Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the user wants organic social captions for Instagram, LinkedIn, TikTok, Facebook, or X — hook, mobile-readable body, A/B variants, hashtags, written to brand voice and funnel stage. Trigger on 'write a caption', 'Instagram caption', 'LinkedIn post', 'TikTok caption', 'social post copy', or 'organic post'. For paid ad copy use 05-ad-copy-global instead.
.claude/skills/minhnv0807-37-social-caption-global/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 104% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 100% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 95% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 225% | 0% |
Organic captions for owned social posts on Instagram, LinkedIn, TikTok, Facebook, and X. A good caption has a hook strong enough to stop the thumb, a body that earns the scroll, and a CTA clear enough that the reader knows what to do. Written for someone reading on a phone, moving fast.
Not this skill: paid ad copy — different constraints (ad policy review, character limits in placements, primary text truncation, iteration against CPA). Use 05-ad-copy-global.
Read the Brand Voice Document (35-brand-voice-global) and the content brief (36-content-brief-global) first. If the brief already sets angle, hook direction, and CTA, use them — do not re-ask. If both are missing, ask up to 4 questions:
35-brand-voice-global. Default banned list: "game-changer", "revolutionary", "unlock your potential", "don't miss out", "act now", "elevate your", "we're excited to announce", "best-in-class".From 36-content-brief-global: angle, source insight, key message, CTA. From 35-brand-voice-global: tone, vocabulary, sentence rules. Without a brief, gather the 4 answers above first.
The hook must stand alone with no context.
| Stage | Hook direction | Typical form | |-------|----------------|--------------| | TOFU | Pain / curiosity / contrarian | A question that names the frustration, or a statement that inverts the category's default belief | | MOFU | Social proof / concrete result | A number, a short testimonial, a before-and-after | | BOFU | Offer / genuine urgency | A specific discount or bonus, a real deadline, a real capacity limit |
The two variants must differ structurally, not by a few swapped words:
Each variant: Hook (opening line) then Body (2-4 short paragraphs, 1-3 sentences each) then CTA (a specific action).
State plainly: use A when context or audience], use B when context or audience], and how to mix (hook A with body B, or the reverse) if that is sensible.
Hashtags are a discovery and categorization signal now, not a reach lever. Fewer and more targeted outperforms volume on every major platform.
| Channel | Hashtags | Adaptation | |---------|----------|------------| | Instagram | 3-5, targeted to the actual topic | The first line carries the hook — it is what shows before "more". Visual leads; the caption supports it | | LinkedIn | 3, professional and topical | First 2-3 lines are visible before "see more". First-person point of view, no hard sell, external links often better in the first comment | | TikTok | 3-5, mixing one trending tag with niche tags | 1-2 lines, keyword-forward so it surfaces in TikTok search. The real hook lives in the first second of video; the caption supports it | | Facebook | 0-2, or none | Longer captions work. Conversational. Line breaks matter more than hashtags | | X | 1-2 maximum | One idea, no wind-up. If it needs more, it is a thread — write it as one |
Truncation points and character limits change; check the platform preview before publishing rather than trusting a remembered number.
