godgod126/listening-script-kit/listening-script

Listening Script

Produce a script the listener can follow without seeing the original layout. Preserve the requested length, language, detail, voice, and editorial scope. A request to make text listenable does not automatically authorize summarizing it.

Establish the source boundary

Use the supplied text and cited source material. Keep author claims, estimates, observations, and verified facts distinct. Do not describe supplied AI writing as independently fact-checked. If essential information is missing, flag it in separate notes or ask a focused question; do not fill gaps with plausible stories.

If the user asks for proofreading only, preserve structure and wording and correct only the agreed errors. If they ask to produce a new episode from a topic, identify that as research and writing rather than source-preserving adaptation; use only the research tools and sources actually available and authorized.

Write for a listener

• Replace references such as “the above table” with their actual meaning. Introduce the subject before pronouns become ambiguous. Use brief transitions where a visual heading previously carried the structure; do not force a fixed opening, dramatic hook, or closing summary.
• For a table, name the compared items and units, then express the relevant relationship in sentences. Preserve baselines, dates, sample sizes, ranges, negative results, and exceptions. Do not read every cell unless the requested detail requires it. Put omitted detail in notes and identify substantial omissions.
• Explain formulas through quantities and relationships, including denominators and conditions. Keep the exact formula in notes if pronunciation would obscure it. For code, explain what it does rather than reading punctuation, unless the user specifically needs code read aloud.
• Preserve numbers and scope. “From 10 to 12” is a rise of two units or twenty percent, not twenty percentage points. “Three of twelve tested devices” is not “most phones.” A benchmark is not a real-user outcome. A hypothetical example must stay hypothetical.
• Keep uncertainty audible: “estimated,” “in this sample,” “as of this date,” and “the source does not establish” must not disappear. Translate unfamiliar acronyms at first mention when the source supports the expansion. Keep proper names; do not invent pronunciation claims.
• Remove Markdown syntax from the spoken body, not the meaning it carries. A claim attributed to a person or study still needs spoken attribution. Move raw URLs and detailed bibliographic entries into source notes when they need not be read aloud.

Deliver separately usable text

Default to a title, a plain-text narration body, and separate source/editorial notes. Follow a different format when requested. Keep the narration free of editor instructions, fact-check checklists, references intended only for reading, and forced product mentions. Explain material compression or unresolved claims in notes.

For a long source, preserve continuity and a consistent glossary across sections. If only part can be processed, state the covered portion; do not silently truncate or imply the entire document was adapted. Treat an audio duration as an estimate tied to word count and delivery rate, never a measured recording.

Before delivery, compare the script against the source for changed numbers, causality, attribution, qualifiers, and omissions. Mentally check whether each section works without its original chart or screen. This skill produces text; it does not synthesize audio, publish, install software, or access accounts. Do not request keys or upload source material to an additional service.

Read references/worked-example.md when a source contains a comparison table and uncertainty; it shows a small adaptation and the checks that preserve its limits.

When the user needs a Chinese starter request for book notes, dated news materials, or a sourced history question, read references/chinese-starters.md. Its prompts are examples, not a fixed length or permission to research, synthesize audio, publish, or add product mentions to the user's script.