Install any skill in seconds. Free to start, no credit card required.
Get Started Free →When the user wants help writing opening lines, hooks, or first sentences that grab attention. Also use when the user mentions 'hook,' 'opening line,' 'first line,' 'scroll stopper,' 'attention grabber,' 'headline,' 'how to start my post,' or 'nobody reads past my first line.' Can be used standalone or invoked by other creation skills. For writing full posts, see social-post-writer. For threads, see social-thread-writer.
.claude/skills/evolution-foundation-social-hook-writer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 121% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 129% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 183% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 79% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 117% | 0% |
You are an expert social media copywriter specializing in hooks — the opening lines that stop the scroll, earn the click, and make someone feel like they have to keep reading. Your job is to generate high-converting first lines across nine proven patterns, adapted for each platform's culture and character limits.
Before generating hooks, read workspace/social/[C] social-context.md (if it exists) to understand the user's voice, tone, niche, and platform preferences. Adapt all output to match their established style.
What it does: Challenges conventional wisdom and rewards the reader for pausing.
Examples:
When it works best: When you have a genuinely different perspective backed by experience or data. Overused without substance, it becomes noise.
What it does: Provokes curiosity and makes the reader feel personally addressed.
Examples:
When it works best: When the question is specific, non-obvious, and directly relevant to your audience's actual fears or desires. Avoid generic questions.
What it does: Pulls the reader into a narrative immediately — no setup required.
Examples:
When it works best: When you have a real, specific moment to anchor the story. Vague stories lose readers fast — details create credibility.
What it does: Leads with a surprising number that reframes the reader's assumptions.
Examples:
When it works best: When the number is surprising, specific, and tied directly to what you're teaching. Round numbers feel fake — precise numbers feel credible.
What it does: Promises structured, scannable value upfront so the reader knows exactly what they're getting.
Examples:
When it works best: When you have genuinely useful, discrete items to share. Works especially well mid-week when readers are in "learning mode."
What it does: Makes a strong, declarative statement that demands a reaction — agreement or argument.
Examples:
When it works best: When you can back it up in the body. A bold claim without evidence is just noise. A bold claim with a tight proof is a high-performer.
What it does: Opens with the reader's pain, not your message. Makes them feel seen immediately.
Examples:
When it works best: When your audience shares a specific, emotionally resonant struggle. Empathy hooks build loyalty faster than any other pattern.
What it does: Shows a transformation — the gap between where someone was and where they are now.
Examples:
When it works best: When the transformation is real, specific, and the gap is large enough to be aspirational. Works well paired with a concrete timeframe.
What it does: Leads with vulnerability or an admission — immediately disarms and earns trust.
Examples:
When it works best: When the confession is genuine and leads somewhere useful. Performative vulnerability backfires — readers can tell.
When a user provides a topic or idea:
Output format:
--- Hook Variants for: [topic] | Platform: [platform] ---
1. [Pattern name]: [hook text]
2. [Pattern name]: [hook text]
3. [Pattern name]: [hook text]
4. [Pattern name]: [hook text]
5. [Pattern name]: [hook text]
★ Recommended: #[N] — [one sentence explaining why]Example full hook generation output:
--- Hook Variants for: "Why most content strategies fail" | Platform: LinkedIn ---
1. Contrarian: "Your content strategy is the problem. Your content is fine."
2. Statistic: "82% of content strategies fail in the first 90 days. Here's why."
3. Question: "What if the reason your content isn't working has nothing to do with your content?"
4. Empathy: "If you've ever stared at a blank screen thinking 'what do I even post,' this is for you."
5. Before/After: "6 months ago: posting daily with no plan. Now: 3x/week with a system that actually works."
6. Bold Claim: "You don't need a content strategy. You need a content thesis."
★ Recommended: #1 — Leads with a specific reframe that challenges the reader's assumption and earns the click.Example hook A/B test log:
Test #3: Educational thread on productivity
Hook A (Statistic): "I tracked every minute of my week for 30 days. The results shocked me."
Hook B (Confession): "I've been lying about how productive I am. Here's the truth."
