How to Automate Content Repurposing With Claude Code Skills

The Problem With Manual Content Repurposing

You record a 30-minute YouTube video. The ideas are good. The delivery is solid. And then it sits there, doing roughly one-tenth of the work it could be doing — because turning that transcript into a LinkedIn post, an X thread, and a newsletter section takes another two to three hours you don’t have.

Most content creators already know this. The bottleneck isn’t ideas or production. It’s distribution. Specifically, the manual labor of adapting one piece of content into formats each platform actually rewards.

Claude Code Skills fix this. With the right skill setup, you drop in a YouTube transcript and get back a LinkedIn post, an X thread, and a newsletter draft — all written in your brand voice, formatted for each platform’s conventions, ready to edit or publish. This guide walks through exactly how to build that skill, from file structure to brand context to the actual process logic.

What Claude Code Skills Are (Quick Version)

If you’re new to this, a Claude Code Skill is a structured, reusable agent task. It’s not a prompt. It’s a defined process that Claude follows consistently every time it runs — with its own file structure, reference context, and step-by-step instructions.

The key distinction: a prompt gives Claude general direction. A skill gives Claude a repeatable workflow with guard rails, brand context, and predictable outputs.

For content repurposing, that distinction matters a lot. You don’t want Claude improvising a new tone every time. You want it to follow the same voice, the same formatting rules, and the same quality bar — run after run.

What You’ll Build

This skill takes a single input — a raw YouTube transcript — and outputs three distinct content pieces:

LinkedIn post — A 150–300 word professional post with a hook, core insight, and a call to action. Formatted with line breaks for readability.
X thread — 6–10 tweets, each under 280 characters, structured to build narrative momentum and end with a strong closer.
Newsletter section — A 300–500 word section formatted as a standalone email block, suitable for dropping into a weekly digest or dedicated send.

All three are written in your brand voice — not generic AI content. That’s the part most people skip, and it’s the part that makes the difference between output you can actually use and output you have to rewrite anyway.

Prerequisites

Before building, make sure you have:

Claude Code access with skill creation enabled
A working brand-voice.md file (or brand context you can document — more on this below)
A YouTube transcript to test with (exported manually, via a tool like Tactiq, or from YouTube’s built-in transcript export)

You don’t need prior coding experience for this. The skill process logic is written in plain steps. Where code snippets appear, they’re simple and annotated.

Step 1: Set Up Your File Structure

A well-organized skill separates the process steps from the reference context. This matters more than it sounds. If you mix everything into one file, Claude has to do cognitive work sorting out what’s instruction versus what’s background. Keeping your skill.md focused on process steps — and storing brand context in separate reference files — produces more consistent output.

Here’s the directory structure for this skill:

/content-repurposing-skill/
├── skill.md
├── brand-voice.md
├── platform-formats.md
└── examples/
    ├── linkedin-example.md
    ├── twitter-example.md
    └── newsletter-example.md

skill.md — The process file. Contains the numbered steps Claude follows. Nothing else.

brand-voice.md — Your brand’s tone, vocabulary, topics it does and doesn’t discuss, and any style rules.

platform-formats.md — Specific formatting requirements and conventions for each output format.

examples/ — Optional but valuable. Real examples of good posts from your brand that Claude uses as format anchors.

Step 2: Write Your Brand Voice File

This is the most important piece and the one most people rush. A weak brand voice file produces generic output. A specific one produces content that actually sounds like you.

Your brand-voice.md should cover:

Tone and Personality

Be specific. “Professional but approachable” tells Claude almost nothing. Instead:

Tone: Direct and practical. No corporate speak. Speaks to operators and builders, not executives looking for reassurance.

Avoid: Buzzwords like "leverage," "synergy," "ecosystems." No rhetorical questions that don't get answered.

Voice: First person ("I" and "we"). Never third person ("the company believes...").

Vocabulary and Phrasing

List specific words and phrases you use — and ones you avoid:

Use: "build," "ship," "run," "test," "workflow"
Avoid: "implement," "utilize," "operationalize," "best-in-class"

Audience Context

Who reads your content and what do they care about?

Audience: Founders, content leads, and operators running lean marketing teams. They're time-constrained and skeptical of hype. They want specific, usable information.

Topics and Positioning

What do you stand for, and what do you stay away from?

Core topics: AI workflows, content automation, productivity systems
Avoid: Political commentary, financial advice, comparisons that punch at competitors by name

The more concrete this file is, the less variance you’ll see in the output. This is the closest thing to teaching Claude how to write like you.

