An agent with a security-review skill checks for things it might otherwise skip.
The right skills for every task. Nothing extra.
A skill is a set of written instructions that teaches your AI coding agent one job, like securing a login flow or designing a database. AI Skills keeps a library of them on your machine and gives your agent only the skills that match the task, so the rest don't crowd its context.
What's a skill?
Think of it as a checklist from an expert. It tells your agent how to do one kind of job well: what to look for, which steps to follow, and when it's finished. A skill is a short file of instructions. No code required.
---
name: Authentication Security
description: Review authentication flows,
credential handling, sessions, tokens,
recovery, and identity boundaries.
---
## Check
- How passwords and tokens are stored
- Session expiry and logout
- Password reset can't be abusedMost setups give every installed skill and tool to the agent at the start of every chat. The more you add, the more context gets used before work starts, and the more likely the agent picks the wrong tool.
Keep as many skills as you like. Each task gets only the ones it needs.
More skills shouldn't mean a worse agent.
Three problems show up as soon as you give an agent more than a handful of skills and tools.
Context fills up
Instructions for jobs you aren't doing take up room in every chat. With AI Skills: only matched skills are loaded.
without it: every skill, every chatWrong tool, wrong advice
When dozens of options are in view, agents pick the wrong one more often. With AI Skills: your task and project decide what's offered.
Skills from strangers
A skill from GitHub is instructions your agent will follow. With AI Skills: it's scanned, pinned to a version, and blocked until you approve it.
Three steps. Then you just describe your task.
AI Skills runs in your terminal next to your agent. It changes what your agent starts with, not how you work.
npm i -g @prof-bilal/aiskills@betaaiskills init --projectInstall and set up
One global install, then one setup command in your project. All 43 built-in skills are ready to use.
aiskills session --agent=claude "add login to my app"Describe the task
AI Skills matches your words and your project (for example, Supabase in your dependencies) against your skills.
Your agent gets to work
Claude Code or Codex opens with just those skills. If the task changes, it can look up more as it goes.
43 skills built in. Add your own anytime.
Skills are grouped into packs, so you can turn on a whole area at once, such as aiskills packs install security.
Core
How to workBuild
Write the codeThink
Decide what to buildQA
Check the workSecurity
Find the holesUp and running in a minute.
Needs Node.js 20+ and Claude Code, Codex, OpenCode, or Gemini CLI. Want a skill from GitHub? Install it, review it, then turn it on for your project.
Read the full guide# 1. install
npm i -g @prof-bilal/aiskills@beta
aiskills init --project
# 2. use it (built-in skills work right away)
aiskills session --agent=claude "add login"
# optional: add a skill from GitHub
aiskills install anthropics/skills \
--path=skills/frontend-design
aiskills skills review frontend-design --yes
aiskills skills enable frontend-designCommon questions
Still unsure? Ask on GitHub.
Is this another AI agent?
No. AI Skills doesn't write code itself. It sits in front of the agent you already use and decides which skills that agent gets for the task.
Which agents does it work with?
aiskills run works with Claude Code, Codex, OpenCode, and Gemini CLI. Interactive sessions (aiskills session) currently support Claude Code and Codex.
Do I have to write skills myself?
No. 43 skills come built in. You can also add skill folders from GitHub or your own disk, including the common SKILL.md format.
Does it replace MCP servers?
No, it manages them. Approved MCP tools stay switched off until a task needs them, then start behind a sandbox called Warden.
Is a skill from GitHub safe?
Not automatically. AI Skills scans each skill without running it and waits for your approval before your agent can use it. The scan catches common red flags but can't prove a skill is safe, so read what you approve.
What does it cost? What do I need?
It's free and MIT-licensed. You need Node.js 20 or newer and at least one supported agent installed. It's currently in public beta.
Every skill.
Only when needed.
Install once and keep working the way you already do.