claudekit / tools / agent-skills
[ Plugin · Development ]

Agent Skills

Production-grade engineering skills for AI coding agents — 24 skills encoding the workflows, quality gates, and best practices senior engineers use when building software. Organized by lifecycle phase (Define, Plan, Build, Verify, Review, Ship), skills activate automatically based on what you're doing: designing an API triggers `api-and-interface-design`, building UI triggers `frontend-ui-engineering`. "Process, not prose" — skills are workflows agents follow, not reference docs they read, and every skill carries evidence requirements like passing tests or runtime data.

addyosmani/agent-skills ·published ·updated
$ /plugin marketplace add addyosmani/agent-skills && /plugin install agent-skills@addy-agent-skills copy

An open-source project by Addy Osmani, Director at Google working on Gemini and Google Cloud (addyosmani/agent-skills).

What it does

Coding agents produce code fast, but they don’t follow the process senior engineers do — refining the spec first, verifying with tests, checking security and performance before shipping. Prompting the process each time leads to omissions, and long guideline documents in context get skimmed rather than followed.

Agent Skills encodes that process as 24 skills. The design philosophy is “Process, not prose” — skills are workflows agents follow, not reference docs they read. Every skill carries evidence requirements like passing tests or runtime data (“Verification is non-negotiable”), and anti-rationalization tables document the common excuses agents make for skipping steps, with counter-arguments.

What’s inside

The 24 skills are organized by development lifecycle phase.

PhaseSkillsExamples
Define3interview-me, idea-refine, spec-driven-development
Plan1planning-and-task-breakdown
Build7test-driven-development, context-engineering, frontend-ui-engineering, api-and-interface-design
Verify2browser-testing-with-devtools, debugging-and-error-recovery
Review4code-review-and-quality, code-simplification, security-and-hardening, performance-optimization
Ship6git-workflow-and-versioning, ci-cd-and-automation, documentation-and-adrs, shipping-and-launch

Alongside the skills:

  • Four specialist agent personas — code-reviewer, test-engineer, security-auditor, web-performance-auditor
  • Four reference checklists — testing-patterns, security-checklist, performance-checklist, accessibility-checklist
  • Hooks — session lifecycle handling
  • Per-tool slash commands — 7 for Claude Code (.claude/commands/), 7 for Gemini CLI, 8 for Antigravity

Quick start

  1. Add the marketplace and install the plugin — /plugin marketplace add addyosmani/agent-skills && /plugin install agent-skills@addy-agent-skills
  2. Work as usual — skills activate automatically based on what you’re doing. Designing an API triggers api-and-interface-design; building UI triggers frontend-ui-engineering.
  3. To drive phases explicitly, use the slash commands — /spec, /plan, /build, /test, /review, /code-simplify, /ship.

Notes

  • MIT license — open-source project addyosmani/agent-skills.
  • Not Claude Code-only — per-tool setup is documented in docs/: Cursor (copy SKILL.md files into .cursor/rules/), Windsurf (rules configuration), OpenCode (AGENTS.md and the skill tool), GitHub Copilot (.github/copilot-instructions.md).
  • Mind the marketplace name — the plugin identifier is agent-skills@addy-agent-skills. Running install without first adding the marketplace (/plugin marketplace add) will not find it.
  • Evidence-based verification — every skill specifies evidence requirements such as passing tests or runtime data, blocking the pattern where an agent just reports “done” without proof.
§ 5

See also

same category · curated
[01]
[Tool] Paseo · Run any coding agent from your phone, desktop, or terminal. Self-hosted, multi-provider, open source. A local daemon runs the agents on your machine and every client — desktop, mobile, web, CLI — connects to it, with the native mobile app at full feature parity with desktop.
tool · claudekit.io / tools / paseo
[02]
[Tool] Orca · Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. It runs each agent in an isolated git worktree, lets you compare outputs side by side, and merge the ones you want. Works with Claude Code, Codex, Grok, and 30+ CLI agents. Available on desktop and mobile.
tool · claudekit.io / tools / orca
[03]
[Skill] Matt Pocock's Skills · Skills For Real Engineers. Straight from Matt Pocock's .claude directory. 20+ agent skills for real engineering — not vibe coding — covering alignment (grilling), specs, TDD, code review, bug diagnosis, and architecture. Install as an editable fork via skills.sh or as a read-only plugin bundle.
tool · claudekit.io / tools / mattpocock-skills
§ 6

Frequently Asked Questions

frequently asked
§ 6.1
What is Agent Skills?
Production-grade engineering skills for AI coding agents — 24 skills that encode the workflows, quality gates, and best practices senior engineers use when building software, organized by lifecycle phase (Define, Plan, Build, Verify, Review, Ship) and activated automatically based on what you're doing.
§ 6.2
Which agents and tools does it support?
Claude Code, Cursor, Antigravity CLI, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro, and generic agents accepting Markdown instruction files. Per-tool setup guides live in docs/.
§ 6.3
How do I install it?
In Claude Code, add the marketplace and install: `/plugin marketplace add addyosmani/agent-skills && /plugin install agent-skills@addy-agent-skills`. For local development, clone the repo and point `claude --plugin-dir` at it.
§ 6.4
How do skills activate?
Skills activate automatically based on what you're doing — designing an API triggers api-and-interface-design, building UI triggers frontend-ui-engineering. You can also invoke phases directly with slash commands like /spec, /plan, /build, /test, /review, /code-simplify, and /ship.
§ 6.5
What ships besides the skills?
Four specialist agent personas (code-reviewer, test-engineer, security-auditor, web-performance-auditor), four reference checklists (testing, security, performance, accessibility), and session lifecycle hooks.
§ 6.6
Is it free?
Yes — open source under the MIT license.