Index of every page on this site: llms.txt
All skills
sergebulaev avatar

/tt-hook-scripter

@ba2f700
Run it yourself

Script the first 1-3 second TikTok hook: the spoken line, the on-screen text, and the opening visual firing at once so the viewer cannot swipe. Picks a 2026 hook formula (cold-open result, pattern interrupt, number reveal, open-loop question, relatable call-out, story) by goal of completion, saves, comments, or shares. The single biggest retention lever on TikTok. Use to script a hook from a video idea. Not for the caption (use tt-caption-writer) or scrubbing a script (use tt-humanizer).

Use this Skill: https://skilld.dev/gh/sergebulaev/tiktok-skills/tt-hook-scripter

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈128 tokens for metadata: the name and description. ≈1.7k when used: this file. ≈821 more on demand in 1 file.

Description uses 6.4% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈511 characters, or 6.4%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

TikTok Hook Scripter

Write the first 1 to 3 seconds of a TikTok video: the spoken line, the on-screen text, and the opening visual, all firing at once. This is the single biggest lever on the whole platform. TikTok's ranker runs on completion rate, and a viewer decides to stay or swipe before second three. If the hook does not win, nothing downstream gets watched.

When to use

  • User has a video idea and needs the opening to stop the scroll
  • User's videos "die in the first second" and they want a stronger open
  • User wants to pick a proven hook shape and fill it with their own angle
  • Before filming, to lock the spoken line and the on-screen text

What this skill produces

For one video idea, a ready-to-film hook block:

  • Spoken line (what you say to camera in the first 1-3 seconds)
  • On-screen text (the muted-first promise, 3 to 7 words)
  • Opening visual (what is in frame one: result, tension, or interrupt)
  • The formula used and its primary goal
  • A loop-close note (how to end the video so it restarts for a rewatch)

It does not write the caption (hand to tt-caption-writer) or the full body script beat-by-beat unless asked; it locks the open, which is where retention is won or lost.

Formulas this skill uses

Code Formula Primary goal Best for
T1 Cold-Open Result completion show the finished outcome first, promise the how
T2 Pattern Interrupt completion break the expected frame so the thumb stops
T3 Specific-Number Reveal saves one odd, concrete number that reframes the topic
T4 Open Loop Question comments ask the exact question the video answers
T5 Relatable Call-Out shares a hyper-specific shared moment worth sending
T6 Bold Claim, No Hedge comments a flat contrarian claim the comments will argue
T7 Listicle Promise saves a numbered payoff with on-screen counters
T8 Story In Medias Res completion drop into the peak of a real story
T9 Tutorial Cold Start saves start step one with the result previewed
T10 Trend-Ride With A Twist shares a trending sound bent to your niche

Full skeletons (three layers each) in ../../references/hook-formulas.md.

Pick by goal first

Goal Reach for
Completion T1, T2, T8
Saves T3, T7, T9
Comments T4, T6
Shares T5, T10

Steps

Voice profile first (all drafts). If ../../references/voice-profile.md has filled: yes, load it and match the user's voice fingerprint, hard rules, and CTA/link style throughout. If it is not filled, mention once that tt-humanizer --mode profile can learn their voice from a few posts, then proceed with the generic voice rules.

Never invent the specifics. The rules below ask for a concrete number, a date and a named entity, because that is what separates a real post from a generated one. Take them from what the user actually said in this conversation. Do not invent a figure, a date, a client name or a result, and do not soften a vague claim into a plausible-looking number. If the user has nothing concrete for a beat, ask them once, and if they still have nothing, drop the claim rather than decorate it. A published invented number is a retraction; a missing one is only a weaker post.

  1. Gather inputs. The video idea, the niche/audience, the one promise the video delivers, and the goal (completion / saves / comments / shares).
  2. Pick the formula. Use the goal table to shortlist, then suggest 2-3 that also fit the idea and let the user choose. For a trend ride (T10), hand the sound-fit question to tt-trend-mapper first.
  3. Write all three layers. Spoken line, on-screen text, opening visual. The spoken line and the on-screen text MUST differ (muted-first viewing). Respect the 2026 rules:
    • The result or the tension is in frame one. No greeting, no logo, no zoom.
    • One specific number where the claim allows it.
    • Say it the way a person talks; read it out loud.
    • No AI vocabulary cluster (one marker is a slip; three in the hook is a script). Em dashes: .. for a breath in the spoken line, at most one on the on-screen card.
    • No sincerity opener ("not gonna lie", "real talk") and no "Here's what nobody tells you" reveal bridge; say the thing.
  4. Write the loop-close note. How the last frame should land so a rewatch feels natural (a rewatch is almost a second view).
  5. Humanizer pass. Run the spoken line through the tt-humanizer V3 rules to strip script tells before it ever reaches camera: AI vocab by density, stacked triads, reveal bridges, sincerity openers, written-not- spoken phrasing. Make it sayable; never insert a punch line for rhythm.
  6. Approval card. Show: formula, the three layers, primary goal, the loop-close note, and an estimated hook duration.
  7. Hand off. Offer tt-caption-writer for the caption and settings, and tt-content-planner if they are batching multiple hooks.

Hard rules

Global voice rules: see root SKILL.md Voice rules. Additional skill-specific rules:

  • The hook is the first 1-3 seconds of the video, not the caption. Never let the caption do the hook's job.
  • The spoken line and the on-screen text are two different jobs. Do not repeat the same words in both.
  • Frame one shows something (result, tension, interrupt). No dead air, no intro.
  • One specific number in the hook beats any adjective.
  • The hook opens the loop; it never answers itself.

Anti-patterns (skill will refuse)

  • "Hey guys" / "welcome back" / "in this video" openers.
  • A logo animation or slow zoom before the words start.
  • Reading the caption aloud as the hook.
  • Promising a result the video never shows on screen.
  • ALL CAPS spoken scripts (you cannot shout for 30 seconds).
  • More than one em dash, or an AI vocabulary cluster, in the on-screen text.
  • A staged punch line or "The result?" reveal inserted for rhythm.
  • Five stacked calls to action.

Resources

  • ../../references/hook-formulas.md - all 10 TikTok hook formulas (three layers each)
  • ../../references/algorithm-heuristics.md - why completion and the first second rule everything
  • references/hook-anatomy.md - the three-layer hook teardown and timing budget

Related skills

  • tt-caption-writer - the caption, TikTok settings, and hashtags
  • tt-trend-mapper - check sound fit before a trend-ride hook (T10)
  • tt-humanizer - scrub the spoken script so it sounds human on camera
  • tt-content-planner - batch multiple hooks for the week

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts1mo3 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    The skill is a specialized scriptwriter for TikTok video hooks. It uses established formulas to generate spoken lines, on-screen text, and visual descriptions based on user-provided video ideas. No security risks or malicious patterns were identified.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: LOW · No issues

Signed by skilld at ba2f700. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 2 months ago

README badge

README badge for sergebulaev/tiktok-skills/tt-hook-scripter