seahi/baoyu-skills
1
0
Fork 0
mirror of https://github.com/JimLiu/baoyu-skills.git synced 2026-08-07 01:13:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

feat(baoyu-youtube-transcript): add sentence segmentation and improve caching

Browse Source
This commit is contained in:
Jim Liu 宝玉
2026-03-21 22:42:43 -05:00
parent e52f92b193
commit e413ade164
5 changed files with 244 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files
skills/baoyu-youtube-transcript/SKILL.md
+9 -5
View File
@@ -98,11 +98,12 @@ youtube-transcript/
├── .index.json # Video ID → directory path mapping (for cache lookup)
└── {channel-slug}/{title-full-slug}/
├── meta.json # Video metadata (title, channel, description, duration, chapters, etc.)
├── transcript-raw.srt # Raw transcript in SRT format (cached, token-efficient for LLM)
├── transcript-raw.json # Raw transcript snippets from YouTube API (cached)
├── transcript-sentences.json # Sentence-segmented transcript (split by punctuation, merged across snippets)
├── imgs/
│ └── cover.jpg # Video thumbnail
├── transcript.md # Markdown transcript
└── transcript.srt # SRT subtitle (if --format srt)
├── transcript.md # Markdown transcript (generated from sentences)
└── transcript.srt # SRT subtitle (generated from raw snippets, if --format srt)
```
- `{channel-slug}`: Channel name in kebab-case
@@ -114,11 +115,14 @@ The `--list` mode outputs to stdout only (no file saved).
On first fetch, the script saves:
- `meta.json` — video metadata, chapters, cover image path, language info
- `transcript-raw.srt` — raw transcript in SRT format (pre-computed timestamps, token-efficient for LLM processing)
- `transcript-raw.json` — raw transcript snippets from YouTube API (`{ text, start, duration }[]`)
- `transcript-sentences.json` — sentence-segmented transcript (`{ text, start: "HH:mm:ss", end: "HH:mm:ss" }[]`), split by sentence-ending punctuation (`.?!…。？！` etc.), timestamps proportionally allocated by character length, CJK-aware text merging
- `imgs/cover.jpg` — video thumbnail
Subsequent runs for the same video use cached data (no network calls). Use `--refresh` to force re-fetch. If a different language is requested, the cache is automatically refreshed.
SRT output (`--format srt`) is generated from `transcript-raw.json`. Text/markdown output uses `transcript-sentences.json` for natural sentence boundaries.
## Workflow
When user provides a YouTube URL and wants the transcript:
@@ -148,7 +152,7 @@ Speaker identification requires AI processing. The script outputs a raw `.md` fi
- Chapter list from description (if available)
- Raw transcript in SRT format (pre-computed start/end timestamps, token-efficient)
After the script saves the raw file:
After the script saves the raw file, spawn a sub-agent (use a cheaper model like Sonnet for cost efficiency) to process speaker identification:
1. Read the saved `.md` file
2. Read the prompt template at `{baseDir}/prompts/speaker-transcript.md`