Agent Skills

Workflow

Addy Osmani's 24-skill development-lifecycle pack: spec, plan, build, test, review, ship with enforced quality gates.

by Addy Osmaniaddyosmani/agent-skills

Stars109.5k

Forks11.4k

Watchers546

Open issues147

LicenseMIT

Last push8 days ago

Install

install

$ /plugin marketplace add addyosmani/agent-skills

Update

update

$ /plugin marketplace update addy-agent-skills

README documents no explicit update command; this follows the marketplace slug `addy-agent-skills`. After updating, run /reload-plugins (or restart Claude Code) to pick up new skill versions.

When to use it

You want an AI agent to follow senior-engineer discipline across the whole lifecycle instead of taking the shortest path.
Starting a new project or feature and want a spec and task breakdown enforced before code is written.
You want test-driven, incrementally committed implementation with verification gates rather than a single large diff.
Preparing to ship and need pre-launch checklists, staged rollout, and rollback procedures baked into the workflow.

Commands & flags

/plugin install agent-skills@addy-agent-skills: Install the pack from the marketplace after adding it.
/spec: Write a PRD covering objectives, structure, code style, testing, and boundaries before any code.
/plan: Decompose the spec into small, verifiable, dependency-ordered tasks with acceptance criteria.
/build auto: Approve the plan once, then implement every task test-driven and committed individually, autonomously.
/review: Five-axis code review with severity labels and ~100-line change sizing before merge.
/ship: Run pre-launch checklists, staged rollouts, feature-flag lifecycle, and rollback procedures.

How to use it

Install in Claude Code via the marketplace, then install the plugin:

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

Drive the lifecycle with the 7 slash commands (/spec, /plan, /build, /test, /review, /code-simplify, /ship). Skills also auto-activate by task — building UI triggers frontend-ui-engineering, designing an API triggers api-and-interface-design. Use /build auto to approve a plan once and run every task autonomously (still test-driven and committed per task, pausing on failures).

Local/dev install without the marketplace:

git clone https://github.com/addyosmani/agent-skills.git
claude --plugin-dir /path/to/agent-skills

If the marketplace clone fails over SSH, use the HTTPS URL form: /plugin marketplace add https://github.com/addyosmani/agent-skills.git. Setup guides for Cursor, Gemini CLI, Antigravity, Windsurf, Copilot, Kiro, and OpenCode live under docs/.

Why it earns its place

From Addy Osmani (Google Chrome engineering lead, prolific web-performance author), so the practices carry real weight: skills bake in concepts from Software Engineering at Google and Google's eng-practices guide — Hyrum's Law, the test pyramid and Beyonce Rule, Chesterton's Fence, trunk-based development, Shift Left. It's not generic prompts but 24 process-driven workflows with anti-rationalization tables (excuses agents use to skip steps, plus counter-arguments) and non-negotiable verification gates. The 4 specialist review personas and 5 reference checklists compose with the slash commands for targeted quality passes.

Try it

claude

> /spec a rate-limiting middleware for our Express API, then /plan and /build it