Now, Claude can add "skills"!
Anthropic has just launched the Skills feature for Claude. This new capability allows Claude to dynamically load specialized instructions, scripts, and resources, transforming it from a general AI assistant into a professional assistant for various fields.
This feature upgrade will further change the way we customize AI.
What are Agent Skills?
Skills are essentially folders containing instructions, scripts, and resources that Claude can dynamically discover and load when needed to complete specific tasks.
Imagine it like preparing an onboarding guide for a new employee:
You no longer need to build scattered, customized Agents for each use case. Now, anyone can specialize their Agent with composable capabilities by capturing and sharing their procedural knowledge.
Core Features
Skills have several key features:
Composable: Skills can stack together, and Claude automatically identifies which ones are needed and coordinates their use.
Portable: Using the same format, build once and use in Claude apps, Claude Code, and API.
Efficient: Loads only what's needed when needed.
Powerful: Can include executable code for tasks where traditional programming is more reliable than token generation.
How It Works
Let's look at a real example of how Skills work: the PDF skill that supports Claude's recently launched document editing capability.
Skill Structure
The simplest Skill is a directory containing a SKILL.md file.
This file must start with YAML frontmatter containing required metadata: name and description. At startup, the Agent preloads the name and description of each installed skill into the system prompt.
This metadata is the first layer of progressive disclosure: it provides enough information for Claude to know when each skill should be used without loading everything into context.
The actual content of the file is the second layer of detail.
If Claude deems the skill relevant to the current task, it loads the skill by reading the full SKILL.md into context.
As skills become more complex, they may contain too much context for a single SKILL.md or context only relevant in specific scenarios. In these cases, skills can bundle additional files in the skill directory and reference them by name from SKILL.md.
In the PDF skill, SKILL.md references two additional files (reference.md and forms.md), which the skill author chose to bundle with the core SKILL.md.
By moving form-filling instructions to a separate file (forms.md), the skill author keeps the core skill lean, trusting Claude to read forms.md only when filling forms.
Progressive Disclosure Design
Progressive disclosure is the core design principle that makes Agent Skills flexible and scalable.
Like a well-organized manual starting with a table of contents, then specific chapters, and finally detailed appendices, Skills allow Claude to load information only when needed:
Agents with file system and code execution tools don't need to read the entire skill content into their context window for specific tasks.
In other words, the amount of context that can be bundled into a skill is effectively unlimited.
Skills and Context Window
The diagram below shows how the context window changes when a user message triggers a skill:
The operation sequence is as follows: Initially, the context window contains the core system prompt, metadata for each installed skill, and the user's initial message; Claude triggers the PDF skill by calling the Bash tool to read pdf/SKILL.md; Claude chooses to read the bundled forms.md file; Finally, Claude proceeds with the user's task after loading relevant instructions from the PDF skill.
Skills and Code Execution
Skills can also include code that Claude can decide to execute itself as tools.
Large language models excel at many tasks, but some operations are better suited for traditional code execution. For example, sorting a list via token generation is much more expensive than simply running a sorting algorithm.
Beyond efficiency, many applications require the deterministic reliability that only code can provide.
In this example, the PDF skill includes a pre-written Python script to read the PDF and extract all form fields.
Claude can run this script without loading the script or PDF into context. Since the code is deterministic, this workflow is consistent and repeatable.
Developing and Evaluating Skills
Anthropic provides practical guides for developing and testing Skills:
Start with Evaluation
Run Agents on representative tasks and observe where they struggle or need extra context to identify specific gaps in Agent capabilities. Then iteratively build skills to address these gaps.
Structure for Scale
When SKILL.md becomes unmanageable, split its content into separate files and reference them. If contexts are mutually exclusive or rarely used together, keeping paths separate reduces token usage. Code can serve as both executable tools and documentation. Make it clear whether Claude should run scripts directly or read them as reference into context.
Think from Claude's Perspective
Monitor how Claude uses your skills in real scenarios and iterate based on observations: note unexpected trajectories or over-reliance on certain contexts. Pay special attention to your skill's name and description, as Claude uses these to decide whether to trigger the skill for the current task.
Iterate with Claude
When working with Claude, ask it to capture successful methods and common errors into reusable context and code in skills. If it veers off track while using a skill to complete a task, ask it to self-reflect on what went wrong. This process helps discover what context Claude actually needs, rather than trying to predict in advance.
