Article To Podcast Script

Article To Podcast Script

Workflow

1. Read the input Markdown article and identify its title, thesis, key sections, facts, examples, and conclusion.
2. Confirm the requested speaker count is an integer from 1 to 4. If it is missing or outside this range, ask for a valid count before writing the script.
3. Write a podcast-ready Chinese dialogue script by default, using only labels in the form Speaker N: where N is from 1 to the requested count. If the user explicitly specifies another language, write the script in that language instead. Transform the article's ideas into the speaker's own first-person viewpoint, as if the speaker is directly sharing their thinking on the topic, not reviewing, summarizing, or reacting to an article.
4. Save the result as a .txt file in the current project's markdown directory. Create that directory if needed.
5. Report the output path and any important assumptions.

Use scripts/save_podcast_script.py to write the final text:

python3 /path/to/article-to-podcast-script/scripts/save_podcast_script.py ARTICLE.md SPEAKER_COUNT < draft.txt

The script validates the speaker count, creates ./markdown, builds a stable output filename from the article name, and writes stdin to the .txt file.

Script Requirements

• Start directly with Speaker 1:. Do not add a title block, cast list, markdown formatting, stage directions, or bullets.
• Default to natural spoken Chinese for the dialogue content unless the user requests another language. Keep the Speaker N: labels in English exactly as shown.
• Preserve English terms that appear in the article, including acronyms, product names, model names, paper titles, project names, protocol names, company names, and technical terms. Do not translate them into Chinese unless the article itself already provides an established Chinese rendering.
• Keep every line as one speaker turn:

Speaker 1: 大家好，欢迎收听本期节目。今天我想聊一个很关键的判断...
Speaker 2: 没错，真正的核心不是表面的新闻，而是背后的传导链条...

• Use exactly the requested number of speaker labels across the script. For one person, use only Speaker 1:.
• Rotate speakers naturally for 2 to 4 people; do not force equal length if the conversation would sound unnatural.
• Preserve the article's core claims and important evidence. Do not invent facts, names, numbers, or citations.
• Convert dense exposition into spoken language: short sentences, natural transitions, reactions, clarifying questions, and concise analogies.
• Make the script self-contained for listeners who have not read the article.
• Avoid excessive filler, repeated greetings, ads, music cues, timestamps, sound effects, and production notes.
• Avoid any AI-generated trace, meta-writing, source-processing disclosure, or reading-response framing. Do not mention "我作为AI", "AI生成", "资料边界", "公开资料显示", "不是基于全文精读", "这篇文章", "本文", "这篇读书笔记", "原文", "作者认为", "评论者认为", "根据材料", "我读完", "读到这里", "我理解这篇文章", "文章里提到", "这篇报告", or similar source-framing phrases. It is fine to say "AI" when AI is the actual topic or an English term from the source.
• Never present the episode as a book report, article review, article summary, reading note, or post-reading reflection. The listener should feel the speaker is sharing their own structured view directly, not explaining how they understood a source text.
• Put the host inside the source article's knowledge and evidence boundary while removing the source frame. The host may say "我的判断是", "我会这样看", "这里真正的问题是", "如果把它落到实践里", but must not pretend to have experiences, authority, interviews, fieldwork, proprietary data, or personal relationships that are not in the article.
• Rewrite source-boundary paragraphs into natural first-person caveats or omit them when they do not help listeners. For example, instead of saying "这本书的公开资料边界也要先讲清楚...", say "我更愿意把它当成一套思考复杂问题的框架，而不是照搬案例的操作手册。"
• Keep the voice like a real human host preparing a thoughtful episode: specific, opinionated within the article's evidence, and free of template-like transitions.