CommunityWriting & Editinggithub.com

Livin21/scripting-explainer-shorts

Use when writing or fact-checking the narration for a short (under 60 seconds) educational video about how something works, when choosing or verifying a hook, when a source won't open or contradicts itself, or when writing on-screen labels, descriptions or web copy that state facts about an episode.

What is scripting-explainer-shorts?

scripting-explainer-shorts is a Claude Code agent skill that use when writing or fact-checking the narration for a short (under 60 seconds) educational video about how something works, when choosing or verifying a hook, when a source won't open or contradicts itself, or when writing on-screen labels, descriptions or web copy that state facts about an episode.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/Livin21/behind-the-things-skills/tree/HEAD/skills/scripting-explainer-shorts

Installed? Explore more Writing & Editing skills: steipete/notion, langchain-ai/langchain, bytedance/podcast-generation · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Scripting explainer shorts

Overview

An explainer is only as good as its weakest claim. Every factual sentence and every on-screen number needs a source you actually opened. A hook that sounds great is usually the least true line in the draft, so verify it first and rewrite it if it fails.

Process

  1. Verify the hook first. Find a primary source (a standards body, manufacturer, government agency, university or published source code) that states it. Then:

    • It fails: rewrite it to what the source says. "The stripes aren't the number" became "the barcode doesn't know the price"; "a speck of radioactive metal" became "something radioactive" (it's an oxide, too small to see).
    • It holds with a nuance: keep it, adjust words to match the source, and carry the nuance in a beat ("doesn't cook from the inside out" → the waves do heat about an inch in).
    • It's provable only for some cases: keep it general, attribute every specific number to its source ("Android's default…", "Windows…"), and record the gap as partly verified.
  2. Research the beats. Open each source rather than trusting a search snippet, and quote the supporting line. Pick the strongest source that states the claim:

    RankSourceExample
    1Standards bodies, specifications, published source codeEMVCo, ISO/IEC catalogue page, NFC Forum, GS1, AOSP
    2Government agencies, regulators, central banksNRC, EPA, FDA, USDA, Federal Reserve, ECB
    3The maker's own documentationVisa, Mastercard, NXP datasheets, Microsoft Learn
    4Universities, textbooks, peer-reviewed papersuniversity course notes, conference papers
    5Reputable explainers and pressonly when 1–4 don't state it

    Never cite a mirror or re-host (IPFS or other Wikipedia copies, PDF re-uploads), a content farm, an SEO blog or vendor marketing. If only a mirror opens, find the original page or an archived copy of it. Wikipedia (or a copy of it) can lead you to a source: follow the article's reference, open it, and cite that.

    • A page won't open: try another fetcher, then an archived copy. If you use an archive, label it. If nothing loads, hedge the line or cut it.
    • The source contradicts itself or physics: speak the correct figure plainly ("about five inches") and show the working in sources.md. Note the source's own figure under "Simplified on purpose". Example: c ÷ 2450 MHz = 12.2 cm ≈ 5 in, not the USDA's rounded "4 in".
    • An inference (true by reasoning, but no source says it in one line): mark it as an inference in sources.md.
  3. Write the narration (shape below).

  4. Fit the budget. The cap is 54 × (the voice's words per second), taken from the project's CHANNEL.md; it's about 140 words for a 2.6 words-per-second voice. The spoken sign-off is included. Count with wc -w; a hyphenated word counts as one.

  5. Scene table and packaging (see the template in templates.md).

  6. Write sources.md with the template: one row per claim, including every number shown on screen in tags or readouts, then "Simplified on purpose" and "Partly verified".

Narration shape

  1. Hook (0–3 s): a surprising true fact, said over the object, which is fully drawn in frame one.
  2. The reveal: open the object and show what's inside. If the "thing" is abstract (an icon, a signal), reveal the device that produces it, such as the phone's radio for Wi-Fi bars. Pick the drawn view so the reveal exposes the insides, and make the verb match the motion ("lift the cover").
  3. Three beats: what's inside → how it works → what nobody expects.
  4. Twist line: the fact people repeat to a friend.
  5. Sign-off: the channel's fixed line.

Write for the ear: short sentences, and spell numbers as spoken. Add voice notes for risky words, which become respellings in the TTS prompt.

Every other text

On-screen tags, YouTube descriptions, captions and site pages may only state claims that are already in sources.md. Adding a "small" fact while writing copy is how errors ship.

Common mistakes

MistakeFix
Keeping a catchy hook that's "basically true"Rewrite it to what the source says
Repeating a source's rounding errorCheck against physics or a second source; show the working
Tags with unsourced numbersAdd a source row for every on-screen figure
148 words for a slow voiceUse the voice's measured pace; cap at 54 × pace
Silently dropping an unopenable sourceHedge or cut the line; record it as partly verified
Citing a vendor blog or a Wikipedia mirrorGo up the table: the spec, the regulator, the maker

Related Skills