This is a Claude skill, not a prompt pack. It bundles Python scripts that build the prompts for you, inject LinkedIn's SEO rules, and remember what performed — so output quality compounds instead of resetting every session.
Seven coordinated commands cover the full content workflow: single posts, carousels, newsletters, 30-day calendars, and a reinforcement-learning memory system. Requires Python 3.8+. No API keys, no external services, no network access.
Start with the setup section — the skill defaults to an "AI & Technology" niche, and it will keep generating content for that audience until you change it.
What the skill does and how it works
Source: github.com/sickn33/antigravity-awesome-skills — skills/linkedin-content-generator
Original author: sarveshkishortalele (github.com/sarveshtalele/linkedin-content-skill). License: MIT.
Install
# Via antigravity CLI (recommended)
antigravity install linkedin-content-generator
# Or manual
cp -r skills/linkedin-content-generator ~/.claude/skills/
The seven commands
A full LinkedIn content-creation suite that turns a topic and niche into publish-ready posts, multi-slide carousels, long-form newsletter editions, and 30-day content calendars.
What makes it different from a prompt: it bundles Python scripts that build richly engineered prompts, inject your saved preferences, and enforce LinkedIn SEO rules before Claude generates output. A local memory.md file persists your style, tone, successful hooks, and top-performing formats across every session.
| Command | Purpose |
|---|---|
/generate-post |
Single ready-to-publish LinkedIn post |
/generate-carousel |
Numbered slide content + caption |
/generate-newsletter |
Long-form newsletter edition |
/generate-calendar |
30-day posting calendar with Markdown table |
/show-memory |
Display current preferences and feedback log |
/feedback |
Save what worked for future outputs |
/clear-memory |
Reset memory to factory defaults |
Architecture: how memory and SEO injection work
User command (/generate-post ...)
│
▼
SKILL.md parses $ARGUMENTS
│
▼
Python script builds prompt
• Injects LinkedIn SEO rules
• Injects memory.md preferences
│
▼
Claude generates publish-ready output
│
▼
/feedback saves what worked → memory.md
(every future output improves)
The reinforcement learning loop: every time you run /feedback, the pattern is appended to memory.md and automatically injected into all future generation prompts. After a few weeks of use, the skill knows your best-performing hook types, which content formats your ICP responds to, and what tone gets the most comments in your niche.
One-time setup: set your niche
Open ~/.claude/skills/linkedin-content-generator/scripts/memory.md and update the Primary Niche field:
## Core Identity & Tone
- **Primary Niche:** AI & Technology ← change this to your niche
This field is injected into every prompt. Without it, the skill defaults to "AI & Technology".
For B2B GTM and SaaS founders: set your niche to match your ICP's world, not your own role. If you help SaaS founders with outbound, your niche is "B2B SaaS Sales", not "GTM Engineering". Your content should live in your buyer's feed, not your own professional category.
Command 1: /generate-post
Single LinkedIn posts with SEO-optimised hooks, niche injection, and 5 tone options. Output is publish-ready: hook, body, takeaway, CTA, and hashtags in one pass.
