Claude Code Team Publicly Shares Internal Skill Usage: Hundreds in Use, Categorized into 9 Major Types

According to 1M AI News monitoring, Thariq, a member of the Anthropic programming tool Claude Code team, wrote a post sharing the team's experience with internal Skill usage. Skills are the most commonly used extension mechanism in Claude Code, with currently hundreds of active ones within Anthropic. The team has categorized them into nine types:

1. Library and API Reference: Teaching the Agent how to correctly use internal libraries, CLI, and SDK, with code snippets and common pitfalls
2. Product Validation: Using tools like Playwright and tmux to validate code output, able to record videos as evidence, and insert assertions at each step
3. Data Retrieval and Analysis: Connecting to monitoring systems, built-in credentials, dashboard IDs, and commonly used query templates
4. Business Process Automation: One-command completion of repetitive tasks such as standup summaries, ticket creation, and weekly report generation
5. Code Scaffold: Generating project templates according to team standards, suitable for scenarios with natural language requirements that cannot be purely covered by code
6. Code Quality and Review: Can be run automatically as a hook or GitHub Action, including spawning child Agents for adversarial reviews
7. CI/CD and Deployment: Monitoring PRs, retrying CI, progressive deployment, and automatic rollbacks
8. Runbook: Troubleshooting from alerts or error signatures through multi-tool collaboration, outputting structured reports
9. Infrastructure Operations: Operations with safety guardrails, such as orphaned resource cleanup, dependency approvals, and cost investigations

The article emphasizes several key practices: a Skill is not just markdown but an entire folder that can contain scripts, assets, and data, leveraging the file system for progressive disclosure; the most valuable content is the Gotchas section, accumulating failure patterns of the Agent continuously; the description field of a Skill is a trigger condition for the model, not a human-readable summary; historical logs can be stored within the Skill directory for cross-session memory. The team measures the activation rate and effectiveness through Skill usage data recorded by the PreToolUse hook.