Step 3: Write Your Platform Formats File

Each platform has conventions that affect whether content performs. Your platform-formats.md file codifies those rules so Claude applies them automatically.

Here’s a working version:

# LinkedIn

- Length: 150–300 words
- Hook on the first line (no context-setting before the hook)
- Use line breaks between paragraphs — no walls of text
- Avoid hashtags in body copy; if used at all, 1–2 at the end
- End with a direct CTA or a genuine question that invites comments
- First line must stand alone — it's what shows before "see more"

# X / Twitter Thread

- First tweet: standalone hook that works without context
- Tweets 2–8: one idea per tweet, direct and specific
- Each tweet under 280 characters
- No fillers like "Let me explain" or "Here's a thread:"
- Final tweet: call to action or memorable summary statement
- Thread should feel like a conversation, not a slide deck

# Newsletter Section

- Length: 300–500 words
- Written in second person ("you") or first person — not both
- Structured: context → insight → implication or action
- One key takeaway per section
- No excessive linking — max 1–2 links per section
- Subheadings optional but helpful for longer sections

Step 4: Write the Skill Process (skill.md)

This is where the actual logic lives. The process should be sequential and unambiguous. Claude follows it top to bottom. Writing code scripts rather than vague markdown instructions produces more reliable results, but for a content skill, structured prose steps work well as long as they’re specific.

Here’s a working skill.md for this use case:

# Skill: YouTube Transcript → Multi-Platform Content

## Input
- `transcript`: Full YouTube video transcript (raw text)

## Reference Files
- brand-voice.md: Brand tone, vocabulary, and positioning
- platform-formats.md: Format requirements for each output
- examples/: Sample posts for format reference (if present)

## Process

### Step 1: Extract Key Ideas
Read the full transcript. Identify:
- The central argument or main point (1 sentence)
- 3–5 supporting ideas or insights
- Any specific data points, stories, or examples worth preserving
- The actionable takeaway, if one exists

Do not paraphrase yet. Just extract and list.

### Step 2: Write the LinkedIn Post
Using brand-voice.md for tone and platform-formats.md for structure:
- Write a hook based on the most compelling idea from Step 1
- Develop the post using 2–3 of the supporting ideas
- End with a CTA or question
- Review: Does the first line work as a standalone hook? Is it under 300 words?

### Step 3: Write the X Thread
Using brand-voice.md for tone and platform-formats.md for structure:
- Tweet 1: Hook (standalone)
- Tweets 2–8: One insight per tweet from the Step 1 list
- Final tweet: CTA, summary, or question
- Review: Is every tweet under 280 characters? Does each stand alone?

### Step 4: Write the Newsletter Section
Using brand-voice.md for tone and platform-formats.md for structure:
- Write a short intro that sets context (1–2 sentences)
- Develop the main insight from the transcript (the "so what")
- Add a practical implication or next step
- Review: Is it 300–500 words? Does the section read as self-contained?

### Step 5: Output
Return all three pieces, clearly labeled:
- LINKEDIN POST
- X THREAD
- NEWSLETTER SECTION

Each separated by a horizontal rule. No meta-commentary about what was written.

This is intentionally plain. The skill doesn’t need complexity — it needs clarity. Claude executes these steps reliably because each one has a clear input, process, and check.

Step 5: Add Examples (Optional but Worth It)

The fastest way to improve output quality is to show Claude what “good” looks like for your brand. Drop a few real examples into your /examples/ folder and reference them in the skill.

A quick format note: you don’t need to annotate the examples or explain what makes them good. Claude picks up on patterns. A set of 3–5 examples per format is enough to anchor tone and structure.

This is especially valuable for the LinkedIn format, where the hook style varies dramatically between voices. If you tend toward story-led hooks, your examples will teach that. If you lead with counterintuitive data, that’ll come through too.

Step 6: Run It and Evaluate the Output

Once the skill is configured, test it with a real transcript. The first run is diagnostic. You’re not looking for perfection — you’re looking for where the output diverges from what you’d actually post.

Common issues and fixes:

Problem: Output sounds generic, not like you. Fix: Review brand-voice.md. Add more specific vocabulary rules. Add better examples.

Problem: LinkedIn post is too long or has hashtag spam. Fix: Review platform-formats.md. Be more explicit about the length cap and hashtag rules.

Problem: X thread is too formal. Fix: Check that brand-voice.md includes a note about conversational tone in threads. X is typically more casual than LinkedIn.

Problem: Newsletter section reads like a blog post, not an email. Fix: Add a note to platform-formats.md about email-native tone: shorter paragraphs, more direct language, written to one person not an audience.

