How to Repurpose Content With Claude Agent Skills (2026)

You wrote a good blog post last week. Maybe 1,500 words, a real argument, a few hard-won examples. It got some traffic, sat on your site, and that was that.

Here's what should have happened to it: eight LinkedIn posts, a couple of X threads, a newsletter issue, a 60-second video script, three image captions. The same idea, recut for the places your audience actually hangs out. Almost nobody does this. Not because they don't know they should, but because doing it by hand means re-reading the post, picking out the quotable bits, rewriting each one for a different platform's rhythm, and pasting it all into five different tools. By the time you've done that for one post, you'd rather write a new one.

This is the exact gap Claude agent skills close. You build the chain once, then run one source asset through it and get the variants back in a few minutes. This walkthrough takes a single blog post and turns it into a full set of derivative content using a handful of skills you install yourself.

What an agent skill actually is

A skill is a folder. Inside it sits a file called SKILL.md with instructions Claude reads only when the task calls for it. Think of it as a specialist you keep on a shelf: Claude doesn't load the whole thing into memory all the time, it pulls the skill in when your request matches what the skill does. A repurposing skill might hold the format rules for a LinkedIn carousel, the character limits for X, your house style for subject lines. You write those once, and every future run inherits them.

Skills live at ~/.claude/skills/<name>/SKILL.md. You can write your own, copy ones other people share, or mix both. They work in Claude Code (the terminal tool) and anywhere the Claude agent runtime reads a skills directory.

The repurposing pipeline at a glance

Before the step-by-step, here's the shape of the whole chain. Each stage takes the output of the one before it and hands a tighter, more channel-specific asset to the next. You don't need all of these on day one. Start with the social cut-downs and add the rest as you go.

The source post is the hub. Everything downstream is a spoke. Get the hub right and the spokes mostly write themselves, which is the whole reason this works.

Step 1: Install the skills

Skills are just folders, so installing one means putting it where Claude looks. If a skill lives in a public repo, clone it straight into your skills directory:

# Make the skills directory if it doesn't exist yet
mkdir -p ~/.claude/skills

# Clone a shared skill into place
git clone https://github.com/some-author/typefully-social \
  ~/.claude/skills/typefully-social

# Or copy one you wrote locally
cp -r ./my-skills/brand-guidelines ~/.claude/skills/brand-guidelines

After that, confirm the file is where it should be:

cat ~/.claude/skills/typefully-social/SKILL.md

If you see the skill's instructions print out, Claude can find it. Restart your Claude Code session so it picks up the new folder, and you're set. Repeat for each skill in the table you want.

Step 2: Point Claude at the source post

Drop your blog post somewhere Claude can read it. A plain markdown or text file is fine. Then open a session in that folder and tell Claude what you're working with:

> Read teardown-of-our-onboarding-flow.md. That's the source post for today's repurposing run. Summarize its core argument and pull out the 5 most quotable lines before we start cutting it up.

Two things happen here. Claude grounds itself in your actual words instead of inventing a generic summary, and you get a quick sanity check that it understood the piece. If the summary is off, fix it now. Every downstream variant inherits whatever Claude thinks the post is about, so a wrong read at this stage poisons the whole batch.

Step 3: Generate the LinkedIn and X variants

Now the social cut-downs. This is where typefully-social earns its place: it knows that a LinkedIn post wants a strong first line and white space, while an X thread wants tension between tweets and a hook that survives being seen with no context.

> Using the typefully-social skill, turn the source post into 6 standalone LinkedIn posts and one 7-tweet X thread. Each LinkedIn post should make one point from the article and work even if someone never clicks through. For the thread, open with the most contrarian line you pulled in step 1.

Ask for standalone posts on purpose. A common mistake is to generate six posts that all say "check out my new blog," which is one post repeated six times. You want six different arguments mined from one article, each able to stand alone in a feed.

When the drafts come back, read them as a reader, not as the author. Cut any that just restate the headline. Keep the ones that pick a fight or teach something specific.