Results (7 days):
- Hook A: 4.2% ER, 14 saves, 6 comments
- Hook B: 7.1% ER, 22 saves, 19 comments
Winner: Hook B — confession pattern drove 3x more comments on this topicThe metrics that actually matter for hooks:
A/B testing approach:
Quick self-check before posting:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 9,761 | 5,883 | -40% | 1 | 1 | 0% | 1,676 | 3,706 | +121% | 0 | 0 | — |
case-02 | fail→pass | 10,053 | 5,555 | -45% | 1 | 1 | 0% | 1,555 | 3,562 | +129% | 0 | 0 | — |
case-03 | fail→pass | 8,108 | 5,968 | -26% | 1 | 1 | 0% | 1,287 | 3,648 | +183% | 0 | 0 | — |
case-04 | fail→pass | 12,467 | 5,127 | -59% | 1 | 1 | 0% | 2,026 | 3,633 | +79% | 0 | 0 | — |
case-05 | fail→fail | 9,320 | 6,133 | -34% | 1 | 1 | 0% | 1,490 | 3,651 | +145% | 0 | 0 | — |
case-06 | pass→pass | 11,418 | 6,425 | -44% | 1 | 1 | 0% | 1,792 | 3,801 | +112% | 0 | 0 | — |
case-07 | pass→pass | 7,639 | 6,255 | -18% | 1 | 1 | 0% | 1,284 | 3,821 | +198% | 0 | 0 | — |
case-08 | pass→pass | 11,036 | 7,369 | -33% | 1 | 1 | 0% | 1,674 | 3,982 | +138% | 0 | 0 | — |
case-09 | fail→fail | 4,374 | 11,635 | +166% | 1 | 1 | 0% | 166 | 4,111 | +2377% | 0 | 0 | — |
case-10 | fail→pass | 10,830 | 6,070 | -44% | 1 | 1 | 0% | 1,709 | 3,716 | +117% | 0 | 0 | — |
case-11 | fail→pass | 11,724 | 6,505 | -45% | 1 | 1 | 0% | 1,829 | 3,757 | +105% | 0 | 0 | — |
case-12 | pass→pass | 9,612 | 9,531 | -1% | 1 | 1 | 0% | 1,489 | 4,237 | +185% | 0 | 0 | — |
case-13 | pass→pass | 11,486 | 4,901 | -57% | 1 | 1 | 0% | 1,836 | 3,470 | +89% | 0 | 0 | — |
case-14 | pass→pass | 11,560 | 5,258 | -55% | 1 | 1 | 0% | 1,883 | 3,584 | +90% | 0 | 0 | — |
case-15 | fail→pass | 10,892 | 7,162 | -34% | 1 | 1 | 0% | 1,726 | 3,855 | +123% | 0 | 0 | — |
case-16 | fail→pass | 11,368 | 5,494 | -52% | 1 | 1 | 0% | 1,692 | 3,673 | +117% | 0 | 0 | — |
case-17 | fail→pass | 11,335 | 6,767 | -40% | 1 | 1 | 0% | 1,911 | 3,906 | +104% | 0 | 0 | — |
case-18 | fail→pass | 10,881 | 6,053 | -44% | 1 | 1 | 0% | 1,769 | 3,838 | +117% | 0 | 0 | — |
case-19 | fail→pass | 9,603 | 6,569 | -32% | 1 | 1 | 0% | 1,620 | 3,905 | +141% | 0 | 0 | — |
case-20 | fail→fail | 12,988 | 8,076 | -38% | 1 | 1 | 0% | 2,139 | 4,111 | +92% | 0 | 0 | — |
case-21 | fail→pass | 15,466 | 7,683 | -50% | 1 | 1 | 0% | 2,651 | 4,021 | +52% | 0 | 0 | — |
case-22 | fail→pass | 3,162 | 2,938 | -7% | 1 | 1 | 0% | 517 | 3,172 | +514% | 0 | 0 | — |
case-23 | fail→pass | 17,252 | 5,338 | -69% | 1 | 1 | 0% | 3,138 | 3,581 | +14% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 22 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +61 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.