File name: social-caption-[topic]-[YYYYMMDD].md
markdown# Social Caption — [Topic] ## Context | Item | Detail | |------|--------| | Funnel stage | TOFU / MOFU / BOFU | | Channel | | | Objective | | | Target length | | | Source brief | [link to content brief if any] | | Disclosure required | none / paid partnership / gifted / affiliate | ## Variant A — [hook direction] **Hook:** [Opening line — stands alone, creates pain / curiosity / surprise] **Body:** [2-4 short paragraphs, 1-3 sentences each, white space between] **CTA:** [Specific action] ## Variant B — [different hook direction] **Hook:** [Different angle — if A used pain, B uses curiosity or social proof] **Body:** [Different structure from A — if A used a list, B uses narrative] **CTA:** [May match or differ from A] ## Selection notes - Use **A** when: [context / audience] - Use **B** when: [context / audience] - Mixing A+B: [hook A with body B or the reverse, if sensible] ## Hashtags [Per platform norm — Instagram 3-5 / LinkedIn 3 / TikTok 3-5 / Facebook 0-2 / X 1-2] ## Brand voice check - Matches Brand Voice Document: Yes / No - Banned words: none present - Read aloud: sounds natural - Every claim substantiable: Yes
35-brand-voice-global: read before writing (tone, vocabulary, banned words); run its section VII checklist before handoff.36-content-brief-global: supplies angle, source insight, key message, and CTA.01-content-calendar-global: captions are written against a calendar slot (channel, funnel stage, pillar).05-ad-copy-global: when the post will run as a paid ad, rewrite to ad-copy standards. Never push an organic caption into an ad unchanged.42-image-brief-global, 43-carousel-brief-global: the accompanying visual brief, produced after the caption is approved.35-brand-voice-global).| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→pass | 15,181 | 14,187 | -7% | 1 | 1 | 0% | 1,717 | 3,502 | +104% | 0 | 0 | — |
case-04 | pass→pass | 23,084 | 24,905 | +8% | 1 | 1 | 0% | 2,682 | 6,052 | +126% | 0 | 0 | — |
case-05 | pass→fail | 25,530 | 38,786 | +52% | 1 | 1 | 0% | 3,421 | 7,703 | +125% | 0 | 0 | — |
case-06 | pass→fail | 33,576 | 30,281 | -10% | 1 | 1 | 0% | 4,424 | 5,874 | +33% | 0 | 0 | — |
case-07 | fail→pass | 17,893 | 16,867 | -6% | 1 | 1 | 0% | 1,884 | 3,760 | +100% | 0 | 0 | — |
case-01 | fail→pass | 16,818 | 17,285 | +3% | 1 | 1 | 0% | 2,818 | 4,020 | +43% | 0 | 0 | — |
case-02 | fail→pass | 12,773 | 11,917 | -7% | 1 | 1 | 0% | 1,976 | 3,851 | +95% | 0 | 0 | — |
case-08 | fail→pass | 12,061 | 11,806 | -2% | 1 | 1 | 0% | 1,176 | 3,821 | +225% | 0 | 0 | — |
case-09 | pass→pass | 13,073 | 14,668 | +12% | 1 | 1 | 0% | 1,972 | 4,142 | +110% | 0 | 0 | — |
case-10 | fail→fail | 11,629 | 15,312 | +32% | 1 | 1 | 0% | 1,077 | 3,569 | +231% | 0 | 0 | — |
case-11 | fail→pass | 17,655 | 17,266 | -2% | 1 | 1 | 0% | 1,870 | 3,827 | +105% | 0 | 0 | — |
case-12 | fail→fail | 10,996 | 21,803 | +98% | 1 | 1 | 0% | 1,008 | 3,889 | +286% | 0 | 0 | — |
case-19 | fail→fail | 13,206 | 20,717 | +57% | 1 | 1 | 0% | 1,275 | 4,306 | +238% | 0 | 0 | — |
case-13 | fail→pass | 12,908 | 19,989 | +55% | 1 | 1 | 0% | 1,253 | 4,355 | +248% | 0 | 0 | — |
case-14 | fail→pass | 16,790 | 15,775 | -6% | 1 | 1 | 0% | 1,722 | 3,648 | +112% | 0 | 0 | — |
case-15 | fail→pass | 12,304 | 15,526 | +26% | 1 | 1 | 0% | 2,139 | 4,301 | +101% | 0 | 0 | — |
case-16 | fail→pass | 15,047 | 16,927 | +12% | 1 | 1 | 0% | 2,563 | 4,746 | +85% | 0 | 0 | — |
case-17 | fail→pass | 13,257 | 11,644 | -12% | 1 | 1 | 0% | 1,321 | 2,832 | +114% | 0 | 0 | — |
case-18 | fail→fail | 8,268 | 21,137 | +156% | 1 | 1 | 0% | 1,423 | 4,842 | +240% | 0 | 0 | — |
case-20 | fail→pass | 11,450 | 16,181 | +41% | 1 | 1 | 0% | 1,382 | 3,674 | +166% | 0 | 0 | — |
case-21 | fail→pass | 8,463 | 10,734 | +27% | 1 | 1 | 0% | 1,322 | 3,758 | +184% | 0 | 0 | — |
case-22 | pass→pass | 15,981 | 16,323 | +2% | 1 | 1 | 0% | 1,676 | 3,748 | +124% | 0 | 0 | — |
case-23 | fail→pass | 14,307 | 16,181 | +13% | 1 | 1 | 0% | 1,688 | 3,703 | +119% | 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. The headline lift of +52 percentage points is the difference between those two pass rates over the 23 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
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.