# Trans

> Transcribe a single Tongji Look lecture video to SRT + TXT, optionally download slide snapshots in parallel.

- Skill: `walkerkiller/trans` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add walkerkiller/trans`
- Raw SKILL.md: https://api.skillmd.com/api/skills/walkerkiller/trans/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Docs & Writing
- License: MIT
- Author: WALKERKILLER (https://skillmd.com/u/walkerkiller)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/walkerkiller/trans

---


# Transcribe

Transcribe one lecture video from look.tongji.edu.cn and optionally download slide snapshots.

## When to Use

- User says `/trans` or "transcribe this lecture".
- User provides a lecture URL or course_id + sub_id.

## Workflow

1. Resolve lecture with URL or IDs:

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

```bash
python "<SKILL_DIR>/../../scripts/look_tongji.py" transcribe --course-id "<ID>" --sub-id "<ID>"
```

2. Transcribe + download slides in one run:

```bash
python "<SKILL_DIR>/../../scripts/look_tongji.py" transcribe --course-id "<ID>" --sub-id "<ID>" --slide
```

3. The CLI prints real-time progress. Stdout contains result paths.

## Non-Interactive / Batch Usage

When running from an automated pipeline, avoid `input()` prompts with these flags:

| Flag | Description |
|------|-------------|
| `--sub-id` | Specify lecture sub_id directly (skips interactive lecture picker) |
| `--course-id` | Specify course ID directly (skips interactive course selection) |
| `--slide` | Download slides after transcription |
| `--concurrency N` | Slide download concurrency (default 4) |
| `--retries N` | Slide download retries (default 3) |

## manifest.json — Required for Wiki Build

**transcribe does NOT automatically create `manifest.json`.** After transcription, you MUST create it manually so the wiki indexer can find this lecture.

Create `manifest.json` in the session's `原始数据/` directory with this structure:

```json
{
  "course_id": "COURSE_ID",
  "sub_id": "SUB_ID",
  "course_title": "课程名称（真实中文名）",
  "session_title": "YYYY-M-D 第X节",
  "base_name": "COURSE_ID_SUB_ID",
  "artifacts": {
    "srt": "F:\\workspace\\raw\\课程名称\\YYYY-M-D 第X节\\原始数据\\COURSE_ID_SUB_ID.srt",
    "txt": "F:\\workspace\\raw\\课程名称\\...\\COURSE_ID_SUB_ID.txt",
    "notes": "",
    "timeline": "",
    "slides": "F:\\workspace\\raw\\课程名称\\...\\slides\\index.json"
  },
  "duration_seconds": 5580
}
```

**Critical path rules:**
- Use **absolute paths** for all artifact paths (relative paths break on Windows)
- `duration_seconds`: must be ≥ 3600 (1 hour); if < 3600, the lecture is marked不合格 and wiki shows it as 缺失
- If no slides were downloaded, leave `slides` field empty (`""`)
- Fields `notes` and `timeline` are filled by `/note` command later

## Artifacts

- `<course_id>_<sub_id>.srt` — subtitle with timestamps
- `<course_id>_<sub_id>.txt` — plain text transcript
- `<course_id>_<sub_id>.json` — metadata
- `slides/` — slide images + index.json (if --slide)
- **`manifest.json` — MUST create manually after transcription**

## One Session Per Call

For batch transcribing, use `/ralphtrans`.

## Where \`<SKILL_DIR>\` Points

\`<SKILL_DIR>\` is the directory containing this \`SKILL.md\`. Shared scripts (\`look_tongji.py\`, \`timeline_tools.py\`, \`tongji_backend/\`) and references live two levels up in the repository root (\`<SKILL_DIR>/../../scripts/\` and \`<SKILL_DIR>/../../references/\`).