Repeatable expert workflows
Teams that repeatedly teach an agent the same process, standards, examples or quality checks.
Independent tool overview
Agent Skills at a glance
Agent Skills are portable folders of instructions, scripts and resources that give compatible AI agents reusable expertise and repeatable workflows.
Visit the official Agent Skills site ↗
- Created by
- Anthropic; now published as an open standard
- Core file
- SKILL.md with name and description metadata
- Best for
- Reusable, version-controlled agent workflows
- Claude support
- Claude.ai, Claude Code and Claude API
Overview
What Agent Skills is
Anthropic introduced Agent Skills as a way to package procedural knowledge in a folder centered on a SKILL.md file. An agent initially sees lightweight metadata, loads the full instructions when a task matches, and can then read supporting resources or run bundled scripts as needed. This progressive-disclosure model lets an agent keep many specialized workflows available without placing every instruction into every prompt.
Agent Skills are now an open standard rather than a Claude-only file format. Anthropic supports them across Claude.ai, Claude Code and the Claude API, and publishes a reference repository with example and document skills. In August 2026, Anthropic moved Agent Skills and the Skills API out of beta on the Claude API.
Use cases
Who Agent Skills is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Organization-specific guidance
Companies packaging brand rules, operating procedures and domain knowledge into version-controlled resources.
Document and technical automation
Users who need instructions plus scripts, templates or reference files for outputs such as spreadsheets, slides, PDFs or code.
Capabilities
Core Agent Skills features
Progressive disclosure
Loads only a skill's identifying metadata at discovery time, then reads its full instructions and resources when relevant.
Simple folder format
Uses a SKILL.md file with required name and description fields, with optional scripts, references, examples and assets alongside it.
Automatic activation
Compatible agents can select a skill from its description when the current task matches, while supported clients may also allow manual invocation.
Scripts and resources
Packages executable helpers, templates and supporting documentation with the workflow instead of relying on prompt text alone.
Versioning and API management
Claude's Skills API supports custom skill creation and versions, while production integrations can pin a specific version for stability.
Open-standard portability
The format is published independently so one skill can be reused across products that implement the Agent Skills specification.
Process
How the Agent Skills workflow works
Step 1
Define one bounded capability
Choose a repeatable task with a clear trigger, inputs, steps and definition of a successful output.
Step 2
Write the SKILL.md metadata
Give the skill a valid lowercase name and a precise description that explains what it does and when an agent should load it.
Step 3
Package instructions and resources
Put the core workflow in SKILL.md and add only the scripts, references, templates or examples needed for execution.
Step 4
Test activation and output
Try positive and negative prompts, inspect which resources load, verify generated files and check that the skill does not over-trigger.
Step 5
Review, version and distribute
Audit every bundled file and dependency, pin production versions, and publish only through a trusted repository or managed workspace.
Cost
Agent Skills pricing and free plan
The Agent Skills specification and Anthropic's public sample repository are free to use. There is no separate per-skill fee in Claude; access follows the relevant Claude plan. API use incurs normal model token charges and code-execution costs when a skill runs in Anthropic's container.
Open standard
Free
Author and store skills using the published file format.
- Portable SKILL.md specification
- Public templates and examples
- No paid registry required
Claude apps
Included with eligible plan
Use built-in or custom skills through supported Claude products.
- Availability depends on Claude product and plan
- Example skills are provided in Anthropic's repository
- No separate skill purchase listed
Claude API
Usage based
Use Anthropic or custom skills through the Skills and Messages APIs.
- Standard model input and output token charges
- Code execution includes 1,550 free organization-hours monthly
- Additional container execution is $0.05 per hour with a five-minute minimum
Pricing checked . Check current pricing at the source ↗
Assessment
Agent Skills strengths and limitations
Where it stands out
- Captures workflow knowledge once instead of repeating a long prompt in every conversation.
- Keeps instructions, scripts and reference material together in a version-controlled package.
- Progressive disclosure limits the up-front context cost of maintaining many available capabilities.
- The open specification reduces dependence on a single agent client.
What to consider
- A skill can contain powerful instructions and executable code, so installing one from an untrusted source creates meaningful security risk.
- Compatibility and behavior still depend on how each agent product implements the open standard and what tools it exposes.
- Descriptions control discovery, so vague metadata can cause a skill to activate too often or fail to activate when needed.
- Claude API skills run in a constrained code-execution container with no network access or runtime package installation.
- Agent Skills on the Claude API are not eligible for Zero Data Retention and follow Anthropic's standard retention policy.
Compare
Agent Skills alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Coding
Codex App
Use Codex skills and plugins when the target workflow centers on OpenAI's desktop, CLI or cloud agent surfaces.
Explore Codex App →Coding
Claude Code
Use Claude Code directly when you need an agentic coding environment and do not yet need to package a portable reusable skill.
Explore Claude Code →Business Operations
ChatGPT
Use custom GPTs when a hosted conversational assistant is more useful than a filesystem-based agent capability.
Explore ChatGPT →Questions
Agent Skills FAQs
What are Agent Skills?
Agent Skills are folders containing instructions, scripts and resources that compatible AI agents can discover and load for specialized tasks.
What goes in a SKILL.md file?
At minimum, SKILL.md contains YAML metadata with a lowercase skill name and a description, followed by the workflow instructions. Supporting files can live beside it.
Are Agent Skills only for Claude?
No. Anthropic created the format and supports it across Claude products, but published Agent Skills as an open standard for other compatible agents to implement.
How much do Agent Skills cost?
The open format and public sample repository are free. Claude product access follows the user's plan, while API usage incurs model charges and, when applicable, code-execution charges.
Are third-party Agent Skills safe?
Not automatically. A skill may include instructions, scripts and dependencies with access to an agent's tools, so users should inspect every file and install only from trusted sources.
Bottom line
Our Agent Skills verdict
Agent Skills are a practical standard for turning repeated prompting and scattered helper files into portable, testable agent capabilities. Their simplicity is the main advantage, but it also makes trust and review crucial: a useful skill is executable operational knowledge, and a malicious or careless one can inherit the reach of the agent that loads it.
Visit Agent Skills website ↗Similar Agents Tools
View all Agents AI tools →
Gemini Computer Use
Gemini 2.5 Computer Use - A specialized model to power agents that can interact with user interfaces
FlowithOS
FlowithOS - An agentic workspace that connects your knowledge, creation, and execution in a single flow
Petri
Petri - Anthropic's open-source agentic system for evaluating LLM safety
GitHub Agent HQ
Agent HQ - A single, unified workflow for developers to be able to orchestrate any agent, any time, anywhere
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.