Addy Osmani's agent skills
addyosmani/agent-skills
25 engineering skills that take a coding agent from spec to ship, with a verification gate at each step.
Addy Osmani's agent skills is a free, MIT-licensed pack of 25 skills for AI coding agents, created by Addy Osmani. Each skill is a step-by-step workflow with exit criteria and a table of excuses agents use to skip steps. The pack adds 9 slash commands for the lifecycle, from /spec to /ship, and installs in Claude Code, Codex, Gemini CLI and other agents.
Facts checked on against the sources listed at the end of this page.
- GitHub stars
- 100,586
- Skills in the pack
- 25
- Agents supported
- 11plus any agent the skills CLI supports
- Instruction size
- 52.5k wordsacross all SKILL.md files
How to install Addy Osmani's agent skills
| Agent | Command | Copy |
|---|---|---|
| Claude Code | | |
| Codex | Needs Codex CLI v0.122 or later. | |
| Cursor | Copy the skills into .cursor/skills/ and put short policies in .cursor/rules/. The repo's docs/cursor-setup.md has the steps. | |
| Gemini CLI | | |
| GitHub Copilot | Use the agent definitions in agents/ as Copilot personas and the skill content in .github/copilot-instructions.md. For the standalone copilot CLI, see docs/copilot-cli-setup.md in the repo. |
Commands for 3 more agents
| Any agent the skills CLI supports | Installs all 25 skills. Add --list to browse first, or --skill <name> to install one. | |
|---|---|---|
| Antigravity CLI | | |
| Command Code | |
Commands as given in the README on 2 October 2026.
What is inside Addy Osmani's agent skills
| Skill | What it makes the agent do |
|---|---|
| using-agent-skills | A meta-skill that maps incoming work to the right skill and sets shared operating rules. |
| spec-driven-development | Writes a PRD covering objectives, commands, structure, code style, testing and boundaries before any code. |
| planning-and-task-breakdown | Splits a spec into small verifiable tasks with acceptance criteria and dependency order. |
| incremental-implementation | Builds in thin vertical slices: implement, test, verify, commit. |
| test-driven-development | Red, green, refactor, with a test pyramid and test sizes. |
| doubt-driven-development | Runs an adversarial fresh-context review of each non-trivial decision while the work is in progress. |
| code-review-and-quality | A five-axis review covering correctness, readability, architecture, security and performance, with severity labels. |
| debugging-and-error-recovery | A five-step triage: reproduce, localize, reduce, fix, guard. |
| security-and-hardening | Applies OWASP Top 10 prevention, auth patterns, secrets management and dependency auditing. |
| shipping-and-launch | Pre-launch checklists, staged rollouts, rollback procedures and monitoring setup. |
The other 15 skills include interview-me, idea-refine, constraint-driven-development, context-engineering, source-driven-development, frontend-ui-engineering, api-and-interface-design, browser-testing-with-devtools, code-simplification, performance-optimization, git-workflow-and-versioning, ci-cd-and-automation, deprecation-and-migration, documentation-and-adrs and observability-and-instrumentation. The repo also has 4 reviewer personas and 7 reference checklists.
What Addy Osmani's agent skills runs on your machine
| Hooks | Yes, as scripts in the repo. hooks/ holds a session-start script, a WebFetch cache and a simplify-ignore hook. The plugin manifest registers no hooks, and the session-start script says it is not wired by the plugin and is meant only for agents without native skill routing. |
|---|---|
| Bundled scripts | Yes. hooks/ contains shell scripts that need jq, and scripts/ holds 13 contributor scripts for validation and evals. |
| Network calls | Yes, in one optional script. The WebFetch cache hook uses curl to revalidate cached pages with the origin server. The browser-testing-with-devtools skill uses the Chrome DevTools MCP server. |
| API keys needed | None named in the README. |
These are facts read from the repo, not a security rating. Read the files before you install any skill.
How much context Addy Osmani's agent skills uses
| All 25 skills | 52,456 words across every SKILL.md file |
|---|---|
| Largest skill | constraint-driven-development, 3,555 words |
| Loaded at start | Only each skill's name and description. The full text loads when a task needs that skill. |
Word counts are measured from the repo every day. Reference files a skill loads on demand are not counted.
How Addy Osmani's agent skills works
Every skill follows one layout: an overview, when to use it, a step-by-step process, a table of common rationalizations with rebuttals, red flags and a verification section that asks for evidence such as passing tests or build output. The README calls this process, not prose. The SKILL.md is the entry point and supporting references load only when needed.
Skills activate by what you are doing, so designing an API triggers api-and-interface-design and building UI triggers frontend-ui-engineering. The 9 slash commands (/spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify and /ship) are entry points to the same skills. /build auto approves a plan once, then implements every task, committing each one and pausing on failures or risky steps.
A good fit for
- Teams that want one process from spec to ship, with checks the agent must pass before it moves on
- Developers who want pieces of a workflow, since any skill installs on its own with the skills CLI
- Codebases that need review, security and performance gates run by the agent, not left to habit
- People who use several coding agents and want the same skills in each
Skip it if
- Quick one-file edits, where a spec and a plan add more steps than the change needs
- Installing a single skill and expecting the shared checklists, because a per-skill install leaves out the repo-level references folder, a gap the README says is tracked
- Anyone who wants the agent to choose its own process, since the skills set hard exit criteria
Is Addy Osmani's agent skills still maintained?
| Last commit | , 1 day ago |
|---|---|
| Latest release | 0.6.11, 26 September 2026 |
| Stars gained in 7 days | Shown after 7 days of tracking |
| Stars gained in 30 days | Shown after 30 days of tracking |
| Open issues and pull requests | 120 |
| Forks | 10,565 |
Tools Addy Osmani's agent skills works with
Questions about Addy Osmani's agent skills
Does it work with Codex and Cursor?
Yes. The README gives a plugin install for Codex CLI v0.122 or later and a manual setup for Cursor. The npx skills command installs into many other agents, including Cursor and Codex.
Is it free?
Yes. It is open source under the MIT license.
Can I install just one skill?
Yes. Run npx skills add addyosmani/agent-skills --skill <name>. The README warns that a per-skill install copies only that skill's folder, so shared checklists in references/ are not included.
How does it compare with Superpowers?
The README links a side-by-side comparison in docs/comparison.md that covers it, Superpowers and Matt Pocock's skills, and says when to reach for each.
Do the slash commands run automatically?
Only when you call them. The skills activate by task, and /build auto runs every task after you approve the plan once, while still testing and committing each task and pausing on failures or risky steps.