Key takeaways
- Stop agents skipping the boring steps: 25 workflows cover spec, plan, build, test, review and ship.
- Nine slash commands map to the development lifecycle, and skills also activate automatically as you work.
- Each skill ends with evidence requirements, so 'seems right' is not enough to call a task done.
- Plain Markdown that works with any agent accepting instruction files, under the MIT licence.
- The catches: native support varies by tool, single-skill installs miss shared checklists, and we have not tested it.
What it is
Agent Skills describes itself as production-grade engineering skills for AI coding agents. It is a collection of Markdown workflows released under the MIT licence, and the code is on GitHub at addyosmani/agent-skills.
The problem it solves
The project's own pitch is that AI coding agents default to the shortest path, which often means skipping specs, tests and security reviews. Anyone who has let an agent loose on a real codebase will recognise that.
The answer here is process rather than prose: skills are workflows agents follow, not reference documents they read.
What you get
- A full lifecycle: the pack includes 25 skills in total, from defining and planning through building, verifying, reviewing and shipping.
- Easy entry points: nine slash commands map to the lifecycle, and skills also activate automatically based on what you are doing.
- Fewer excuses: every skill includes a table of common excuses agents use to skip steps, with counter-arguments.
- Proof, not vibes: verification is described as non-negotiable, and each skill ends with evidence requirements.
- Specialist reviewers: pre-configured personas for targeted reviews, plus an optional automatic build mode where you approve the plan once.
The catches
- Uneven support: the lifecycle slash commands are Claude Code commands and do not appear in Copilot, and the repository does not ship a Cursor folder.
- Known bug: in affected Antigravity CLI releases, wrapper commands are reported as converted but are not discoverable.
- Portability gap: installing a single skill does not carry the repo-level references directory, so shared checklists go missing.
- Instructions, not enforcement: the skills are plain Markdown that steer an agent. How closely a given agent follows them is not something the README measures.
- Autonomy: the automatic build mode removes the human between tasks, though the project says it pauses on failures or risky steps.
What we did not test
We did not install the pack, run it in any agent, or measure whether it improves the code an agent writes. Every claim here comes from a literal quote in the project's README or the GitHub API, and the descriptions of the skills are the project's own.
Verdict
Editors' view: this is for developers who already lean on coding agents and keep finding that tests, specs and reviews get skipped. If you work alone on small scripts, it is probably more process than you need. If you try it, do so on a low-stakes project first and judge the results yourself.
Sources
Need help adopting this, or have a software project in mind?
Tell us what you are building. The initial consultation is free. Email [email protected] or follow @zolnaltd.
Start a conversation →