A Claude Code skill is a SKILL.md file that lives in a named directory. Inside, you write YAML frontmatter that tells Claude when to use the skill, followed by markdown instructions that tell Claude what to do. When Claude detects a request matching the skill’s description, it loads those instructions automatically. Or you can invoke it directly by typing /skill-name.
Think of skills as saved playbooks. Instead of rebuilding your prompting context from scratch every session, you define it once and Claude applies it when relevant. The skill’s body loads only when it’s used, so long reference material costs almost nothing until you need it.
This isn’t just a convenience feature. It’s a fundamental shift in how you interact with AI. You stop being a prompt-writer and start being a workflow-designer. The AI doesn’t just respond—it follows a repeatable, auditable process that you defined.
Anthropic built Claude Code skills on the Agent Skills open standard, which means the same skill format works across multiple AI tools including Gemini CLI, OpenCode, and OpenHands. Write once, use everywhere—that’s the promise, and it’s already happening. For the full technical documentation, see Anthropic’s official skills guide.
Every skill needs two parts: YAML frontmatter between --- markers, and markdown body content with instructions. The frontmatter tells Claude when to use the skill. The body tells Claude what to do when it fires.
Start with a skill that solves your most annoying repetition. If you paste the same style guide into every session, that’s your first skill. If you manually run the same research queries before every article, that’s your second. Solve the pain you actually feel, not the pain someone else invented.
Here’s what a content marketing skill looks like in practice. This example creates a skill that checks any draft against your brand voice rules before publishing:
———
description: Reviews content against brand voice guidelines. Use when the user asks for a voice check, style review, or before publishing any draft.
———
## Brand Voice Guidelines
– Active voice. No hedging (might, could potentially, may help).
– Short paragraphs: 2-4 sentences max.
– Contractions required: don’t not do not.
– No corporate-speak: cut leverage, utilize, unlock, supercharge.
– Every claim backed by data or cut.
## Instructions
Read the provided draft. Flag every violation of the guidelines above. For each flag, suggest the specific fix. Do not rewrite the entire piece — flag and fix individual violations. If no violations found, confirm the draft is clean.
Save that file to ~/.claude/skills/brand-voice/SKILL.md and Claude loads it automatically when you ask for a voice check. Or invoke it directly with /brand-voice. The skill works across all your projects—personal skills live in your home directory and apply everywhere.
Where you store a skill determines who can use it. Personal skills (~/.claude/skills/) apply to all your projects. Project skills (.claude/skills/) stay in a specific repo and travel with the codebase. Enterprise skills deploy across your entire organization through managed settings. Plugin skills bundle with third-party tools. This hierarchy means you can have company-wide editorial standards that override individual preferences, while still letting teams build project-specific workflow skills.
Here are four skills every content team should build first, based on the workflows that consume the most manual effort in a typical content pipeline:
These four skills cover the core content operations loop: research → write → edit → optimize → distribute. Build them once, use them for every piece of content, and you’ve automated the most time-consuming part of content marketing without any code.
Dynamic context injection lets a skill run a shell command before Claude sees the instructions, replacing the command with its output. In the skill file, you write !`command` and Claude Code executes it before the skill loads. For a content team, this means a research skill can pull the latest headlines from an RSS feed, grab real-time analytics data, or check what’s already been published—all before Claude starts reading the skill. The context is always fresh, never stale.
Subagent execution lets skills spawn isolated sub-agents to handle complex tasks without polluting the main session context. Think of it as delegation: the skill tells Claude to create a sub-agent with specific instructions, hand it a chunk of work, and report back when done. For content operations, this means you can run research, editing, and SEO optimization in parallel—each in isolated sessions that don’t interfere with each other.
Used together, these features let you build skills that are genuinely autonomous. A publishing skill could: spawn a sub-agent to research the topic, another to draft the article, a third to run editorial review, and a fourth to handle SEO metadata—all from one command. Each sub-agent runs in isolation with its own context window, so the main session stays clean and responsive.
Anthropic also bundles several skills with Claude Code that demonstrate these patterns, including /code-review, /batch, and /loop. These are prompt-based—they give Claude detailed instructions and let it orchestrate the work. Content marketers can study these bundled skills as reference implementations for their own workflows.
You publish a strong piece on your main blog. Two weeks later, your newsletter goes out with a completely different topic because nobody connected the dots. Your social team doesn’t promote it because they didn’t know it existed. That article—the one you spent six hours on—gets 40 views. Fix the coordination, and the same article gets your newsletter audience, your social following, and cross-links from your other properties.
A Multi-Property Content Orchestrator skill maintains a living inventory of everything you’ve published across properties. It tracks topic coverage by property, flags cross-linking opportunities, and prevents duplicate coverage. When you draft a new article, the skill checks: did we cover this elsewhere? Can we link? Should this be a follow-up instead?
--- description: Orchestrates content across multiple properties. Use before drafting, when planning editorial calendar, or when checking for cross-linking opportunities. tools: [Read, Glob, Grep] --- ## Property Inventory - Property A (Primary Blog): /path/to/content/inventory-a.json - Property B (Industry Site): /path/to/content/inventory-b.json - Newsletter: /path/to/newsletter-archive.json ## Instructions 1. When given a topic, search all property inventories for existing coverage. 2. Flag any post on the same topic across properties (potential duplicate). 3. Identify related posts for cross-linking (≥2 recommendations). 4. Check category/pillar balance: which topic areas are under-covered? 5. Output: coverage map, cross-link candidates, and gap recommendations.