If you want to go further with systematic improvement, building a learning loop into your skill lets Claude record what worked and what didn’t, then apply those learnings automatically on future runs.

Step 7: Chain It Into a Larger Workflow

A single content repurposing skill is useful. Chaining it into a larger workflow is where things get genuinely efficient.

A natural extension for this skill is a content calendar pipeline:

Transcript Ingestion Skill — Pulls the transcript from YouTube automatically using a URL input.
Content Repurposing Skill (this one) — Generates LinkedIn, X, and newsletter content.
Scheduling Skill — Pushes content into a scheduler like Blotato or a project management tool for review and approval.

Chaining skills into end-to-end workflows like this means the whole pipeline runs from a single trigger — a new YouTube video URL — with no manual steps in between.

You can also connect this to a broader content marketing skill system where multiple skills share the same brand context and compound each other’s value over time.

Making Brand Context Consistent Across Skills

One trap people fall into: they configure brand voice for one skill, then rebuild it from scratch when they add a second. That’s avoidable.

The right approach is a shared business brain — a central brand context file that every skill references. The business brain pattern means your newsletter skill, your LinkedIn skill, and your X skill all draw from the same source of truth. You update the brand voice once, and every skill stays in sync.

This becomes especially important as your content operation grows. If you’re running multiple skills across multiple platforms, having two distinct memory layers — a shared brand context and a skill-specific context folder — keeps output consistent without requiring you to micromanage each skill individually.

Where Remy Fits

If you want to take this further — building a full web application around this workflow — Remy is worth looking at.

Remy compiles annotated markdown specs into full-stack applications: backend, database, auth, deployment. The content repurposing skill described in this guide is a Claude Code workflow. But if you wanted to wrap it in a proper product — a content dashboard where your team submits video URLs, reviews AI-generated drafts, approves them, and pushes to a scheduler — that’s a full-stack app.

Building that app from scratch means wiring up a frontend, a backend API, a database to store drafts, auth for team access, and deployment. With Remy, you’d describe what the app does in a spec, and the code compiles from that. The spec stays in sync as the product evolves.

You can try Remy at mindstudio.ai/remy.

Common Questions

Can this skill work with transcripts from sources other than YouTube?

Yes. The skill takes raw transcript text as input. As long as you can get a text transcript — from a podcast, a Zoom recording, a webinar export — it’ll work the same way. YouTube is just the most common source for long-form video content.

How do I get a YouTube transcript?

YouTube has a built-in transcript feature. Go to any video, click the three-dot menu under the video, and select “Show transcript.” Copy the full text. Tools like Tactiq and Whisper can also auto-generate transcripts or export them in cleaner formats.

How do I keep the output consistent across multiple runs?

Consistency comes from the reference files, not the process steps. If you’re seeing variance between runs, improve your brand-voice.md and platform-formats.md. More specific rules produce less variance. Adding real examples helps too.

Can I add more output formats to the skill?

Yes. Add a new section to platform-formats.md for the format you want (e.g., Instagram caption, blog intro), then add a corresponding step to skill.md. The structure scales cleanly. If you’re adding many formats, consider splitting into separate platform-specific skills and chaining them together rather than putting everything into one long process file.

What if the transcript is very long?

Transcripts from 30–60 minute videos can be 8,000–15,000 words. That’s fine for Claude — it handles long context well. If you notice the output drifting toward the end of a long transcript, add a Step 1 note instructing Claude to prioritize the first and last 20% of the transcript for key ideas, since that’s typically where the core argument lives.

How is this different from just prompting Claude directly?

A direct prompt produces one-off output. A skill produces repeatable, consistent output that maintains your brand voice, follows your formatting rules, and can be triggered programmatically. Prompts degrade in quality when you’re inconsistent about what you include. Skills are stable because the instructions and context live in structured files that don’t change unless you change them. That’s the practical difference.

Key Takeaways

A content repurposing Claude Code skill takes a YouTube transcript and outputs a LinkedIn post, X thread, and newsletter section automatically.
The quality of the output depends almost entirely on your brand voice and platform format reference files — not the process steps themselves.
Separating process logic (skill.md) from reference context (brand-voice.md, platform-formats.md) produces more consistent results than mixing everything together.
Examples are the fastest quality lever — show Claude what good looks like for your specific voice.
This skill can chain into a larger pipeline: transcript ingestion → content generation → scheduling and publishing.
A shared brand context file keeps output consistent if you’re running multiple skills across platforms.

If you want to go beyond skills into a full web application — a content dashboard, a team approval workflow, a proper publishing system — Remy lets you build that from a spec rather than from scratch.