How to give Claude a LinkedIn skill that actually sounds like you

A LinkedIn agent skill for Claude is a small package of instructions that teaches Claude how to do LinkedIn work your way: how to write a hook, how long a post should be, where the "see more" fold falls, which phrases you never use. Claude reads the short description of every installed skill, and when your request matches one, it loads the full instructions and follows them.

Installing one takes about a minute. The quality comes from two things most people skip: what you put in the skill, and how you correct it over time.

I use Claude every day for my own LinkedIn and for building Liftli, and I've spent more time than I'd like to admit tuning skills. Here is what works.

What is an agent skill, and how is it different from a prompt or an MCP?

People mix these three up constantly, so let's separate them.

A skill is a reusable way of working. An MCP is a pair of hands. For LinkedIn, you usually want both: a skill so the drafts are good, and a connector so the good draft actually gets posted without copy and paste.

The mechanics of a skill are simple. At the top of SKILL.md there's a short header with a name and a description. Claude keeps those descriptions in view all the time, and only reads the rest of the file when your request fits. That's why one Claude can carry dozens of skills without getting confused. It's also why the description matters so much: if it's vague, Claude won't reach for the skill at the right moment.

How to give Claude a LinkedIn skill

There are three routes, depending on where you use Claude.

In Claude Code. Skills live in folders. Personal skills go in ~/.claude/skills/<skill-name>/SKILL.md and are available in every project. Project skills go in .claude/skills/ inside a repo. To install a ready-made set, you can use a package command. For example, our free set installs with npx skills add liftli-ai/linkedin-agent-skills, and the full list is on the LinkedIn agent skills page.

In the Claude apps (web and desktop). Skills can be uploaded as a zipped folder from Claude's settings, under the capabilities or skills area. Anthropic moves this menu occasionally, so look for "Skills" in settings if it isn't where you expect.

Write your own. Create a folder, add a SKILL.md with a name, a one-line description of when to use it ("Use when the user asks to draft, rewrite or critique a LinkedIn post"), and then the instructions in plain language. Add example posts as separate files in the same folder and tell the skill to read them.

If you're starting from zero, install a ready-made set first and read the files. You'll learn more about what a good skill looks like in ten minutes of reading than in an hour of writing one blind.

One thing I learned from my own internal skills: the value has to come before the convenience. When we first built a LinkedIn-expert skill and an architecture-expert skill for ourselves, neither triggered on its own. I had to remember to call them in the middle of a conversation, and at first I often didn't. That changed once I saw the gap between the answers I got with the expert and without it. The advice and the action items were so much better that I started calling it all the time, clunky or not. Only after that did it make sense to smooth the triggering, which is what the one-line description in SKILL.md is for. So get the instructions good first, then make the description specific enough that Claude reaches for the skill without being asked.

What a good LinkedIn skill contains

Most LinkedIn skills I've seen are just a long prompt with formatting rules. That gets you clean posts that sound like everyone else. The pieces that make a difference:

1. A voice file. Your real posts, or a distilled description of how you write: sentence length, how you open, words you use, words you'd never use. Build it from your own writing.
2. Your strategy. Who you're writing for, the three or four topics you want to be known for, and what you're trying to get from LinkedIn (hires, customers, investors). Without it, Claude writes about whatever you mention, with no direction.
3. Platform rules. Post length, where the "see more" fold lands (roughly the first 210 characters on desktop), your rule on links in the body, how to format lists. You can check a draft's fold with a post preview tool.
4. A banned list. The phrases that make you cringe. Mine includes em-dashes and "it's not X, it's Y" constructions. I don't want either anywhere near my posts.
5. Good examples with reasons. Two or three posts that worked, each with a note on why it worked. This is the part almost everyone leaves out, and I'll come back to it.
6. A source of truth for facts. Instructions to only use stories and numbers you've supplied. Claude will happily invent a customer anecdote to make a post land. The skill should forbid that.

A skill that has all six will still produce drafts you edit. It'll just produce drafts worth editing.

