Every Monday I sit down to plan the week's content. For three years that planning meeting was a 90-minute slog of opening 12 Chrome tabs, scrolling through what Jordan Crawford, Eric Nowoslawski, and Amos Bar-Joseph posted last week, copying the hooks I liked into a Notion page, then trying to derive a pattern.
Two months ago I wrote a Claude skill that does the entire thing in 30 seconds. This morning it pulled 25 posts from those same three creators, ranked the top 2 by comment count, classified every post on four dimensions, and handed me a one-page brief with the exact hook pattern and gate mechanic to use this week.
I'm releasing the skill today. It's called viral-intelligence, it runs on Trigify's profile-posts API, and you can drop it into .claude/skills/ and run it in your own niche in about 30 seconds.
The slop problem nobody is solving
Most "what should I post on LinkedIn" advice is universal. Use a hook. Pattern-interrupt. Tell a story. Post on Tuesday and Thursday.
That advice is correct and also useless. The hooks that work in dev-tools-x-Claude are not the hooks that work in B2B-sales-ops. The lead magnet types that print in May are not the types that printed in January. Niche-specific patterns drift on a weekly cadence, and "post a story about your failure" is the kind of advice that survives by being too vague to be wrong.
The Trigify team frames this better than anyone in their May Influencer Brief: the best content is reactive, not planned. Calendar content gets buried. Reactive content rides the wave the creators in your niche already started.
The problem is that being reactive at the speed of LinkedIn is a full-time job. Unless you automate it.
What the skill actually does
The input is three things: a niche keyword, 3 to 5 LinkedIn profile URLs of the people who set the tempo in your space, and an optional lookback window (default 21 days).
The output is a single Markdown file. It contains:
The top 2 posts from the lookback window, ranked by comment count, with the full text, the hook, the CTA, the lead magnet type, and the gate mechanic the creator used to capture leads.
Four aggregate tables: how often each lead magnet type was offered, which hook patterns dominated, which topics kept appearing, and which gate mechanics (keyword + must-be-connected + DM vs link-in-comments) were used.
A recommendations block that turns the raw counts into a content decision. Match this LM type. Use this hook pattern. Ride this hot topic. Copy this gate.
That last block is the value layer. Everyone else's "viral hook list" gives you a menu. This brief gives you a Monday morning decision.
What this morning's run produced (real numbers)
I run it against three creators in the AI-for-GTM space: Jordan Crawford, Eric Nowoslawski, and Amos Bar-Joseph. This morning's brief, 21-day lookback:
25 unique posts analyzed across 31 scraped (Jordan: 15, Eric: 9, Amos: 7)
Top post by comments: Eric Nowoslawski, 426 comments on "I put 29 Claude skills into this repository so anyone can launch a cold email campaign automatically that ALSO self improves." Lead magnet type: Skill File / CLAUDE.md. Hook pattern: Specific number + Free framing. Gate: comment any favorite restaurant, DM delivery of the repo link.
Second: Amos Bar-Joseph, 108 comments on "60 days ago we announced our $6M raise. today we're on a billboard in Times Square." Lead magnet type: brand-story, not a gated offer. Hook pattern: Specific number.
Dominant hook pattern in the corpus: Standard (12 of 25), then Contrarian (5), then Specific number (3). The Specific number pattern is what landed both top-2 posts.
Hottest topic: Claude (12 of 25 posts mention it). Then AI Agents (8), GTM (7), Outbound (6).
Lead magnet types in the corpus skew toward
Unknown(12) because most posts in the set were observations or stories without a gate. Among the gated posts, Skill File / CLAUDE.md tied with Course/Training and Playbook.
The recommendations block converted those numbers into one sentence: this week, ship a Skill File lead magnet (matching Eric's winning structure), use a Specific number hook, tie the framing to Claude × your specialty, and gate it with a comment-based DM delivery.
I'm following the recommendation. The post you're reading is the result. The skill itself is the Skill File lead magnet.
The four classification dimensions, briefly
If you want to understand what's actually getting tagged, here is the logic.
Lead magnet type is detected by regex against the full post text, but only if a CTA gate is also present. So a post that name-drops "playbook" but never gates is correctly classified as None. The skill knows nine types: Playbook, System / Blueprint, Skill File / CLAUDE.md, Prompt System, Swipe File, Automation / Script, Course / Training, Guide / Framework, and Unknown.
