AI-powered LinkedIn content suite: generate posts, carousels, newsletters, and 30-day calendars with niche-specific SEO rules and a reinforcement-learning personal memory system.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
AI-powered LinkedIn content suite: generate posts, carousels, newsletters, and 30-day calendars with niche-specific SEO rules and a reinforcement-learning personal memory system.
A full LinkedIn content-creation suite for Claude Code that turns a topic and niche into
publish-ready posts, multi-slide carousels, long-form newsletter editions, and 30-day content
calendars — all wired through a personal reinforcement-learning memory system so every output
improves as you give feedback.
Seven coordinated commands cover the full content workflow:
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
All helper scripts are bundled inside skills/linkedin-content-generator/scripts/ and ship
alongside this SKILL.md. They 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.
When to Use This Skill
Use when you need a ready-to-paste LinkedIn post with SEO-optimised hooks and hashtags.
Use when building a multi-slide carousel deck for LinkedIn Documents.
Use when writing a long-form LinkedIn Newsletter edition with structured sections.
Use when planning an entire month of content with format variety and pacing rules.
Use when you want content that adapts to your personal voice over time via saved feedback.
Use when working in any niche (AI, SaaS, Marketing, Finance, Healthcare, etc.) and need
platform-native formatting that avoids common LinkedIn algorithmic pitfalls.
Prerequisites
Python 3.8 or later must be available in your shell path.
The skill is self-contained. Install it from the antigravity skills library:
# Or copy manually into your Claude Code skills directory
cp
All six Python scripts and the default memory.md are bundled inside the
scripts/ subdirectory of this skill. No additional cloning or downloads are required.
No API keys, external services, or network access are needed.
How It Works
Architecture
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
(loop — every future output improves)
Step 1: Set Up Your Niche (One-Time)
Open ~/.claude/skills/linkedin-content-generator/scripts/memory.md and update the
Primary Niche field:
## Core Identity & Tone-**Primary Niche:** AI & Technology ← change this
This field is injected into every prompt. Without it, the skill defaults to
"AI & Technology".
Step 2: Generate Content
Run any of the seven commands described in the Commands Reference section below.
Claude reads the script output and produces the final content directly in the chat.
Step 3: Save What Works
After each output, save successful patterns with /feedback:
/feedback the storytelling hook in this post got 3x more comments than usual
The feedback is appended to memory.md and automatically injected into all future
generation prompts.
Commands Reference
/generate-post — Single LinkedIn Post
Generates a scroll-stopping, SEO-optimised LinkedIn text post.
/generate-calendar niche: Marketing Agencies days: 14 frequency: 5 times a week goal: leads
Output: Markdown table (# | Day | Format | Topic / Angle | Hook | CTA) + monthly theme +
top 5 SEO keywords + format breakdown summary.
/show-memory — Display Preferences
Displays current memory contents: niche, tone, style, and all saved feedback entries.
Usage:
/show-memory
Output: Full memory.md content with entry count, primary niche, and tone summary.
/feedback — Save Successful Patterns
Appends a labelled feedback entry to memory.md. Future outputs automatically incorporate
saved patterns.
Usage:
/feedback <what worked well>
Examples:
/feedback the contrarian hook "everyone is wrong about X" drove 400% more impressions
/feedback myth-busting carousels in the DevOps niche get 3x more saves than listicles
/feedback storytelling tone with a personal failure story outperforms data-driven in my audience
/clear-memory — Reset Memory
Resets memory.md to factory defaults. The command asks for confirmation before
executing.
Usage:
/clear-memory
LinkedIn SEO Rules (Enforced Automatically)
The skill injects these rules into every prompt via the bundled scripts/utils.py. They are not
optional; they are part of the prompt engineering that makes outputs platform-native.
Hook Engineering
Line 1 must be scroll-stopping (bold claim, surprising stat, provocative question, or
personal story opener).
Line 2 must create a pattern interrupt that forces "see more".
Forbidden openers: "In today's...", "I am excited to...", "Happy to share...",
"Thrilled to announce...".
Readability Rules
Maximum 2 sentences per paragraph.
Aggressive line breaks — white space wins on LinkedIn.
1–2 community hashtags (#LinkedInTips, #PersonalBranding).
Hard limit: never more than 5 total.
Best Practices
✅ Set Primary Niche in memory.md before generating any content.
✅ Run /feedback after any post that performs well — the memory compounds over time.
✅ Use /generate-calendar first when planning a content sprint; it provides topics
for /generate-post and /generate-carousel runs.
✅ Mix carousel styles across a calendar period: listicle, myth-busting, and
story-arc perform differently and prevent audience fatigue.
