AI Skills

Reusable prompt templates for AI coding assistants

AI skills are the missing layer between a raw AI assistant and a productive coding workflow. Instead of typing the same multi-step instructions every time you want to run tests, create a PR, or review code for security issues, skills package those workflows into reusable templates that any AI assistant can follow.

What are AI skills?

Skills are markdown files that teach AI assistants specific workflows. They live in tool-specific directories and are loaded automatically when the AI assistant starts a session:

• .claude/skills/ — skills for Claude Code, loaded as slash commands or context
• .cursor/skills/ — skills for Cursor, available as workflow templates
• Each skill is a structured prompt with clear steps, expected inputs, and success criteria — not a vague instruction but a precise playbook

Example skills

Skills cover the workflows developers repeat daily. Here are some of the most popular community skills:

• Run tests and fix failures — executes the test suite, analyzes failures, and fixes the implementation (not the tests) until all tests pass
• Create PR with changelog — generates a pull request with a structured summary, test plan, and changelog entry based on the diff
• Security review — audits code changes for common vulnerabilities: hardcoded secrets, SQL injection, XSS, missing auth checks, and insecure dependencies
• Refactor module — breaks a large file into smaller, focused modules while preserving all exports and updating import paths across the codebase

Discover and install skills with Caliber

Caliber connects you to a growing ecosystem of community-built skills. The caliber skills command lets you browse, search, and install skills from sources like skills.sh:

The workflow is straightforward:

1. 1Browse — run caliber skills to see available skills with descriptions, scores, and compatibility info
2. 2Install — select skills to install. Caliber writes them to the correct directory for your AI tool
3. 3Use — skills are immediately available in your AI assistant. Invoke them by name or let the assistant pick the right skill for the task
4. 4Share — commit skills to git so your entire team gets the same workflows

Bootstrap skills for your stack

Don't want to pick skills manually? The caliber bootstrap command analyzes your project's tech stack and installs a recommended set of skills automatically. A React + TypeScript project gets different skills than a Go microservice. Bootstrap looks at your dependencies, framework, testing setup, and CI configuration to select the most relevant workflows.

SkillsBench: scored and benchmarked skills

Not all skills are created equal. Caliber uses SkillsBench to score and benchmark skills against real codebases. Each skill in the marketplace has a quality score based on how well it performs across different project types, how consistently it produces correct results, and how well-structured its instructions are. This means you can install skills with confidence — the top-rated ones have been validated on real projects.

Skills vs other config types

Caliber manages several types of AI configuration. Here's how skills fit in:

• Skills — task-specific workflow templates (“how to do X”)
• Instructions (CLAUDE.md / .cursorrules) — project-wide context and conventions (“what this project is”). See the Claude Code setup and Cursor setup guides
• Agents (AGENTS.md) — specialized personas for different roles. See the AGENTS.md generator guide
• MCP servers — external tool integrations (databases, APIs, deployment)

Caliber generates and manages all of these from a single caliber init scan. Learn more about the unified AI setup process.

Frequently asked questions