Jake Schincariol releases eleven free Claude skills for LinkedIn, and one of them scores an AI-written draft before it goes out: Reveneau

Why it mattersA single Claude command can now check whether a LinkedIn draft still reads like AI on numbers a reviewer can see, so the fix is a rewrite rather than a hope that the audience will not notice.

A LinkedIn post written by a language model still reads as one to most people who write LinkedIn posts as their day job, and Jake Schincariol has released eleven Claude Code skills under the MIT licence that flag the specific signals before the draft is sent. His linkedin-agent-skill repository, published on 7 September and now at 655 stars on GitHub, ships eleven separate skills that a person invokes by name in Claude Code, each doing one narrow job on a LinkedIn account.

The eleven, as the README lists them: li-post for writing a post from an idea, li-comment for replies to other people's posts, li-reply for the writer's own thread, li-profile for scoring a profile against a twelve-part rubric out of 100, li-plan for weekly content, li-human for the humanizer described below, li-carousel for slide-by-slide document posts, li-repurpose for turning one piece into a week of posts, li-dm for outreach messaging, li-inbox for triaging incoming messages, and li-audit for a post-mortem ranked by engagement.

The humanizer, and what it counts

The li-human skill is what makes this a workflow story rather than another template pack. It scores a draft on five things the author names in the README: burstiness, meaning the range of sentence lengths, because a language model writes them all the same length; specificity, meaning numbers, names and concrete markers per 100 words; slop density, meaning matches from the author's own list of 113 stock words and phrases per 100 words; fingerprint, meaning invisible characters, em dashes and curly quotes per 1,000 characters; and voice, meaning contractions, use of "I" and "you", and the sentence structures that identify AI writing. The output is a 0 to 100 human score.

It rewrites what it can. The author says it removes invisible characters, fixes typography and swaps the 113 flagged words for plain-English equivalents. It flags structural patterns for the writer rather than trying to fix them, because a sentence that starts "Not X, but Y" needs a person to decide what the sentence was actually meant to say.

What ships with the post skill

li-post returns three hook options selected from 21 formulas that the author has recorded in the repo, along with one full draft. The draft has already run through the humanizer before the writer sees it. Nothing in the repository connects to LinkedIn: the skills produce text that a person then pastes into the site by hand, and there is no API key, no signup and no account link. The whole thing runs locally, and the repository states that "no uploads" happen at any step.

Traction, briefly

The repository is 17 days old and carries 655 GitHub stars, 96 forks and an MIT licence, plus install paths for direct paste, git clone and Claude Code's plugin marketplace. Whether the humanizer's scoring holds up against a person who writes many LinkedIn posts a week is a check the writer has to run on their own drafts. The shape of the release, one measurable score plus a list of things to remove, is more useful for a team already writing with Claude than another prompt library.

Source

Jake Schincariol, Jakeschincariol/linkedin-agent-skill on GitHub, MIT licence.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.