Command and parameters
/generate-post <topic> [in <niche>] [tone: controversial|storytelling|educational|motivational|professional]
| Parameter | Default | Options |
|---|---|---|
| topic | required | any subject |
| niche | from memory.md |
any industry |
| tone | professional | professional, storytelling, controversial, educational, motivational |
| style | list-based | list-based, text-only, storytelling, data-driven, contrarian |
Output structure
- Scroll-stopping hook (2 lines, written to trigger the See More click)
- Context / problem setup (2–3 short sentences)
- Core value (numbered list or bullets, max 7 items)
- Key takeaway (1–2 punchy sentences)
- Specific call to action
- 3–5 hashtags (broad + niche + community mix)
Copy-paste examples
# Storytelling tone for a GTM audience
/generate-post why most founders fail at outbound in B2B SaaS tone: storytelling
# Controversial take on a common belief
/generate-post the real cost of technical debt in SaaS tone: controversial
# Educational list post
/generate-post 5 things I wish I knew before my first outbound campaign in Outbound Sales tone: educational style: list-based
# Contrarian take with data
/generate-post cold email is not dead in B2B Sales tone: controversial style: data-driven
# Motivational for founders
/generate-post why slow pipeline growth is not a leads problem in Entrepreneurship tone: motivational
GTM-specific post prompts
# Proof content (proof posts convert best for pipeline generation)
/generate-post how we booked 43 calls in one month from LinkedIn in B2B Sales tone: storytelling style: data-driven
# Framework content (builds authority, ICP associates your name with expertise)
/generate-post the 3-layer LinkedIn system that replaced our SDR team in GTM tone: educational style: text-only
# Contrarian content (gets the most DMs and replies)
/generate-post why hiring more SDRs made our pipeline worse in B2B SaaS tone: controversial style: contrarian
# Signal-based outreach insight
/generate-post how we use buying signals instead of job titles for outbound in Revenue Operations tone: educational style: list-based
Command 2: /generate-carousel
Numbered slide content plus a ready-to-use LinkedIn caption. 5 carousel styles, 3–12 slides. Carousel posts get 2.5–3.5x the reach of text-only posts on LinkedIn.
Command and parameters
/generate-carousel <topic> [in <niche>] [<n> slides] [style: how-to|listicle|myth-busting|framework|story-arc]
| Parameter | Default | Options |
|---|---|---|
| topic | required | any subject |
| niche | from memory.md |
any industry |
| slides | 7 | 3–12 |
| style | listicle | how-to, listicle, myth-busting, framework, story-arc |
Style guide
| Style | Structure |
|---|---|
| how-to | Slide 1 = problem, slides 2–N = steps, last = result/CTA |
| listicle | Each slide = one item with bold title + 1–2 sentence explanation |
| myth-busting | Each slide = MYTH: [belief] then TRUTH: [reality] |
| framework | Introduce a proprietary framework; each slide = one component |
| story-arc | Slide 1 = before, middle = journey, last = after + CTA |
Output: slides numbered 1 through N, followed by a LinkedIn caption with hook, teaser context, "Swipe →" prompt, and hashtags.
Copy-paste examples
# Myth-busting (high saves, high shares)
/generate-carousel 10 LinkedIn outreach mistakes 8 slides style: myth-busting
# How-to for GTM audience
/generate-carousel how to build a LinkedIn content system in B2B Sales 7 slides style: how-to
# Framework carousel (best for authority building)
/generate-carousel the 3-layer LinkedIn GTM system in Revenue Operations style: framework
# Story-arc (most shareable)
/generate-carousel how I went from 0 to 40 calls per month on LinkedIn in B2B SaaS 6 slides style: story-arc
# Listicle (easiest to consume, good for reach)
/generate-carousel 7 LinkedIn signal plays that book meetings in Outbound Sales 7 slides style: listicle
# Myth-busting for ICP-specific content
/generate-carousel 5 myths about AI outbound in B2B Technology 5 slides style: myth-busting
Command 3: /generate-newsletter
Complete long-form newsletter edition structured for the LinkedIn Newsletter editor. Short (~700 words), medium (~1,200 words), or long (~2,000 words).
