in <skill_root>/.env (auto-ignored by .gitignore). Auth cache goes in <skill_root>/state/.
• The workspace config (course wiki path) is stored outside the skill tree so skill updates do not destroy it.
• Lecture artifacts follow this layout: raw/<course_name>/<session>/原始数据/{transcript,slides,materials,manifest.json}.

Testing Requirements

• Full end-to-end testing requires a valid Tongji IAM account and access to look.tongji.edu.cn.
• Unit-testable components: URL parsing (_extract_ids_from_url), timeline normalization (scripts/timeline_tools.py), env-file parsing (_parse_env_lines), agent-name normalization (_normalize_agent_name).
• Slide download can be tested for concurrency/throttling behavior with --concurrency 2 --retries 5.
• ASR testing requires actual lecture media with audio streams.

Common Patterns

• SKILL.md defines agent-facing workflow. CLI commands in scripts/look_tongji.py are the backend.
• Lecture identification: prefer --lecture-url for best-effort parsing, or explicit --course-id + --sub-id.
• Course search: prefer list --all --query "<keyword>" for accuracy over the recent-courses list.
• The note command runs transcript + slide download in parallel by default.
• Notes are written by the AI agent after CLI artifacts complete; the CLI never generates study-note content itself.
• Site rebuild: run index then build after writing notes, then serve --port 8765 for local preview.

Dependencies

Internal

• scripts/tongji_backend/ provides auth, client, transcriber, workspace, and site-builder modules consumed by scripts/look_tongji.py.
• llmwiki/ (vendored) provides the static site generator consumed by workspace helpers.
• scripts/timeline_tools.py provides SRT sampling and timeline normalization utilities.

External

• requests — HTTP client for Tongji API and slide downloads
• playwright — Browser automation for IAM SSO login
• python-dotenv — .env file loading
• markdown / markitdown[all] — Supplementary material Markdown conversion
• ffmpeg (system) — Audio extraction from lecture video streams
• BcutASR (Bilibili cloud API) — Free cloud speech-to-text for transcription