| name | ra-逐字稿提取skill |
| description | 视频内容逐字稿提取专用 skill。Use when the user provides a 抖音 or 小红书 URL and asks for逐字稿、视频转文字、听写视频或提取口播内容。Only extract transcript content; do not use this skill as the production subtitle clock, and do not summarize, rewrite, translate, title, or start production. Uses qushuiyin plus Paraformer ASR. |
ra-逐字稿提取skill
Goal
Extract a faithful Markdown transcript from a 抖音 or 小红书 video URL.
This skill is intentionally narrow and lightweight:
- resolve the video URL with qushuiyin
- send the resolved media URL to 阿里云百炼 Paraformer
- write a Markdown transcript to
01-内容生产/视频工作台/.internal/洗稿/(源隐私区——提取件默认按他人内容对待,源标题不得落入公开目录)
- print the complete transcript in the conversation
For final-video subtitle synchronization, route to ra-audio-to-subtitles.
Paragraph-level transcript timestamps from this skill are not production
caption timing.
Do not summarize, rewrite, translate, adapt, title, distribute, or start video production.
Locate The Skill
ZT_HOME="$(
for d in "$(pwd)/.codex/skills/ra-逐字稿提取skill" \
"$(pwd)/.agents/skills/ra-逐字稿提取skill" \
"$(pwd)/.claude/skills/ra-逐字稿提取skill" \
"$HOME/.codex/skills/ra-逐字稿提取skill" \
"$HOME/.agents/skills/ra-逐字稿提取skill" \
"$HOME/.claude/skills/ra-逐字稿提取skill"; do
[ -f "$d/SKILL.md" ] && echo "$d" && break
done
)"
export ZT_HOME
Use "$ZT_HOME/scripts/transcript.py" for every operation.
Workflow
-
Optional health check:
python3 "$ZT_HOME/scripts/transcript.py" --doctor
-
Extract transcript:
python3 "$ZT_HOME/scripts/transcript.py" "<抖音或小红书视频URL>"
-
If the URL is not 抖音/小红书, or if qushuiyin/Paraformer fails, stop and report the exact error. Do not try another extraction pipeline inside this skill.
Required Config
Keys are read from the project root .env first:
QUSHUIYIN_API_BASE=https://api.guijianpan.com
QUSHUIYIN_API_KEY=...
DASHSCOPE_API_KEY=...
PARAFORMER_MODEL=paraformer-v2
PARAFORMER_POLL_INTERVAL_SECONDS=3
PARAFORMER_TIMEOUT_SECONDS=240
PARAFORMER_API_KEY may be used instead of DASHSCOPE_API_KEY.
Hard Rules
- Preserve spoken content as ASR text. Do not rewrite it into a better script.
- Keep paragraph-level timestamps when Paraformer returns sentence timing.
- Save transcripts by default to
01-内容生产/视频工作台/.internal/洗稿/(隐私区). Only write into 逐字稿/ when the user explicitly says the video is their own work(此时用 --output-dir 指定);逐字稿/ 的写入白名单见其 README.
- Show the full transcript in the conversation after extraction.
- Do not call other creation, rewriting, title, image, distribution, or video-production skills during this task.
Command Options
python3 "$ZT_HOME/scripts/transcript.py" "<url>" \
--title "<optional title>" \
--output-dir "<optional output dir>"
Use --no-save only when the user explicitly does not want a .md file.