GitHub - addyosmani/agent-skills: Production-grade engineering skills for AI coding agents. · GitHub
Agent Skills is an MIT-licensed, cross-agent pack of 25 structured engineering workflows that pushes AI coding agents toward specs, tests, reviews, security checks, and safe releases. It offers lifecycle slash commands, native adapters for major coding agents, and evaluation-backed skill routing.
GitHub - addyosmani/agent-skills: Production-grade engineering skills for AI coding agents. · GitHub
Author: Addy Osmani | Published: Unknown | Generated: 2026-05-07 | Domain: github.com
Tags: ‘#ai-agents’ ‘#coding-agents’ ‘#engineering-workflows’ ‘#developer-tools’ ‘#claude-code’ ‘#agent-skills’
TLDR
Agent Skills packages senior-engineering practices as 25 portable Markdown workflows for AI coding agents, spanning requirement definition through production launch. The pack adds nine lifecycle commands, 24 specialized lifecycle skills plus a routing meta-skill, four reviewer personas, and seven checklists—all designed around explicit process steps, anti-rationalization safeguards, and verifiable evidence rather than generic prompting. It supports 70+ agents through the Skills CLI and native integrations for tools including Claude Code, Codex, Gemini CLI, Cursor, Copilot, and OpenCode.
Key Takeaways
- Full development lifecycle: Nine commands—
/spec,/plan,/build,/test,/constraints,/review,/webperf,/code-simplify, and/ship—map to define, plan, build, verify, review, and deployment stages;/build autogenerates and executes an approved plan while preserving per-task testing, commits, and failure pauses. - Structured, evidence-driven skills: Each
SKILL.mdincludes triggering conditions, a step-by-step workflow, rationalization rebuttals, red flags, and verification requirements; “seems right” is explicitly not accepted as evidence. - 25 skills and specialist reviews: The repository provides 24 lifecycle skills plus
using-agent-skillsfor routing, alongside four targeted personas: code reviewer, test engineer, security auditor, and web-performance auditor. - Cross-tool portability:
npx skills add addyosmani/agent-skillsinstalls all 25 skills into 70+ agent environments. Native installation guidance is included for Claude Code, Codex CLI v0.122+, Gemini CLI, Antigravity, Cursor, Windsurf, OpenCode, GitHub Copilot, Kiro, and Command Code. - Quality practices encoded directly: Workflows incorporate concepts such as test-driven development and an 80/15/5 test pyramid, WCAG 2.1 AA accessibility, OWASP Top 10 prevention, Core Web Vitals measurement, trunk-based development, feature flags, and staged rollouts.
- Repository signals: At capture time, the project reports 94.4k stars, 10k forks, 67 contributors, 512 commits, 11 releases, and latest release
0.6.9dated 2026-09-05.
Images & Media
- Addy’s Agent Skills overview graphic — Promotional illustration for the Agent Skills pack.
- Development lifecycle diagram — Inline ASCII workflow depicting Define → Plan → Build → Verify → Review → Ship, with corresponding slash commands.
- Skill anatomy diagram — Inline ASCII diagram showing a
SKILL.mdstructure: frontmatter, overview, triggers, process, rationalizations, red flags, and verification.
Referenced Links
- Skills CLI — Open CLI used to install the pack into 70+ supported agents.
- Adoption Guide — Rollout guidance for greenfield projects and incremental adoption in established codebases.
- Software Engineering at Google — Source for engineering principles reflected in the workflows.
- Google Engineering Practices — Google’s engineering-practice guidance referenced by the project.
- Comparison with Superpowers and Matt Pocock’s skills — Side-by-side comparison of the skill packs and intended use cases.
- Controlled head-to-head experiment — Linked experiment comparing Agent Skills and Superpowers.
- Portability gap issue #361 — Tracks missing shared
references/files when a single skill is installed vianpx. - Claude Code setup — Universal getting-started guidance for agents that accept prompts or instruction files.