This is the setup guide for the LinkedIn skill pack from the reel. It is a free, MIT-licensed project on GitHub that turns Claude into a LinkedIn writing assistant. One thing to know up front: it does not log in to LinkedIn or post for you. It writes drafts, and you paste them in yourself. That is by design, and it keeps your account safe.
What you need
Claude. Works best in Claude Code, because the humanizer uses two small Python scripts. You can also paste a single skill file into a normal Claude chat, but you lose those scripts.
The skill pack: Jakeschincariol/linkedin-agent-skill on GitHub. Free, MIT license.
Three of your own LinkedIn posts, to teach it your voice.
Python on your computer, if you want the humanizer scripts to run.
Step by step
Install the pack. Pick one of the three methods from the repo README.
Option A, ask Claude Code to do it. Paste this:
https://github.com/Jakeschincariol/linkedin-agent-skill
Install this skill, then confirm /li-post works.Option B, copy the files yourself:
git clone https://github.com/Jakeschincariol/linkedin-agent-skill.git
cp -r linkedin-agent-skill/skills/li-* ~/.claude/skills/Option C, install as a plugin inside Claude Code:
/plugin marketplace add Jakeschincariol/linkedin-agent-skill
/plugin install linkedin-agentSet up your voice file. Every skill reads
voice.md. The README says skipping this makes the output generic. The quick way is to paste three of your posts into Claude with this prompt:
Here are three LinkedIn posts I wrote myself. Read them and write my
voice.md file for the linkedin-agent-skill pack, following the format in
templates/voice.md. Capture my sentence length, the words I use, the
topics I cover, how I open and close posts, and anything I never do.
Save it to ~/.claude/linkedin/voice.md.
[paste post 1]
[paste post 2]
[paste post 3]Know the 11 commands. Here is what each one does, per the README:
/li-post: Three hook options from 21 hook formulas, plus one full draft
/li-comment: Drafts a comment on someone else's post
/li-reply: Sorts comments on your post and drafts replies, most important first
/li-profile: Scores your profile out of 100 and rewrites the weakest parts
/li-plan: Weekly plan: what to post, when, and 10 people to engage with
/li-human: The humanizer (cleans and scores a draft)
/li-carousel: Slide copy and a PDF for a document post
/li-repurpose: Turns a video, newsletter or transcript into a week of posts
/li-dm: Connection note, first message and two follow-ups
/li-inbox: Sorts your inbox into leads, recruiters, peers and spam
/li-audit: Ranks your past posts by engagement rate, not raw impressions
Write your first post. Try this in Claude Code:
/li-post
Idea: I moved my team's weekly report from Excel to a shared dashboard
and we stopped spending Monday mornings on it. Audience: operations
managers in India. Keep it under 200 words.Pick the hook you like, edit the draft so it sounds like you, then paste it into LinkedIn.
Run the humanizer on anything you wrote with AI help.
/li-human
[paste your draft]It swaps em dashes, curly quotes and 113 stock "AI-sounding" phrases for plainer ones, strips invisible characters, and flags patterns like "not just X, it's Y" for you to rewrite. It then gives the draft five local scores (sentence variety, specifics, stock-phrase density, formatting fingerprints, voice).
Comments and replies: review every one.
/li-commentand/li-replyonly draft. Read each one, change it, and post it yourself. Skip any comment you would not write by hand.
Good to know
It does not run your account. The repo says there is no signup, no API key and nothing to connect. It cannot post, comment or reply on its own.
Do not wire it to a bot. LinkedIn's rules ban software that automates activity on the site, and accounts can be restricted. The author says the same. See LinkedIn's prohibited software page.
No "undetectable" promise. The five checks are the author's own local scores, not real AI detectors like GPTZero or Turnitin, and the README says nobody can honestly promise a draft won't be flagged. It does not claim to remove any cryptographic AI watermark. Use the humanizer to make writing plainer, not to pass off AI text as fully your own. The best fix is still to rewrite drafts in your own words.
It won't invent results. If a draft needs a number you didn't give, it leaves a
{{your number}}placeholder. Fill it with a true figure or cut the line.Free, but Claude isn't. The pack costs nothing. Claude Code needs a paid Claude plan or API billing.
Sources
Found this useful? Follow @kevinsingh.me on Instagram for a new AI tool every day, and subscribe so you don't miss the next one.