Extract Douyin/Xiaohongshu Transcripts with a Narrow Agent Skill

Where this fits

A content researcher pastes a Douyin link into their agent and asks for the spoken script. They don't want a summary, a rewrite, or a translation—they need the exact words, in Markdown, fast. This skill answers that specific request: it takes a 抖音 or 小红书 URL, resolves the video source via qushuiyin, runs it through Alibaba Cloud's Paraformer ASR, and returns a faithful transcript. It deliberately stops at extraction—no summarization, no title generation, no subtitle timing.

Why agents benefit

• Narrow scope prevents scope creep: The skill refuses to summarize, rewrite, translate, or produce titles, so your agent won't accidentally drift into content production when the user only asked for a transcript.
• Clean output format: It writes a Markdown transcript to a designated private directory and also prints the full transcript in the conversation, making it easy to pipe into downstream analysis or editing steps.
• URL handling is built in: No need to manually scrape or download videos; the skill resolves the URL and feeds the media directly to the ASR service.
• Privacy-aware storage: Transcripts land in an internal .internal/ folder, treating extracted content as third-party material—important for teams that separate source research from production assets.
• Clear separation from subtitle work: The skill explicitly warns that its paragraph-level timestamps are not production caption timing, routing users to ra-audio-to-subtitles for that job.

Practical scenarios

Competitive research on short-video scripts

A marketer collects competitor Douyin videos and needs the exact spoken copy to analyze hooks, pacing, and CTAs. The agent fetches each transcript without altering the wording, letting the marketer compare scripts side by side.

Content repurposing (with human review)

A writer wants to adapt a Xiaohongshu video into a blog post. The agent extracts the transcript, then the writer decides what to keep, rewrite, or expand—the skill never pre-empts that editorial judgment.

Archive for internal reference

An editorial team keeps a library of video transcripts for reference. The skill stores them in a private directory, avoiding accidental exposure of source titles in public folders, and keeps the conversation transcript accessible for quick lookup.

Add it to your agent workflow

Install with npx:

npx skills add Pluviobyte/rnskill --skill ra-逐字稿提取skill

Then invoke it naturally in your agent prompt:

User: Here's a Douyin URL: https://v.douyin.com/xxxx. Extract the 逐字稿.
Agent: (runs the skill, returns the full Markdown transcript)

The skill locates itself in standard skill directories (.codex/skills, .agents/skills, .claude/skills) and will resolve automatically after installation.

Compare before adopting

Before committing, compare against other transcription skills on three axes:

• Quality signals: Does the ASR handle Chinese dialects, filler words, and background noise well? Paraformer is strong, but test with your actual video types.
• Maintenance freshness: Check the GitHub repo (currently 1425 stars) for recent commits and issues—ASR endpoints and URL resolvers drift over time.
• Workflow fit: Does your agent already have a subtitle pipeline? If so, ensure this skill's strict separation doesn't conflict with your existing production flow.

Also consider alternatives like local Whisper-based skills if you need offline processing or custom model tuning.

Why it is worth tracking

This skill is a good example of a single-purpose, well-bounded agent capability. Its 1425 stars and clear README show community interest and active maintenance. Evaluate it when you frequently handle Douyin/Xiaohongshu transcription requests and want a reliable, privacy-conscious extraction step that doesn't overstep into content creation. For more agent-skill discoveries, check the OpenAgentSkill directory for similar narrow-purpose tools.