Step 4: Build the newsletter issue

Newsletters want a different muscle than social. Longer leash, a personal voice, room to tell the story behind the post rather than just the conclusion.

> Now draft a newsletter issue from the same source. Around 400 words. Open with the personal reason this topic came up, walk through the one idea most worth a subscriber's time, and link back to the full post at the end. Give me 4 subject line options, plain and curiosity-driven, no clickbait.

The newsletter is where you can be most yourself, so this is the draft you'll edit most by hand. Treat what comes back as a strong first pass, not a finished issue.

Step 5: Write the short-video script

Last stop, the video. A short script is its own format: spoken-not-read, one idea, a hook in the first three seconds because that's all the attention you get.

> Using the source post, write a 60-second vertical video script. One idea only, the strongest one. First line has to stop the scroll. Mark the on-screen text cues in brackets. Then give me caption text I can run through videocaptioner.

If you produce the video with code, remotion-video can take that script and turn the beats into a rendered composition. For burned-in captions on footage you already shot, videocaptioner handles the timing. Either way the script came from the same hub post, so the message stays consistent across every format without you re-deciding what the point is each time.

Step 6: Run the brand pass

One source post has now become roughly a dozen assets across four channels. Before any of it ships, run the lot through your brand skill:

> Run everything we just made through brand-guidelines. Flag anything off-voice, any banned words, any place the terminology drifts from our house style. Don't rewrite. Just list what needs a human eye.

This catches drift. When you generate a dozen variants in one sitting, small inconsistencies creep in: a product named two different ways, a tone that's chummy in one post and stiff in the next. The brand skill is cheap insurance against publishing those.

Where the AI still needs you

The chain gets you 80% of the way in minutes. The last 20% is the part that decides whether the content lands, and it's still yours.

Voice. Generated copy tends toward a flat, agreeable middle. Your best posts have an edge, an opinion, a way of phrasing things that's recognizably you. Add it back. The AI gives you structure and coverage; you give it a pulse.

Facts. Anything with a number, a name, a date, or a claim about how something works needs checking against the source. Models will state a confident figure that's slightly wrong, and a wrong stat in a LinkedIn post is your name on it, not the model's.

Hooks. First lines are everything on social, and they're the thing AI is weakest at. It writes serviceable openers, rarely great ones. Rewrite the first line of every post by hand. It's the best-spent 30 seconds in the whole process.

One more thing worth saying on a site about content quality: read the output for that generic "AI voice" before you post. If a sentence could have come from any brand on earth, it shouldn't go out under yours. The same scrutiny you'd apply to a freelancer's draft applies here. The skill drafts; you edit; nothing publishes that you haven't read out loud.

If you write for a more technical audience, the same chain adapts — there's a companion piece on skills for technical writers that swaps the social skills for docs-focused ones.

FAQ

Do I need Claude Code, or does this work in the Claude app? The skills-folder approach is built for Claude Code and the agent runtime, which read ~/.claude/skills/. The chat app doesn't load local skill folders the same way, so if you want this exact setup, Claude Code is the place to run it.

How long does one full run take? After the skills are installed, a single blog post through all five stages takes a few minutes of generation plus your editing time. The editing is the real cost, and it's worth it — that's the pass that keeps the output from sounding machine-made.

Can I chain my own skills instead of the ones listed here? Yes. The table is a starting set, not a requirement. Any skill with a clear SKILL.md will slot into the chain. Write a skill for your own newsletter format and Claude will use it the same way it uses a shared one.

Won't all the variants sound the same since they come from one post? They share an argument, which is the point, but the formats pull them apart. A 7-tweet thread, a 400-word email, and a 60-second script each force different choices. The step-6 brand pass plus your own edit keep them distinct rather than copy-pasted.

Is this just spamming the same idea everywhere? Only if you let it. Repurposing done badly is one post broadcast five times. Done well, it's one idea argued five different ways for five different rooms. The difference is entirely in the editing pass you do at the end.