Playbooks · Coding
Addy Osmani's 25 agent skills for coding agents
Written November 11, 2026
The pack turns the development lifecycle into 25 skills an agent can follow, from spec to ship, with review gates before merge.
The problem
An AI coding agent left to itself takes the shortest path: it skips the spec, writes code before tests, and merges without a review. The result works once and breaks later. Addy Osmani's agent-skills pack gives the agent a written workflow for each phase, so the discipline a senior engineer brings to production code is in the instructions, not left to chance.
What the pack is
The pack holds 25 skills: 24 lifecycle skills plus a using-agent-skills router that maps incoming work to the right skill. Each skill is a structured workflow with steps, verification gates, and a table of common excuses for skipping steps, with the counterargument written next to each one. Every skill ends with evidence requirements: tests passing, build output, runtime data. "Seems right" is never sufficient, per the pack's own design notes. The repo is MIT licensed, and shows more than 115,000 stars on GitHub.
Alongside the skills sit 4 reviewer personas (code reviewer, test engineer, security auditor, web performance auditor) and 7 reference checklists covering definition of done, testing patterns, security, performance, accessibility, observability, and orchestration. Skill routing is tested in CI against 25 eval cases, and the release notes report rank-1 routing at 94 of 95.
The 9 slash commands
The commands map to the lifecycle, and each one activates the right skills on its own:
| Command | What it does |
|---|---|
/spec | Define what to build, spec before code |
/plan | Break the spec into small, atomic tasks |
/build | Build incrementally, one slice at a time |
/test | Prove it works, tests are proof |
/constraints | Set the quality bar once, enforce it everywhere |
/review | Review before merge, improve code health |
/webperf | Audit web performance, measure before optimizing |
/code-simplify | Simplify the code, clarity over cleverness |
/ship | Ship to production, staged rollouts with a rollback plan |
In Claude Code these install as plugin commands. Where a built-in command has the same name, as /review does, use the namespaced form /agent-skills:review. Skills also activate on their own from what you are doing: designing an API triggers the API design skill, building UI triggers the frontend skill.
How to install it
The fastest path works in 70+ agents through the skills CLI:
npx skills add addyosmani/agent-skills # install all 25 skills
npx skills add addyosmani/agent-skills --list # browse before installing
To install one skill only:
npx skills add addyosmani/agent-skills --skill code-review-and-quality
One catch, stated in the repo's own install notes: installing a single skill copies only that skill's directory, not the shared references folder, so paths to supplementary checklists are unavailable. A whole-repo install, a clone, or copying the needed checklist into the skill avoids that gap. Native installs exist per host, including a Claude Code marketplace install, a Codex plugin install, and a Gemini CLI install.
What 0.6.15 changes
Version 0.6.15 is the latest release. Its headline fix is a security one: the security skill's staged-secret check used to print every matching line of a staged diff, so a staged key landed in the terminal and in any agent's transcript. The check now prints only file names, and a git failure exits nonzero by itself. The release also fixes two simplify-ignore bugs that could corrupt protected code, and makes the constraints skill's security and accessibility gates fail CI when they should, instead of passing while the written bar is broken.
Limits
- The repo page does not publish per-skill install counts, so this page does not rank the skills by installs.
- Skills are plain Markdown workflows. They make the agent follow a process; they do not guarantee the code is good. The verification gates at the end of each skill are what has to catch bad work.
- Host support varies. The lifecycle slash commands are Claude Code commands and do not appear in Copilot, where skills are invoked by name instead. The repo lists only hosts a maintainer has run in its install section.
Related playbooks
- Fifty agents built a bureaucracy; four beat the game
- Stroq: audit what your coding agent touched
- Agent Skills has a neutral home — and still no security story
- MCPHallu: agents hallucinate across 358 tasks
Sources
Back to: Coding playbooks hub · Also: Skills index · Agent Skills joins the Agentic AI Foundation · Agent Skills has a neutral home — and still no security story · Kimi Code 2.5.3: steer a running session