A LinkedIn Content Engine in 22 Claude Skill Files

The content engine, as Claude skill files

22 files that turn Claude into a writer who sounds like you.

We analyzed 1,000+ high-performing LinkedIn posts plus a full pass over the Kleo swipe corpus, and reduced what recurred into a working system: the hook types that actually earn the click, the post structures that actually hold attention, and the mechanisms underneath both.

Then we wired it to your voice. Not a prompt pack, not templates you fill in: a system that learns how you write from your own posts, drafts against your rules, and checks every line before you ever see it.

Everything here is plain markdown. No database, no API keys, no accounts, nothing to install. Open it, read it, change anything you disagree with.

---

01Why this is different from a prompt pack

A prompt pack gives you someone else's voice. This builds yours.

The first thing you do is paste 5-10 of your own posts. Claude reads them for invariants: what you always do, what you never do: and turns that into a set of rules it can mechanically check. Line length. How you open. Whether you use questions. Which markers you use. Whether you concede the counter-argument.

Then every draft gets linted against those rules before it reaches you.

And it keeps learning. Correct the same thing twice and Claude proposes a new rule. Mention a durable fact about yourself and it proposes saving it. It always asks first: nothing gets written to your profile silently.

---

02What's inside

The hook library: from 1,000+ viral posts

13 hook archetypes. Not thirteen because it's a nice number: thirteen because that's what actually recurred at volume across the corpus. Each one with its shape and what it's for:

bold-claim · curiosity-gap · number-led · confession · contrarian · pain-callout · story-open · question · social-proof · before-after · direct-address · list-promise

Plus the part most hook guides skip: what's banned and why:

• Audience-shaming openers ("most founders don't realise..."): you're telling the reader they're stupid before you've earned anything
• Follower-count storytelling: reads as fabricated
• A hook the post doesn't pay off: lifts the click, tanks the dwell time, and the platform measures that
• Mirroring the visual: if the image says it, the hook says something else

And a diagnostic for fixing a weak hook, in order. The one that works most often: delete the first sentence, then check whether the second one was the real opener all along.

The swipe file

A system for capturing other people's posts as analyzed structure instead of screenshots.

Every swipe gets reduced to four things: its post type, its hook type, its beat-by-beat skeleton, and one paragraph on the mechanism that earned the engagement: not a summary of what it said.

post_type: framework
hook_type: number-led
topics: [hiring, sales]

## Structure skeleton
number-led contrast hook → the reframe → 3 numbered components → proof → imperative close

## Why it works
The hook pits a large external number against a small personal one, so the reader
assumes a contrarian argument is coming and stays for it. Proof arrives after the
framework rather than before, so it reads as evidence instead of a credential.

Ships with 16 archetypes distilled from the corpus: the recurring structures behind the 1,000+ posts we analyzed, reduced to reusable patterns. So matching works on day one, before you've captured anything. Then you add your own as you see good posts.

The rule that makes a swipe file safe to keep: swipes lend structure, never content. Someone else's numbers are not your proof. Someone else's phrasing in your post is plagiarism.

The post taxonomy: the shapes that recur

12 post types from the same corpus analysis, each with its telltale skeleton:

Type	Telltale skeleton
story	hook → context → turning point → lesson
framework	hook → principle → numbered components → proof
case-study	result hook → situation → what was done → numbers
contrarian-take	consensus → "wrong" → counter-argument → reframe
listicle	list-promise hook → N items → takeaway
how-to	outcome hook → steps in order → result

...plus analysis, observation, one-liner, announcement, lead-magnet, promo.

One shared vocabulary across swipes, templates, and drafting. That's what makes matching work.

Four deep playbooks

Full architecture for the four post types that carry the most risk and the most upside:

• Case study: anonymization rules, the two structures (result-first vs narrative), how to resolve conflicting numbers, and the traps that make a case study read as a brag
• Lead magnet: the 8-step caption architecture, the gated-vs-open decision (don't default to gatekeeping), keyword case variants, the DM, and ~10 comment replies that don't read as a bot
• Partner / sponsor: evidence-before-features structure, and the affiliate honesty rules: never claim a promo you haven't verified, never say "free credits" if the referral only pays you
• Thought leadership: the founder-thinking-out-loud register. Why "the problem emerges mid-story" beats announcing your conclusion, and why specificity beats credibility badges

The fabrication guard

The single most valuable file if you write about client work.