Security Considerations
Skills provide Claude with new capabilities through instructions and code.
While powerful, this means malicious skills could introduce vulnerabilities in environments using them, or instruct Claude to leak data and take unintended actions.
Anthropic recommends installing skills only from trusted sources. When installing from less trusted sources, thoroughly review before use. First read the contents of bundled files in the skill to understand its functionality, especially noting code dependencies and bundled resources like images or scripts.
Likewise, watch for instructions or code in skills that direct Claude to connect to potentially untrusted external network sources.
Full Platform Support
Claude Apps
Skills are now available to Pro, Max, Team, and Enterprise users.
Anthropic provides skills for common tasks like document creation, customizable examples, and the ability to create your own custom skills.
Claude automatically invokes relevant skills based on your task without manual selection. You can even see skills in Claude's chain of thought while it works.
Creating a skill is simple:
Anthropic's "skill-creator" skill provides interactive creation guidance: Claude asks about your workflow, generates folder structure, formats the SKILL.md file, and bundles needed resources.
No manual file editing required—just talk.
Claude Developer Platform (API)
Agent Skills can now be added to Messages API requests, with the new /v1/skills endpoint giving developers programmatic control over custom skill versioning and management.
Skills require the Code Execution Tool beta, which provides the secure environment needed for execution.
Skills created by Anthropic allow Claude to read and generate professional Excel spreadsheets with formulas, PowerPoint presentations, Word documents, and fillable PDFs.
Developers can create custom Skills to extend Claude's capabilities for their specific use cases.
Claude Code
Of course, there's Claude Code:
Skills let you extend Claude Code with your team's domain knowledge and workflows.
Install skills via plugins from the anthropics/skills marketplace, and Claude loads them automatically when relevant. Share skills with your team via version control. You can also manually install skills by adding them to ~/.claude/skills.
Claude Agent SDK provides the same Agent Skills support for building custom Agents.
User Reactions
Satvik (@satvikmaker) commented:
Building specialized Agents with folders and files is a game-changer. Keep it up Anthropic team. 🔥👑
Carlos Lebron (@carlosml) joked:
Agent Skills sound cool. Finally a way to get Claude to do my laundry
Vijay Krishna S (@vjk_2k5) pointed out:
Claude just doubled down on brute-force training methods and beat every other AI company on coding. Insane.
Enterprise Application Prospects
Several enterprises are already exploring the potential of Skills:
Box
Skills teach Claude how to handle Box content. Users can convert stored files into PowerPoint presentations, Excel spreadsheets, and Word documents that follow their organization's standards, saving hours of work.
Notion
With Skills, Claude collaborates seamlessly with Notion, getting users from question to action faster. Reduces prompt wrangling on complex tasks for more predictable results.
Canva
Leverage Skills to customize Agents and extend their capabilities. This unlocks new ways to bring Canva deeper into Agent workflows, helping teams capture their unique context and easily create stunning high-quality designs.
Rakuten
Skills streamline our management accounting and finance workflows. Claude handles multiple spreadsheets, captures key anomalies, and generates reports using our templates. What used to take a day, we now do in an hour.
Future
Anthropic is working on simplified skill creation workflows and enterprise-wide deployment capabilities, making it easier for organizations to distribute skills across teams.
In the coming weeks, they'll continue adding full lifecycle features for creating, editing, discovering, sharing, and using Skills. They're especially excited about Skills helping organizations and individuals share their context and workflows with Claude.
They'll also explore how Skills can complement Model Context Protocol (MCP) servers by teaching Agents more complex workflows involving external tools and software.
Looking further ahead, Anthropic even hopes to enable Agents to autonomously create, edit, and evaluate Skills, encoding their own behavioral patterns into reusable capabilities.
Skills is a simple concept with a correspondingly simple format. This simplicity makes it easier for organizations, developers, and end users to build custom Agents and give them new capabilities.
[1] Anthropic Engineering Blog: https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills
[2] Skills Announcement: https://www.anthropic.com/news/skills
[3] Documentation: https://docs.claude.com/en/docs/agents-and-tools/agent-skills/overview
[4] Cookbook: https://github.com/anthropics/claude-cookbooks/tree/main/skills
[5] Example Skills: https://github.com/anthropics/skills