Command and parameters
/generate-newsletter <topic> [in <niche>] [length: short|medium|long] [title: "<series title>"]
| Parameter | Default | Options |
|---|---|---|
| topic | required | any subject |
| niche | from memory.md |
any industry |
| length | medium | short (~700w), medium (~1,200w), long (~2,000w) |
| title | auto-generated | optional series name |
Output structure
- SEO-optimised H1 headline
- Opening hook (personal anecdote, statistic, or bold claim)
- Body sections with H2 subheadings
- Key takeaways (3–5 bullets)
- One specific action step for this week
- Engagement question to spark comments
Copy-paste examples
# Standard medium edition
/generate-newsletter how AI is changing B2B outbound in Sales Technology length: medium
# Long-form thought leadership with series name
/generate-newsletter the state of LinkedIn GTM in 2026 in B2B Sales length: long title: "Pipeline Weekly"
# Short punchy edition for busy readers
/generate-newsletter 3 signal plays that booked 20 calls this month in Revenue Operations length: short
# Deep-dive with series name
/generate-newsletter why cold email reply rates are falling and what to do about it in Outbound Sales length: long title: "The Outbound Brief"
Command 4: /generate-calendar
30-day Markdown table calendar with monthly theme, SEO keywords, format breakdown, and 5 goal modes. Plan a month of content in one command instead of writing each post topic manually.
Command and parameters
/generate-calendar [niche: <niche>] [days: <n>] [frequency: <freq>] [goal: awareness|engagement|leads|authority|growth]
| Parameter | Default | Options |
|---|---|---|
| niche | required | any industry |
| days | 30 | any positive integer |
| frequency | 3 times a week | any posting cadence |
| goal | growth | awareness, engagement, leads, authority, growth |
Goal guide
| Goal | Strategy |
|---|---|
| awareness | Shareable, relatable, trending. Heavy on carousels and contrarian takes. |
| engagement | Opinion posts, polls, questions, storytelling to maximise comments. |
| leads | Educational value posts + authority-building + clear DM CTAs. |
| authority | Deep insights, data-backed posts, thought leadership, newsletters. |
| growth | Mix viral formats (carousels, lists, contrarian) with high-value education. |
Output: a Markdown table with columns # | Day | Format | Topic/Angle | Hook | CTA, plus monthly theme, top 5 SEO keywords, and a format breakdown summary.
Copy-paste examples
# Full month for a GTM audience, pipeline-generation goal
/generate-calendar niche: B2B Sales days: 30 frequency: 3 times a week goal: leads
# Authority-building calendar for SaaS founders
/generate-calendar niche: B2B SaaS days: 30 frequency: daily goal: authority
# 2-week sprint for outbound content
/generate-calendar niche: Outbound Sales days: 14 frequency: 5 times a week goal: engagement
# Revenue Operations thought leadership
/generate-calendar niche: Revenue Operations days: 30 frequency: 3 times a week goal: authority
# GTM engineering audience awareness campaign
/generate-calendar niche: GTM Engineering days: 30 frequency: 4 times a week goal: awareness
Commands 5–7: the memory system
The reinforcement-learning memory system is the most important feature of this skill. Every piece of feedback improves all future outputs automatically.
How memory works
memory.md lives at ~/.claude/skills/linkedin-content-generator/scripts/memory.md.
What it stores:
- Your primary niche (injected into every prompt)
- Your preferred tone and style
- Every
/feedbackentry you have saved - Successful hook types and formats
How it gets used: the Python scripts read memory.md before building any prompt and inject its contents. Claude generates output that already incorporates your historical preferences without you having to re-specify them.
After consistent use, the skill knows your best-performing hook types, which formats your audience responds to most, what tone drives the most comments in your niche, and which structures to avoid.
/feedback — saving what works
/feedback <what worked well>
Examples:
# Hook performance
/feedback the contrarian hook "everyone is wrong about X" drove 400% more impressions
# Format performance by niche
/feedback myth-busting carousels in the B2B Sales niche get 3x more saves than listicles
# Tone performance
/feedback storytelling posts about personal failures get 5x more comments than educational posts
# CTA performance
/feedback asking "what would you add?" gets more comments than "drop a comment below"
# Negative feedback (what to avoid)
/feedback data-heavy posts with 5+ statistics underperform against single-stat proof posts
# GTM-specific learning
/feedback posts about LinkedIn signal plays perform 2x better than posts about cold email
# Hook learning
/feedback posts starting with a specific number ("43 calls in 30 days") outperform posts starting with questions
/show-memory — review your preferences
/show-memory
Displays full memory.md contents: entry count, primary niche, tone summary, and all saved feedback entries. Run this before a content session to review what the skill knows about your preferences.