✅ Test the controversial tone on topics where you have a genuine, defensible stance;
avoid it for topics where nuance is more valuable than edge.
❌ Do not skip the /feedback loop — without it, every output starts from generic
LinkedIn best practices rather than your specific audience data.
❌ Do not post more than 5 hashtags; LinkedIn's algorithm penalises hashtag stuffing.
❌ Do not use the data-driven style without real statistics to cite; fabricated
numbers destroy credibility faster than any other LinkedIn mistake.
❌ Do not generate a 30-day calendar without specifying goal; the default growth
goal mixes formats broadly and may not match a specific campaign objective.
Limitations
This skill does not publish to LinkedIn directly. All output is copy-paste ready but
requires manual posting via the LinkedIn web or mobile app.
The memory system is file-based and local. It is not shared across machines or team
members without manually syncing memory.md.
The skill does not verify real-time LinkedIn algorithm changes. SEO rules are based on
documented best practices as of mid-2025 and may need manual updates as the platform
evolves.
Calendar output does not auto-schedule posts or integrate with scheduling tools
(Buffer, Hootsuite, etc.). It produces a Markdown table for manual import.
The slides parameter is clamped to the range 3–12. Carousels outside this range
will silently be adjusted to the nearest boundary.
memory.md grows unbounded as feedback accumulates. Very large memory files (500+
entries) may exceed prompt context limits and cause truncation. Periodically archive
old entries using /clear-memory and re-seed with your top learnings.
Does not work in sandboxed enprojectnments where python3 is unavailable or Bash tool
calls are blocked.
Security & Safety Notes
This skill uses the Bash allowed-tool to run Python scripts bundled at
~/.claude/skills/linkedin-content-generator/scripts/. All scripts are read-only
operations except memory_manager.py, which writes only to memory.md inside that
same bundled scripts/ directory.
No network requests are made by any script.
No credentials, tokens, or secrets are read, written, or logged.
No files outside ~/.claude/skills/linkedin-content-generator/scripts/ are modified.
The clear command in memory_manager.py overwrites only the bundled memory.md;
it does not delete any other files.
All --feedback and --id arguments passed to memory_manager.py are written
verbatim to memory.md. Do not pass shell metacharacters or sensitive data as
feedback strings.
All Bash commands in this skill are local Python invocations with no elevated privileges
required:
Problem: Script exits with ModuleNotFoundError or No module named 'utils'.
Solution: Each script uses sys.path.insert(0, SCRIPT_DIR) to locate utils.py
relative to itself, so they must be invoked with an absolute path — not from inside
the scripts/ directory. Use
python3 "${HOME}/.claude/skills/linkedin-content-generator/scripts/generate_post.py" ....
Problem: Memory is not being applied to generated content.
Solution: Check that memory.md exists at
~/.claude/skills/linkedin-content-generator/scripts/memory.md. Run /show-memory
to confirm. If missing, run any generator command once — it auto-creates the file from
the bundled template.
Problem: Calendar output is missing days or the table is malformed.
Solution: Verify the --days value is a positive integer and --frequency is
quoted if it contains spaces (e.g., "3 times a week"). The script passes these
values directly into the prompt string.
Problem: Carousel slides exceed the requested count.
Solution: The slides value is clamped server-side to [3, 12]. If Claude
generates more slides than requested, it is following the style guide structure
(cover + content + CTA). Specify an exact count and style to get precise control.
Problem: Generated post sounds generic despite feedback being saved.
Solution: Memory entries are injected as context, not as hard rules. Use specific,
actionable feedback: "opening with a personal failure story outperforms stats for my audience" is more useful than "storytelling was good".
Problem:python3 not found on Windows.
Solution: Install Python 3.8+ from python.org and ensure it is on PATH, or run via
py "%USERPROFILE%\.claude\skills\linkedin-content-generator\scripts\generate_post.py" .... On Windows without WSL, the Bash tool invocation
may need adjustment in the SKILL.md allowed-tools context.
Related Skills
@content-creator — Broader brand voice analysis, SEO optimisation, and
cross-platform content frameworks. Use when building a full content marketing system
beyond LinkedIn alone.
@content-strategy — Topic cluster planning, editorial roadmap, and content mix
strategy. Use before running /generate-calendar when you need to define pillar topics
first.
@content-marketer — Campaign-level content planning across channels. Complements
this skill when LinkedIn is one channel in a broader multi-platform launch.
@linkedin-automation — Programmatic LinkedIn post publishing via the Composio/Rube
MCP. Use alongside this skill when you want to automate the publishing step after
generating content here.
@linkedin-profile-optimizer — LinkedIn profile and personal brand optimisation.
Use before running this skill to align generated content voice with your profile's
headline, summary, and featured section.