Curate what your LinkedIn skill reads, like a library

The biggest lesson I've learned about AI and writing is that the input decides the output. And most people never think about the input.

Here's a concrete example from my own setup. For my research notebook, I didn't let the AI pull whatever it found. I hand-picked the voices I wanted shaping its thinking: Adam Robinson, Elena Verna, Andrej Karpathy, Greg Eisenberg, Gal Aga. Then I added a large archive of their long-form writing, batched by topic and by voice. My instruction to the AI doing the packing was to fit them in a way that maximizes the tokens available per single source, so nothing got summarized into mush.

Curating the inputs was harder than building the system. It was also the point. The AI stopped being neutral and started having a perspective, one I'd chosen on purpose.

The same applies to a LinkedIn skill. Before you write instructions, decide what the skill should read:

• Your own best posts, the ones that got real conversations going. Leave out the ones that only got likes.
• Your raw thinking: voice memos, call notes, messages where you explained something well. That's where the stories are. I built my voice memo workflow for exactly this reason.
• Two or three writers you admire, as a reference for structure only.
• Nothing generic. "Top 50 viral LinkedIn hooks" lists pull every draft toward the average.

There's a trap here I learned the hard way. Matching someone's voice perfectly falls short on its own. We once matched a CEO's voice so well that he said the result sounded like every CEO on LinkedIn. The voice was right. The substance was missing. A curated library of your actual thinking is what fixes that.

Train your Claude skill with feedback, including the good kind

A skill is never finished. The way it gets better is feedback, and I noticed something about my own feedback a few months ago that changed how I do it.

I used to give Claude only negative feedback. This is wrong. Too long. Cut the second paragraph. Make it punchier. That's the natural way to correct AI, and it works for fixing one draft.

Then I caught myself complimenting Claude. We record these working sessions because I'm building a version of Claude trained to think like me, and I realized that if every recorded example is a correction, the model learns what to avoid and nothing about what to repeat.

So now, when a draft is good, I say why, specifically. Something like: "This works because the hook states a specific number from my own week, the second line creates tension, and the ending asks a question only an operator would answer." That's a principle I took from management training: identify the components that make something succeed. Once you can name them, you know what to aim for when they're missing.

For a LinkedIn skill, this turns into a simple habit:

1. When a draft lands, write down why in one or two sentences.
2. Add that post and the reason to the skill's examples file.
3. When a draft misses, name the missing component, using the same words you used when it worked.
4. Every few weeks, read the examples file and promote repeating reasons into the main instructions.

After a month of this, the skill carries your taste, written down. That's something a generic LinkedIn prompt can't give you.

LinkedIn skill vs LinkedIn MCP: when you need both

A skill can't publish. It shapes how Claude writes, and then you still have to copy the post, open LinkedIn, paste it, fix the line breaks, and hit post. For a lot of people that's fine.

If you want Claude to do the whole loop (decide what's worth posting this week, draft it, let you approve it, and publish or schedule it), you need a LinkedIn MCP connector alongside the skill. The MCP is the part that talks to LinkedIn. Look for one that uses LinkedIn's official API and sign-in, and that requires your approval before anything goes out. Tools that automate a logged-in browser session put your account at risk.

My honest take on the split:

• Skill only if you post occasionally and you enjoy the final edit in LinkedIn itself.
• Skill plus MCP if you post weekly or more and the copy-paste step is where your consistency breaks.
• Neither if you don't have anything specific to say yet. No skill fixes an empty input. Start by capturing your thinking.

If you're comparing options beyond Claude, I keep a running list of the best AI LinkedIn tools and who each is for.

Why a hands-on setup beats a black box

My belief is that the best leaders in the AI age are hands-on. You can't lead what you've never touched. Setting up your own LinkedIn skill is a small, practical version of that. You open the file, you read what Claude is being told, you change it when a draft goes wrong.

A founder who hands LinkedIn to a tool they never look inside gets posts that belong to the tool. A founder who owns the skill, curates what it reads and tells it why the good drafts were good ends up with something closer to a trained writer who never forgets a note.