agent-skills

This is an open-source skill pack designed for AI coding agents, created by Addy Osmani to enforce production-grade engineering discipline across the full software development lifecycle. It addresses the common issue of AI agents skipping critical steps like spec writing, testing, security reviews and quality gates by providing structured, opinionated workflows that mirror the practices of senior engineers. The pack includes 25 total skills: 24 lifecycle-focused skills organized into 6 development phases, plus 1 meta skill for routing work to the correct workflow. The phases are: Define (covering requirement interviewing, idea refinement, spec-driven development and constraint setting to clarify what to build), Plan (task breakdown to decompose specs into verifiable units), Build (incremental implementation, test-driven development, frontend UI engineering, API design, context engineering, source-driven development and doubt-driven development for writing code), Verify (browser testing with DevTools and debugging/error recovery to prove functionality), Review (code review, code simplification, security hardening and performance optimization to set quality bars), and Ship (git workflow, CI/CD automation, deprecation and migration, documentation and ADRs, observability and instrumentation, and launch processes for production deployment). It also includes 4 pre-configured specialist agent personas (code reviewer, test engineer, security auditor, web performance auditor) for targeted review tasks, and 7 reference checklists covering shared standards like definition of done, testing patterns, security, performance, accessibility and observability. The pack supports integration with 70+ AI coding agents and tools, including Claude Code (with native plugin support and 8 slash commands), Cursor, Antigravity CLI, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, byNara IDE/CLI, Codex and Command Code, as well as any agent that accepts system prompts or instruction files. It provides 9 core slash commands mapped to development phases: /spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify and /ship. The /build auto command supports autonomous execution of plan and implementation tasks after one-time user approval, while retaining per-task test-driven verification, individual commits and automatic pausing on failures or risky steps. Skills also activate automatically based on the current task context, for example triggering frontend UI engineering skills when building user interfaces. Each skill follows a consistent anatomy with frontmatter, step-by-step workflows, anti-rationalization tables that counter common agent excuses for skipping steps (such as "I'll add tests later"), non-negotiable verification requirements that demand concrete evidence rather than "seems right" judgments, and progressive disclosure to minimize token usage by only loading supporting reference material when needed. The workflows incorporate proven engineering practices from Google's engineering culture, including Hyrum's Law for API design, the Beyonce Rule and 80/15/5 test pyramid for testing, ~100 line change sizing norms for code review, Chesterton's Fence for code simplification, trunk-based development for git workflow, Shift Left and feature flags for CI/CD, and a code-as-liability mindset for deprecation. The repository includes an adoption guide for both greenfield projects and incremental rollout for existing codebases, contribution guidelines requiring skills to be specific, verifiable, battle-tested and minimal, and a comparison document with similar skill packs such as Superpowers and Matt Pocock's skills to help users choose the right tool for their needs.