Nine rules on what you can and cannot claim, the anonymization rules, and a fabrication ledger: real drafts that were written, caught, and pulled. An invented client count. A borrowed revenue figure. Quotes attributed to people who never said them. A discount claim nobody verified.

The pattern across all of them: each one felt like it was making the post stronger. That's why the file also carries a rationalization table: the specific thoughts that precede the mistake, next to what's actually true.

Repurposing

• X threads: with the full playbook: 5 format maps, the drop-off curve, and the actual engagement weights from X's open-sourced ranker. Including the one that matters most: a reply the author replies back to is worth roughly 150 likes
• Newsletters: the format runs opposite to a social caption. Section headers instead of flowing prose, no tags, not in a code block, and the CTA in the PS
• Subject lines: always a set of options plus a recommended A/B pairing, never one guess
• Newsletter calendar: a month planned: topic, track, day, author

Ideation

Post concepts that pass four gates before they reach you:

1. Obsession topic: is the audience already asking about this?
2. Concrete noun: does the hook name a tool, role, channel, figure, or year?
3. One-second shape: can the visual be read in one glance?
4. Honest number: is it a number the reader has gut feel for?

Plus the six visual shapes that work, the metaphors that are banned (icebergs, Jenga towers, dominos), and the rejected-idea classes that get killed on sight.

---

03The 15-minute setup

1 · Upload the folder to a Claude Project. Or open it in Claude Code.

2 · Say: Read 00-start/setup.md and run it with me.

3 · Answer the interview. Six steps, resumable across sessions:

Step	What happens	What you need
1 · Identity	Claude interviews you on your background, story, offers, audience	10 minutes
2 · Writing style	You paste 5-10 posts; Claude derives your rules and shows them for editing	your posts
3 · Knowledge	You add what content can draw real claims from	a doc or two
4 · Swipe file	You paste posts you admire; Claude analyzes each	10 posts
5 · Templates	Claude turns your best posts into reusable skeletons	optional
6 · First post	Claude drafts one using everything above	None

Step 2 is the one that matters. Give it the posts that sound most like you at your best, not your highest-engagement ones. A post that went viral because of its topic teaches nothing about how you write.

4 · Then write: Read 04-writing/write-post.md. Write me a post about [thing].

---

04How a draft gets made

1. Loads your profile: identity, every active rule, 2-3 of your exemplars
2. Frames the post with you: the idea, the post type, the topics, which audience it serves
3. Matches a structure: your own templates first, then swipes
4. Loads the playbook for that post type, if one exists
5. Drafts two contrasting takes: so your feedback can be specific
6. Lints every line against your rules and fixes violations before showing you
7. Learns: repeated corrections become proposed rules

The hook is written last. It's easier once you know what you're opening, and a hook written first bends the post to fit it.

---

05The seven rules that hold across everything

1. Your profile wins. Precedence: voice > playbook > template > swipe.
2. Nothing is fabricated. No source, no claim. Cut, never softened.
3. Swipes lend structure, never content.
4. Everything is draft-only. Nothing publishes or schedules. You post.
5. The profile changes by propose-then-approve.
6. A generic draft is an upstream problem: go deepen the profile, don't add adjectives.
7. Length and shape derive from the content: never a default number.

---

06The folder

00-start/        content-engine · setup
01-foundation/   voice-profile · proof-and-claims · hooks · post-formatting
02-reference/    swipe-file · post-taxonomy · template-library
03-ideation/     ideate-post-concepts · ideate-lead-magnets
04-writing/      write-post · 4 playbooks · convert-to-voice
05-repurposing/  x-thread · x-thread-frameworks · newsletter
                 subject-lines · newsletter-calendar
your-profile/    ← your data. identity, rules, exemplars,
                   knowledge, templates, 16 swipe archetypes

16 skills + 6 reference libraries.