Skill of the day: agent-skills, 25 engineering skills for your coding agent

When a coding agent writes faster than anyone reviews, the bottleneck stops being code and becomes process. Addy Osmani's agent-skills, MIT-licensed and last updated on 13 October 2026, bundles 25 skills that encode the engineering lifecycle (define, plan, build, verify, review and ship) and is among the skill repos gaining traction on GitHub this week.

01What it does

It isn't one skill, it's a pack of 25: 24 lifecycle skills plus a meta-skill, using-agent-skills, that picks which one applies to a task. Each SKILL.md is a workflow with steps, verification gates and "rationalization" tables meant to stop the agent from skipping steps. Examples from the README:

• spec-driven-development: write a PRD covering objectives, commands, structure, style, tests and boundaries before touching code.
• code-review-and-quality: five-axis review (correctness, readability, architecture, security, performance) before merge. Its approval standard is to approve if the change improves code health, even if imperfect.
• constraint-driven-development: interviews you for a quality bar, writes a CONSTRAINTS.md, and catches agents silencing checks or skipping tests to get green.
• interview-me: one question at a time until it knows what you actually want.

It also ships nine commands mapped to the lifecycle: /spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify and /ship. With /build auto you approve the plan once and the agent implements every task; each task is still test-driven and committed on its own, and it pauses on failures or risky steps.

02How to install and use it

In Claude Code, as a plugin:

/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

For any agent, with the open skills CLI, install everything or a single skill:

npx skills add addyosmani/agent-skills --list
npx skills add addyosmani/agent-skills --skill code-review-and-quality

Skills activate on their own based on what you're doing (designing an API triggers api-and-interface-design, for example). Where a command clashes with a built-in one, like /review, use the namespaced form: /agent-skills:review. The README also covers Cursor, Codex, Gemini CLI, Copilot, OpenCode, Kiro and more.

03Risks and limits

I read the README and the SKILL.md of code-review-and-quality, and they are plain Markdown: instructions for the agent, with no scripts of their own to run in that skill. I did not review all 25 one by one, and I couldn't query the GitHub API from the routine, so I'm not quoting stars or install counts. Worth knowing:

• Installing a single skill with npx copies only skills/<name>/, not the repo's references/ directory; the skill works, but the shared checklists are missing (repo issue #361).
• In Claude Code the marketplace install clones over SSH; without GitHub SSH keys it fails with Permission denied (publickey) and you need the HTTPS URL.
• These are instructions that change how your agent behaves. Read them before enabling them, as you would a dependency, and don't install them blindly in a repo with credentials.
• A pack this large adds context and opinions: /build auto removes the human between tasks, so try it first on a project that isn't production.

My take: the value isn't any single skill but the idea of putting explicit verification gates around the agent's work. I would start with code-review-and-quality and constraint-driven-development in a throwaway repo and see which ones turn out to be noise.

04Sources

• The addyosmani/agent-skills repository on GitHub (README, MIT licence and the SKILL.md of code-review-and-quality).