The real value of this skill isn’t in the individual links—it’s in the system it creates. Over six months of consistent use, your properties stop being isolated silos and become an interconnected content network. Google notices. Readers notice. And your team stops accidentally writing the same article twice.
Editors get tired. They skim. After the third draft of the day, the hedges start looking normal and the em dashes stop registering. An automated first pass catches what tired eyes miss, and it catches the same things every time—no variability, no fatigue.
This skill runs a structured audit against any draft. It scans for: em dashes (the number one AI fingerprint), hedging language (might, could potentially, may help, tends to), corporate-speak (leverage, utilize, unlock, supercharge), formulaic paragraph structure, and zero-data claims.
--- description: Audits content drafts for AI-generated patterns and voice inconsistencies. trigger: /voice-audit --- ## AI Pattern Detection Rules ### Immediate Flags (Auto-Reject) - Em dashes in body text - "In today’s rapidly evolving landscape" - "Furthermore," "Additionally," "Moreover" as transitions - Lists of 3-5 generic benefits with no specifics ### Soft Flags (Review Required) - Hedging: might, could potentially, may help, tends to - Corporate-speak: leverage, utilize, unlock, supercharge - Overly balanced structures: "On one hand... on the other hand..." - Placeholder examples: "For instance, a SaaS company might..." ### Structural Flags - 3+ consecutive paragraphs following topic-example-explanation pattern - Zero personal anecdotes or specific data points - No contrarian angle or debatable thesis ## Instructions 1. Read the provided draft in full. 2. Run each detection rule. Flag every match with line/paragraph reference. 3. Categorize: Auto-Reject (must fix), Soft Flag (review), Structural. 4. Score the draft 0-10 on voice authenticity. Target: 8+ before publish. 5. Output: flagged items with specific fix suggestions.
Build this skill once, run it before every publish. Over time, your writers internalize the patterns the auditor catches, and the flags decrease. That’s the real win: the skill trains your team, not just your content.
Raw research is noise. You pull data from industry reports, scan competitor coverage, check Reddit and social for practitioner takes, and search for contrarian viewpoints. You end up with 20 tabs open and a Notion page full of bullet points. The synthesis engine takes that chaos and produces: a clear article angle, the three strongest supporting data points, a contrarian hook, and a practitioner example that grounds the piece in reality.
The skill doesn’t just collect—it evaluates. Each source gets a quality tier. Tier 1 sources (original research, official documentation, named practitioner case studies) get weighted higher than Tier 3 sources (aggregator roundups, unsourced claims, AI-generated summaries). The brief includes source grading so writers know what’s solid and what needs verification.
---
description: Synthesizes raw research into structured, angle-driven content briefs.
trigger: /research-synthesis
tools: [Read, Write, WebSearch, WebFetch]
---
## Source Quality Tiers
- Tier 1: Original research, official docs, named case studies
- Tier 2: Industry publications, analyst reports, credible news
- Tier 3: Aggregator roundups, unsourced claims, AI summaries
- Tier 4: Unknown blogs, forum posts (flag, don’t cite)
## Brief Structure
1. **Angle Statement** (1 sentence): What makes this piece worth reading?
2. **Three Supporting Points**: Ranked by source quality and specificity
3. **Contrarian Hook**: The counter-narrative or surprising finding
4. **Practitioner Anchor**: A real example that grounds the piece
5. **Source Map**: All sources with tier, key quote, and URL
6. **Gap Warning**: What the research didn’t find
## Instructions
1. Read all provided research files and source URLs.
2. Grade every source on the tier system above.
3. Draft the angle statement. Reject generic angles ("X is important").
4. Select the 3 strongest supporting points backed by Tier 1-2 sources.
5. Identify the contrarian hook—if none exists, flag as incomplete.
6. Produce the practitioner anchor from best real-world example.
7. Write the gap warning explicitly: what data does not support.
8. Output the complete brief in the structure above.
Build this skill after you have the Research Brief Generator working. The generator handles discovery; the synthesis engine handles the thinking. Together, they turn your pre-writing process from a two-hour research session into a ten-minute automated pipeline that produces better briefs than most humans can in two hours.
For the complete picture on building AI-powered content operations, see Claude Code for Content Operations and The Agentic Content Era.
The content marketing stack of 2026 has too many tools and not enough process. You have a CMS, a design tool, an analytics platform, a social scheduler, an email platform—each doing its job but none talking to each other. Between them sits a human, manually bridging the gaps. Skills automate those bridges.
Start small. Build one skill that fixes one friction point. The brand voice checker from earlier is a perfect starting point. Once that’s working, add the research brief generator. Then the editorial pipeline. Over a month of consistent use, you’ll have a library of skills that handle the repetitive parts of content operations, freeing your team for the work that actually requires human judgment: strategy, voice, and creative direction.
For more on building AI-powered content engines, see our deep dive on Claude Code for Content Operations. If you’re thinking about the bigger picture of how AI agents reshape content teams, The Agentic Content Era covers where this is all heading.