What Is LinkedIn Agent Skill? Writing LinkedIn Posts with Claude, Installation, Cost and LinkedIn Rules

A skill is an instruction file that teaches Claude how to do a job. This collection is for LinkedIn: it turns a raw idea into a post, writes comments on other people's posts, sorts incoming messages and scores your profile. Everything is a draft; you read it, fix it and paste it into LinkedIn yourself.

Here's how it works: first you put writing samples into the voice.md file (the README says to do this first, otherwise the output sounds "like everyone else's"). Then you give it an idea; /li-post offers three opening-line options and one full draft, and if you say "yes" it records the post in the log file. Nothing is posted.

!The README's first sentence can read like "it manages your LinkedIn account", but further down it clearly says nothing is posted automatically. It doesn't manage your account; it only writes.

You don't have to install all of them; you can copy folders one by one. But according to the README, li-post, li-comment, li-reply and li-dm run their output through li-human; if you don't copy li-human too, this step is skipped.

!li-human cleans traces of AI writing (invisible characters, stock words) and scores the text with a local rubric. The README says it doesn't promise the verdict of real detectors like GPTZero. If your school, employer or a platform has a rule on AI use, following it is up to you. In one user issue (issue #2), a Claude session refused the installation for exactly this reason Reported by users.

The commands are taken verbatim from the README. Both routes require Claude Code on your own computer; the skill files go into the ~/.claude/skills/ folder.

After installing, copy templates/voice.md to ~/.claude/linkedin/voice.md and fill it in, or paste three of your posts into Claude and say "write voice.md from these" (the README says it takes ten minutes).

The repository documentation doesn't describe how to remove the plugin after installing it; use Claude Code's /plugin menu. Python 3 is needed only for the li-human scripts and requires no extra packages.

You type the command and describe your idea. The text goes nowhere without your approval; if you say "yes", it is only recorded in the local log file.

According to the README, the skill doesn't invent numbers it doesn't have; it leaves {{your number}} in their place and warns you. Provide your real number yourself.

!The skills are written in English. You can ask for a Turkish post, but the official sources don't state the quality on Turkish text or whether the cleanup word list (113 English terms) works for it.

The plugin definition has no hooks or MCP servers. skills/li-post/hooks.json is not a Claude Code hook but the data file for the 21 post-opening formulas. No advisories are listed in GitHub Security Advisories, GitHub's global advisories or OSV (October 15, 2026); since the repository isn't a published package, the OSV result has limited value. This doesn't mean "safe"; there is no SECURITY.md.

Yes. The README says "free, MIT, no signup, no API key"; the LICENSE file in the repository is also MIT. The real cost is the Claude Code/Claude plan you run the skills on, which is separate.

The repository is new and has a single developer; the developer has not replied to the issues. There is no commitment to updates.

This collection doesn't connect to LinkedIn: no API key, login or browser automation. According to the README, this is a deliberate design; the skill files also tell the agent not to post comments with a browser tool, not to read the inbox through a browser and not to scrape the feed.

!The risk of automatic posting starts with you: if Claude also has a browser tool, a social media MCP or a posting automation connected, that tool may do the job when you say "post this". The skill forbids it, but if you don't set the rule, another tool may take over. Posting manually is safest.

This is not legal advice; LinkedIn's terms may change, so check the current text on LinkedIn. Drafting is one thing, automatic posting is another; this collection only does the first.

1. 1If you copied it manually: Delete the li-* folders under ~/.claude/skills/.
2. 2If you installed it as a plugin: Remove it from Claude Code's /plugin menu.
3. 3Your files remain: voice.md, log.md and plan.md in ~/.claude/linkedin/ are not deleted automatically; delete them yourself if you want.