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

1. Optional health check:

   python3 "$ZT_HOME/scripts/transcript.py" --doctor
   

2. Extract transcript:

   python3 "$ZT_HOME/scripts/transcript.py" "<抖音或小红书视频URL>"
   

3. 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