Anthropic launched Agent Skills for Claude, a system that lets you package instructions, scripts and resources into folders that Claude can load when it needs them.
Skills work across Claude’s products, the web app, Claude Code and the API. You build them once and use them everywhere.
What Skills Actually Are
A Skill is a folder that contains instructions and resources for specific tasks. Think of it as a reference guide that Claude consults when working on something that matches the skill’s domain.
Claude scans available skills while working. When it finds a relevant match, it loads only what it needs for that task. You don’t select skills manually. Claude identifies which ones apply and uses them automatically.
The system is built on four principles:
- Composable: Multiple skills can work together. Claude coordinates them without requiring you to manage dependencies.
- Portable: Skills use the same format everywhere. A skill you create for the web app works in Claude Code and the API without modification.
- Efficient: Claude loads minimal information from each skill. This keeps responses fast while accessing specialized knowledge.
- Powerful: Skills can include executable code for tasks where traditional programming produces more reliable results than token generation.
Where Skills Work
Skills are available across three Claude products, each with different implementation details.
Claude Web App
Pro, Max, Team and Enterprise users can access Skills. Anthropic provides built-in skills for common tasks, such as document creation. You can also create custom skills or modify examples.
Creating a skill is straightforward. Claude includes a “skill-creator” skill that walks you through the process. It asks about your workflow, generates the folder structure, formats the required SKILL.md file, and bundles your resources. No manual file editing needed.
Team and Enterprise users need admin approval before using Skills organization-wide.
API
The Messages API now supports Skills through a new /v1/skills endpoint. Developers get programmatic control over skill versioning and management.
Skills require the Code Execution Tool beta, which provides the secure environment they need to run. This is a hard requirement, not optional.
Anthropic provides pre-built skills for generating Excel spreadsheets with formulas, PowerPoint presentations, Word documents, and fillable PDFs. Developers can create custom skills for their specific use cases.
Skill management happens through the Claude Console, where you can create, view, and upgrade skill versions.
Claude Code
Skills extend Claude Code with team expertise and workflows. You install them as plugins from the anthropics/skills marketplace. Claude loads them automatically when they’re relevant to your current task.
Teams can share skills through version control. You can also install skills manually by adding them to ~/.claude/skills. The Claude Agent SDK provides the same Skills support for building custom agents.
Why DevOps Teams Should Care
DevOps work involves specialized tools, company-specific processes, and infrastructure patterns that generic AI assistants don’t understand well.
Skills change that. You can package your deployment procedures, infrastructure templates, monitoring setup, and incident response workflows into skills that Claude references when needed.
A few practical applications:
Infrastructure as Code: Create a skill that includes your Terraform or CloudFormation patterns, naming conventions, and tagging requirements. When Claude helps you write infrastructure code, it follows your standards automatically.
CI/CD Pipelines: Package your pipeline templates, testing requirements, and deployment gates into a skill. Claude can then help build or troubleshoot pipelines that match your actual setup.
Runbook Automation: Convert operational runbooks into skills. When incidents happen, Claude can reference your documented procedures and help execute them.
Compliance and Security: Embed security requirements, compliance checks, and approved patterns into skills. Claude will apply them consistently across different tasks.
The composability matters here. You might have separate skills for security, infrastructure, and monitoring. Claude can use all three together when helping you deploy a new service.
Mitch Ashley, vice president and practice lead, software lifecycle engineering, The Futurum Group, believes, “Anthropic’s Agent Skills is the practical realization of the agentic development model. AI systems no longer just generate code; they plan, coordinate, and act based on context and intent. By packaging domain knowledge, best practices, and automation into reusable skills, Anthropic is creating a foundation for agent-orchestrated workflows across development and operations. It is another clear step toward intent-driven, outcome-based Intent-Driven Development, what I call ‘human driving the loop.’”
Getting Started
The barrier to entry is low. If you’re using Claude’s web app, enable Skills in Settings and try the skill-creator. Start with something simple—maybe your team’s Git workflow or common kubectl commands.
API developers should check the documentation and explore the /v1/skills endpoint. The Code Execution Tool beta requires you to enable it first.
Claude Code users can browse the anthropics/skills marketplace for existing skills before building custom ones.
One important note: Skills execute code in Claude’s environment. That power comes with responsibility. Stick to trusted sources. Don’t install skills from unknown developers or unverified sources.
What’s Missing
This is an early release. Anthropic is working on simplified creation workflows and enterprise-wide deployment tools.
Right now, distributing skills across large teams requires manual coordination or version control systems. Enterprise admins will likely want centralized management and approval processes that don’t exist yet.
The documentation is solid but still evolving. Expect more examples and best practices to emerge as developers experiment with the system.
The Bigger Picture
Skills represent a shift in how we think about AI assistance. Instead of cramming everything into prompts or expecting the AI to magically understand your context, explicitly package that context into reusable components.
For DevOps teams dealing with complex, organization-specific workflows, explicit knowledge packaging makes AI tools significantly more useful. You’re not fighting the AI’s generic responses. You’re teaching it your way of working.
Test it on a low-stakes task first. If it saves time, expand gradually. Portability means your investment can transfer across different Claude products as your needs change.