Free agent skills for writing X and LinkedIn posts

We published two free skills that help an AI agent write social posts, one for X and one for LinkedIn. They live at postmoderno/postmoderno-skills on GitHub under the MIT license. They run in Claude Code, claude.ai, the Claude desktop app, and any agent that reads the Agent Skills format. You don’t need a postmoderno account to use them.

A skill is a folder of instructions an agent loads when a task calls for it. Ask Claude to “turn these notes into an X post” and it loads the X skill, which knows how X ranks posts, what gets a post labeled as spam, and how X counts characters.

Why we wrote them

Much of the advice about “the algorithm” repeats old numbers. “A reply is worth 27 likes” comes from code X published in 2023. “Put the link in the first reply” works around a link penalty that X’s leadership said in July 2026 had been gone for over a year.

Each skill starts from the best source its platform offers and puts a date on every claim, so you can see which rules come from code and which are rules of thumb.

What the X skill is built on

X has open-sourced its For You ranking code as xai-org/x-algorithm. The skill works from a snapshot of it taken at the end of October 2026. For each person scrolling, a model predicts how likely they are to take each action on a post, and the post’s score is a weighted sum of those predictions:

The weights multiply probabilities, so they aren’t exchange rates. A report doesn’t cancel a fixed number of likes; it carries a big weight because reports are rare. The table does show that a post people send to a friend or answer counts for far more than one they like, and that a post people regret seeing costs a lot.

The code shows a few more things the skill uses:

• Small accounts get a lift. An original post from an account with up to 50,000 followers, less than two hours old and with fewer than 200 impressions, can be moved up to 16th place in someone’s feed.
• Posts that ask for engagement (“like if…”, “RT to…”, “follow for more”) get labeled spam, even from accounts X otherwise trusts.
• A post with two or more @mentions goes through an extra spam check.
• X puts an “AI slop” label on posts and on accounts. For seven days, an account with the label reaches few people beyond its followers.
• Hashtags carry no ranking weight, and stuffing them counts as spam.

The skill also ships a small checker that estimates X’s weighted character count (any link counts as 23, an emoji as 2) and flags bait phrases, extra mentions, hashtags and common AI phrasing before you see the draft.

What the LinkedIn skill is built on

LinkedIn hasn’t published its feed code, so there are no weights to quote. The skill works from what LinkedIn’s engineers have published: a 2026 paper on the model that ranks most of the feed, engineering blog posts on dwell time, and LinkedIn’s March 2026 announcement that it is cutting back engagement bait, automated comments and engagement pods.

The advice that follows from them is plain: write for a specific professional reader, open with the point, back it with your own facts, and leave out “Agree?” and “comment YES”. The skill is just as clear about what no source shows: a penalty on every link, a best time to post, a first-hour cutoff or an ideal length.

What the skills won’t do

They won’t make up facts. If a post needs a number, a customer name or a story you didn’t give, the agent asks you or leaves a visible placeholder. They don’t promise reach. Their reference files keep code, statements by platform staff, and third-party studies apart, so you can check where each rule comes from.

Install

In Claude Code, run:

/plugin marketplace add postmoderno/postmoderno-skills
/plugin install postmoderno@postmoderno-skills

For claude.ai and the Claude desktop app, download the repository, zip the folder of each skill you want (skills/x-posts, skills/linkedin-posts), and upload it under Settings → Capabilities → Skills.

For other agents, copy the skill folders into the folder your agent loads skills from.

Then ask the way you normally would:

Here are my notes from today’s release. Write an X post and a LinkedIn post about the new export.

Why does X keep showing my posts only to my followers?

With postmoderno connected

The skills work on their own. Connect postmoderno as well, at https://mcp.postmoderno.io/mcp, and the agent:

• writes in the voice you saved in postmoderno, whichever AI client you use,
• shows you a preview with the time, the account and what the post uses from your plan, and schedules only after you say yes,
• schedules an X thread as your first post with each reply under the one before it,
• tells you when a link makes an X post use more of your X allowance, and offers a version without it.

The setup guide has the steps for each client.

Keeping them current

Platforms change how they rank posts often. The dated facts sit in each skill’s references folder, apart from the instructions, so an update changes one file. If you spot a claim that’s out of date, open an issue or a pull request with a source.

Questions

Do I need postmoderno to use the skills?

No. They write and check posts on their own. postmoderno adds your saved voice and scheduling.

Which agents can use them?

Claude Code, claude.ai and the Claude desktop app, plus any agent that loads skills in the Agent Skills format: a folder with a SKILL.md file in it.

Does X rank AI-written posts lower?

Nothing in X’s published code penalizes a post for being written with AI help. X’s labels target slop patterns and spam behavior. That’s why the skill writes in your voice and checks the draft for common AI phrasing.

Will the skills make my posts reach more people?

Nobody can promise that. The skills help you avoid what the platforms penalize and write posts worth reading.

Are they free?

Yes. They’re MIT licensed, so you can use, change and share them.

← all posts