LinkedIn Post Generator

Turns any input into 3 ready-to-publish LinkedIn post variants, each with a different hook style. Works standalone or chained with other intelligence skills.

What this skill does

1. Takes any input — topic, news story, lead insight, observation, raw notes
2. Generates 3 post variants using different hook styles
3. Scores each variant on hook strength (1–10)
4. Recommends the strongest variant and best posting time
5. Suggests relevant hashtags

Configuration

Before running, set in CONFIG.md:

YOUR_VOICE_STYLE=          # e.g. direct and data-driven, warm and conversational
YOUR_LINKEDIN_TONE=        # e.g. insight-led, contrarian, story-driven
YOUR_BANNED_PHRASES=       # e.g. synergy, leverage, game-changer, excited to share
YOUR_ICP_INDUSTRY=         # Used to make posts relevant to your audience
ANTHROPIC_API_KEY=

How to invoke

# CLI with a topic
agent run linkedin-post-generator --input "Your topic or observation here"

# CLI with no input — prompts you to enter one
agent run linkedin-post-generator

# Voice (after ACTIVATION.md setup)
"[your wake word] linkedin"

# Chain from another skill
agent run ai-news-digest | agent run linkedin-post-generator
agent run icp-radar | agent run linkedin-post-generator

# Scheduled — set your preferred days/times in crontab or Task Scheduler
# See ACTIVATION.md for scheduling instructions

Hook styles — the 3 variants

Claude system prompt (what runs under the hood)

You are a LinkedIn content strategist. Write 3 post variants from the following input.

Voice style: [YOUR_VOICE_STYLE]
Tone: [YOUR_LINKEDIN_TONE]
Target audience: [YOUR_ICP_TITLES] in [YOUR_ICP_INDUSTRY]
Never use these words or phrases: [YOUR_BANNED_PHRASES]

Input: [USER_INPUT]

Variant 1 — INSIGHT-LED: Lead with a specific number, result, or observation.
Variant 2 — STORY-LED: Open with a scene, moment, or situation.
Variant 3 — CONTRARIAN: Challenge a commonly held belief in your industry.

Rules for all variants:
- First line must stop the scroll — no "I am excited to share"
- No more than 200 words
- One clear point per post — do not try to say everything
- End with a question that invites comments or a statement that lands hard
- No bullet point lists — write in short paragraphs
- No hashtags in the body — add them at the end only

Also return:
- hook_score for each variant (1-10)
- recommended_variant with reason
- best_posting_time (day and time window based on B2B LinkedIn patterns)
- hashtags (5 relevant ones, no more)

Return valid JSON only.

Output format

{
  "posts": [
    {
      "variant": "INSIGHT-LED",
      "text": "Post text here...",
      "hook_score": 8
    },
    {
      "variant": "STORY-LED",
      "text": "Post text here...",
      "hook_score": 7
    },
    {
      "variant": "CONTRARIAN",
      "text": "Post text here...",
      "hook_score": 9
    }
  ],
  "recommended_variant": "CONTRARIAN",
  "recommendation_reason": "Strongest hook — challenges a belief your audience holds.",
  "best_posting_time": "Tuesday or Wednesday, 7:30–9:00 AM your local time",
  "hashtags": ["#AIMarketing", "#B2BSaaS", "#MarketingOps", "#GrowthMarketing", "#ContentStrategy"]
}

Chaining with other skills

# Full morning content workflow
agent run ai-news-digest | agent run linkedin-post-generator

# Lead insight to post
agent run icp-radar | agent run linkedin-post-generator

# Natural language
agent ask "Run the news digest and turn the top story into 3 LinkedIn post variants"

Related skills

• ai-news-digest — feed today's news stories as input
• icp-radar — feed lead insights as input
• competitive-intel — feed competitor observations as input
• See ACTIVATION.md for voice and scheduled setup