| name | hook-writer |
| description | Turn any piece of content (draft post, blog article, video script, podcast transcript, newsletter, case study, or raw idea) into scroll-stopping hooks for a specific platform. Outputs 5 hook options with 1 clear recommendation and the reasoning behind it. Use whenever the user asks for hooks, opening lines, first lines, titles, or intros, says "make this grab attention", "improve my hook", "what should the first line be", "help this stop the scroll", or wants openers for Twitter/X, LinkedIn, Instagram, TikTok, YouTube, or an email newsletter. Trigger even if they paste content and just ask "how do I open this?". NOT for paid ad campaign hooks generated from a business URL (that is the ads-hooks skill). |
Hook Writer
Turn a piece of content into hooks that earn the next 3 seconds of attention. The output is always 5 options plus 1 clear recommendation, formatted so a non-writer can use it immediately.
A hook is not a summary of the content. It is the single most interesting thing in the content, sharpened until a stranger who owes you nothing stops scrolling. Most weak hooks fail because they describe the content ("My thoughts on hiring") instead of opening a gap the reader needs closed ("I hired 3 people wrong before I noticed the pattern").
Process
Work through these five steps in order. Steps 1 and 2 are where hooks are actually won; do not skip to writing.
Step 1: Identify the job
Establish three things before writing anything:
- The content: what piece is this hook for? If the user gave only a topic or idea, work with that, but say the hooks assume the content will deliver on them.
- The platform: hooks do not transfer between platforms. A LinkedIn first line dies on TikTok. If the user named a platform, use it. If not, infer the most likely platform from the content format (short video script implies TikTok/Reels, professional insight implies LinkedIn, essay or article implies Twitter/X or a newsletter) and state the assumption in one line so they can redirect you. Only ask when the content genuinely fits multiple platforms equally and the choice changes everything.
- The audience: who has to stop scrolling? A hook for founders is different from a hook for brides-to-be. Pull this from the content itself or from what you know about the user's business.
Step 2: Mine the content for hook assets
Hooks are found, not invented. Before writing, scan the content and list its raw assets:
- The most surprising fact or counterintuitive claim
- The biggest specific number (money, time, percentage, count)
- The transformation (before X, now Y)
- The mistake or failure story
- The most relatable pain or frustration
- The thing an insider knows that outsiders don't
- The enemy (a common practice, belief, or tool the content argues against)
A hook built on a real asset from the content will always beat a generic template filled with vague words. If the content has no assets (no numbers, no story, no tension), tell the user that plainly and hook on the strongest available angle, but flag that adding one specific number or story to the content would make every hook stronger.
Diagnose the draft. If the user gave you an existing draft with its own opening, find the single most interesting sentence in it and check whether it is the first line. Very often the real hook is buried in the middle (a story, a number, a confession) while the draft opens with warm-up or a resume. When that happens, say so directly and point at the exact sentence that should have led. In testing, this diagnosis was frequently the most valuable part of the whole reply, so do not skip it to get to the list faster.
Step 3: Load the platform playbook
Read the reference file for the target platform before generating. Each file has the platform's mechanics (where the hook lives, truncation limits, timing), its proven formula families with examples, and its specific failure modes:
For multi-platform requests, produce a separate section per platform. Never reuse the same hook across platforms; adapt the underlying asset to each platform's mechanics.
Step 4: Generate wide, then select 5
Draft 10 to 15 candidates internally, pulling from at least 4 different formula families in the platform reference. Then select the 5 strongest. The final 5 must be genuinely different angles (for example: one contrarian, one number-led, one story-led, one fear/mistake, one curiosity gap), not five rewordings of the same idea. Variety is the point: the user or their client will have instincts about their audience that you don't, and a diverse slate lets them exercise that judgment.
Select using the universal checklist (in universal-principles.md). The short version:
- Specific beats vague. A number, a name, or a timeframe beats an adjective every time.
- The first 5 words carry the weight. Front-load the tension.
- The curiosity gap must be one the content actually closes. Bait-and-switch destroys trust and gets punished by every platform's retention metrics.
- Would a stranger care? Not a follower, a stranger.
- It must sound like a person. No em dashes, no "In today's world", no "game-changer", none of the AI tells listed in universal-principles.md.
Step 5: Deliver in the standard format
ALWAYS use this exact output structure:
## Hooks for [Platform]: [short content descriptor]
⭐ **Recommended**
> [the hook, formatted exactly as it would appear on the platform]
**Why this one:** [2-3 sentences. Name the psychological angle, the asset from
the content it uses, and why it fits this platform's mechanics.]
**Four more to test:**
2. > [hook]
[angle name] · [one line on why it could win]
3. > [hook]
[angle name] · [one line on why it could win]
4. > [hook]
[angle name] · [one line on why it could win]
5. > [hook]
[angle name] · [one line on why it could win]
**Delivery notes:** [1-3 platform-specific execution notes: where the hook goes,
truncation limit to respect, pairing with visual/text-on-screen, what to A/B test first.]
Rules for the output:
- The recommendation is a real commitment. Pick one and defend it; do not hedge with "any of these could work".
- Write each hook exactly as it would be published: correct casing for the platform, line breaks where they matter, on-screen text marked as such for video platforms.
- For video platforms (TikTok, Reels, YouTube), a hook is spoken + shown + framed. Give the verbal line, and when it materially helps, the matching on-screen text or visual note in brackets.
- Keep the whole deliverable scannable. The user may paste it directly to a client.
Calibration
- If the user's content is strong but their draft hook buries the lede, say so directly: point at the sentence in their content that should have been the hook. This is often the single most useful thing in the deliverable.
- If the user asks for hooks for "all platforms", push back gently toward their top 2 or 3 platforms, then deliver those in full quality rather than 7 platforms of filler.
- If the user gives feedback ("more edgy", "less clickbaity"), regenerate the full slate of 5, not just one replacement. Feedback shifts the whole calibration.
- When the user's niche is regulated (finance, health, legal), keep hooks bold but factual: curiosity and specificity, never implied guarantees.
What this skill is not
- Not an ad-copy generator: for paid campaign hooks built from a business URL and persona research, use the ads-hooks skill instead.
- Not a content writer: it writes the opener, not the piece. If the user wants the full post or script written, hooks come first, then continue with the body as a separate task.