Hook pattern is detected against the first two non-empty lines only. Eleven patterns: Specific number, Death-of-category, Contrarian, Result-led, Cost replacement, Effort compression, How-to, Steal framing, Free framing, "The exact", and Standard as the catch-all.
Topics are extracted via a YAML dictionary you edit. The shipped dictionary is calibrated to AI-for-GTM (Claude, Outbound, Lead gen, ICP, GTM, Intent signals, etc.). If you sell to RevOps, you replace those entries with Attribution, CRM hygiene, Pipeline metrics. The classifier doesn't care.
Gate mechanic captures the structure: the ALL-CAPS keyword in the CTA, whether the creator required "must be connected", which engagement actions they asked for (like, repost, comment), and the delivery mode (DM vs link-in-comments). That last field is the one most "viral hook" guides ignore, and it's the one that actually predicts conversion.
Why the skill ranks by comments, not reactions
A reaction is one click. A comment requires the reader to form a sentence. Optimizing for reactions optimizes for tepid content that everyone clicks Like on and no one actually engages with. Comments are the signal that the post hit a nerve.
If a creator in your niche gets 500 reactions and 12 comments, they are not your benchmark. They are an algorithm-pleaser. Drop them from your list.
Tune it for your niche in 5 minutes
The skill ships calibrated for AI-for-GTM, but the calibration is three editable blocks:
The topic dictionary (Step 3.3 in the skill file). Replace the YAML with topics relevant to your space. If you write to founders, your topics are PMF, Fundraising, Hiring. If you write to RevOps, your topics are Attribution, CRM hygiene, Pipeline metrics. About a dozen entries is enough.
The lead magnet type regex (Step 3.1). Skill Files don't matter to non-technical audiences. Drop that rule and add ones that match your space. Calculator. Audit. Template. The classifier picks the first matching rule, so order them by what your audience actually buys.
The 3 to 5 creators. Pick people who publish weekly with comments greater than reactions. Re-pick every quarter. The people who were the tempo-setters in Q1 are not the tempo-setters in Q3.
If you do those three things, the brief you get out the other end describes your specific niche, not the internet's average.
How to run it
The full skill file lives here. Copy the content into .claude/skills/viral-intelligence/SKILL.md in your project. Then:
# Install Trigify CLI (one-time)
npm install -g @trigify/cli
export TRIGIFY_API_KEY="sk-tri_..." # get one at app.trigify.io
# Drop the skill file at .claude/skills/viral-intelligence/SKILL.md
# Then in Claude Code:
/viral-intelligence devtools \\
<https://www.linkedin.com/in/creator-1/> \\
<https://www.linkedin.com/in/creator-2/> \\
<https://www.linkedin.com/in/creator-3/> \\
--lookback=21
That's it. The brief lands at .viral-intelligence/{your-niche}/{today}/viral-brief.md. Skim it in two minutes, pick the hook pattern, write the post.
If you want it to run on its own every Monday at 8am, the skill ships with scheduler templates for cron, systemd, launchd, Windows Task Scheduler, and Claude Code scheduled tasks. Pick the one that matches your OS and your brief shows up while you're making coffee.
Why I built this on Trigify
Two reasons. The first is that Trigify's signal layer covers LinkedIn, Reddit, X, YouTube, Hacker News, podcasts, and Substack from one API. So once you have the skill running for LinkedIn, swapping the scrape command in Step 1 gives you the same brief for Reddit threads in your niche, or trending X posts, or the podcast episodes your buyer just listened to.
The second is that their CLI is built for exactly this pattern. Long agent sessions, scheduled runs, clean JSON output. The MCP version works too if you want the conversational flow inside Claude Code or Cursor.
If you're already paying for a scraping stack (Apify, PhantomBuster, a chain of scripts), this collapses it into one CLI. If you're not, this is the cheapest way to get a Monday morning content radar without writing your own scrapers.
What you get this week
The full viral-intelligence skill file, published here and ready to drop at .claude/skills/viral-intelligence/SKILL.md. About 330 lines of executable instructions, full regex tables, scheduler templates for every major OS, and a tunable topic dictionary.
I'm shipping it because the next 12 months of LinkedIn content are going to be won by the people who post on the wave instead of after it. The hooks that worked in March are already cooling. The ones that will print in July are being written by 3 to 5 creators in your niche right now. The skill tells you who they are and what they're doing.
Subscribe below and a fresh brief drops in your inbox every time I refresh the recommendations.
Until next week,
The GTM Architects