Anthropic explained how to build skills for Claude Code and why teams need their own marketplace

Anthropic has published a detailed guide on skills for Claude Code — extensions that help the agent work better with code, data, and internal processes. The company says it's already using hundreds of such skills and has found through practice which ones actually improve work quality and which ones just bloat the context.

What skills work

The main point of the guide: a skill isn't just a markdown file with instructions, but a full-fledged folder with configs, scripts, templates, data, and even hooks. This is why Anthropic recommends first determining the skill type, then writing its content. According to the company's observations, the best skills usually solve one clear task. Those that try to be a reference, autopilot, runbook, and code generator all at once more often confuse the model than help it.

• API, SDK, and internal library references
• Skills for product verification through browser, CLI, and assertions
• Tools for data analysis, metrics, and monitoring
• Templates for scaffolding, tests, and code review
• Skills for CI/CD, runbooks, and infrastructure operations

AnthropIC separately highlights verification and operational skills. In the first case, the agent doesn't just write code, but runs the scenario itself, checks the steps, and records the result. In the second case, it gets safeguards for sensitive actions like resource cleanup, deployment, or alert triage. Such skills are especially valuable where errors are costly: in production, payments, onboarding, and internal platforms. These scenarios best show whether a skill turns the model into a reliable working tool.

How to write a skill

Anthropic recommends not wasting space on obvious things that Claude already knows about code and typical practices. It's much more useful to gather in a skill a section with real "gotchas": typical failures, non-trivial edge cases, and local team rules. Another important principle is gradual context disclosure through the file system. API details can be moved to references, templates to assets, and repetitive logic to scripts. This way the model only reads what it needs for the current step.

"The best way to understand skills is to start, experiment, and see

what works for you."

The authors also recommend planning the initial setup in advance: for example, storing parameters in config.json and asking the user structured questions if the config is empty. The description field should be written not as a marketing summary, but as a trigger condition for the model itself. For memory, you can use logs, JSON, or SQLite, but persistent data is better kept in a stable directory like `${CLAUDE_PLUGIN_DATA}`. Plus, Anthropic separately promotes the idea of storing code and helper libraries inside skills rather than forcing the model to reinvent boilerplate each time. To simplify this process, the company even released Skill Creator.

How to share within a team

For distributing skills, Anthropic describes two basic scenarios: storing them directly in the project repository or packaging them as plugins for the Claude Code marketplace. The first option is convenient for small teams and a couple of repositories, but it has a cost: each committed skill adds context to the model. When there are many skills, it makes sense to move them to a separate internal catalog where teams choose what to install themselves. This is especially important if different projects live in different codebases and teams have different needs.

That said, Anthropic doesn't have one central team that manually approves all skills from the start. Usually, useful developments first appear locally, then are shared via GitHub and Slack, and after organic demand, they're promoted to the marketplace. Usage can be measured through PreToolUse hooks, and skill dependencies are currently resolved by simple name-based links. The idea is to grow the library gradually, without cluttering the system with duplicates and poorly maintained extensions.

What it means

Anthropic's guide shows an important shift: the value of AI tools for development increasingly lies not just in the model, but in how the team packages its knowledge, checks, and safe processes around it. For Claude Code users, this is a signal to invest not in "magic prompts," but in a custom skills layer that makes the agent more predictable, useful, and closer to real work tasks in the engineering environment.