/clear-memory — reset to defaults
/clear-memory
Resets memory.md to factory defaults. Does not delete the niche setting. Use when feedback has accumulated conflicting preferences, or when you're switching to a different content persona or audience.
Quick reference: all commands and GTM prompts
Install
# Via antigravity CLI
antigravity install linkedin-content-generator
# Manual
cp -r skills/linkedin-content-generator ~/.claude/skills/
# Verify installed
ls ~/.claude/skills/linkedin-content-generator/
# Should show: SKILL.md, scripts/
First session setup
# Step 1: Set your niche in memory.md
# Open ~/.claude/skills/linkedin-content-generator/scripts/memory.md
# Change: Primary Niche: AI & Technology
# To: Primary Niche: [YOUR NICHE]
# Step 2: Check memory is loaded correctly
/show-memory
# Step 3: Generate your first post
/generate-post [YOUR FIRST TOPIC] tone: storytelling
# Step 4: Save what works
/feedback [WHAT PERFORMED WELL]
Weekly content workflow
# Monday: plan the week
/generate-calendar niche: [YOUR NICHE] days: 7 frequency: 3 times a week goal: [YOUR GOAL]
# Tuesday: draft the week's posts
/generate-post [TOPIC 1] tone: [TONE]
/generate-post [TOPIC 2] tone: [TONE]
/generate-post [TOPIC 3] tone: [TONE]
# Mid-week: generate carousel if needed
/generate-carousel [TOPIC] [N] slides style: [STYLE]
# After posts go live: save what performed
/feedback [WHAT WORKED]
/feedback [WHAT UNDERPERFORMED - to avoid in future]
# Monthly: generate next month's calendar
/generate-calendar niche: [YOUR NICHE] days: 30 frequency: [CADENCE] goal: [GOAL]
GTM content type quick reference
# Proof post (highest pipeline conversion)
/generate-post [SPECIFIC CLIENT RESULT WITH NUMBER] in B2B Sales tone: storytelling style: data-driven
# Framework post (best for authority)
/generate-post [YOUR METHODOLOGY OR PROCESS] in [YOUR NICHE] tone: educational style: list-based
# Contrarian post (most DMs and replies)
/generate-post [COMMON BELIEF YOU DISAGREE WITH] in [YOUR NICHE] tone: controversial style: contrarian
# Carousel for lead magnet
/generate-carousel [TOPIC YOUR ICP CARES ABOUT] 7 slides style: how-to
# Myth-busting carousel (highest saves)
/generate-carousel [MYTHS IN YOUR SPACE] 6 slides style: myth-busting
# Newsletter for long-term warming
/generate-newsletter [DEEP-DIVE TOPIC YOUR ICP SEARCHES FOR] length: long title: "[YOUR SERIES]"
# Pipeline-focused calendar
/generate-calendar niche: [YOUR ICP'S NICHE] days: 30 frequency: 3 times a week goal: leads
LinkedIn SEO rules the skill enforces
The Python scripts inject these LinkedIn platform rules into every prompt before Claude generates:
- The hook occupies lines 1–2 and must earn the See More click — the most important element
- Hashtag strategy: 3–5 per post, mixing broad, niche, and community tags
- No external links in the post body (LinkedIn suppresses reach for posts with outbound links)
- Paragraph breaks every 1–3 lines (long blocks suppress engagement)
- Carousel posts include a "Swipe →" prompt in the caption
- Newsletter headlines are H1-structured for SEO
- CTAs are action-specific — not "let me know what you think" but specific questions or resource offers
- Format variety is enforced in calendar outputs, to avoid algorithmic penalties from posting the same format repeatedly
