The agent-skills repository is a production-grade collection of engineering workflows and best practices packaged for AI coding agents. Created by Addy Osmani, it encodes the methodologies that senior engineers follow across the full software development lifecycle, enabling AI agents to apply consistent quality standards and proven patterns automatically.
The core functionality centers on eight slash commands that map to distinct phases of development: /spec for specification, /plan for task breakdown, /build for incremental implementation, /test for verification, /review for code quality gates, /webperf for performance auditing, /code-simplify for clarity improvements, and /ship for production deployment. Each command automatically activates the relevant skills for that phase. The /build auto variant removes manual stepping between tasks by generating a plan and implementing every task in a single approved pass, though it maintains test-driven development and individual commits for each task while pausing on failures or risky steps.
The skill pack contains 24 total skills organized across six categories. The Define phase includes interview-me for extracting actual user requirements, idea-refine for structured thinking on rough concepts, and spec-driven-development for comprehensive PRDs. The Plan phase offers planning-and-task-breakdown for decomposing specifications into verifiable units. The Build phase encompasses incremental-implementation for thin vertical slices, test-driven-development following red-green-refactor patterns, context-engineering for feeding agents appropriate information, source-driven-development for framework decisions grounded in official documentation, doubt-driven-development for adversarial review of high-stakes decisions, frontend-ui-engineering for user-facing interfaces, and api-and-interface-design for contract-first API design. The Verify phase includes browser-testing-with-devtools using Chrome DevTools MCP and debugging-and-error-recovery with five-step triage. The Review phase covers code-review-and-quality with five-axis review methodology, code-simplification using Chesterton's Fence principles, security-and-hardening addressing OWASP Top 10 patterns, and performance-optimization with measure-first approaches targeting Core Web Vitals.
The repository supports integration across multiple AI coding platforms including Claude Code via marketplace installation, Cursor through rule files, Antigravity CLI as native plugins, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro IDE, and other agents accepting system prompts or instruction files. Documentation represents the most active issue label. Each skill is structured as a Markdown workflow with explicit steps, verification gates, and anti-rationalization tables designed to prevent common reasoning errors and ensure consistent application of engineering best practices across AI-driven development work.
