Guide · Agent skill
How to use the Learn Agent Skills Tutor agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The learn-agent-skills skill turns your AI agent into a step-by-step tutor for the Agent Skills part of the AI Engineering from Scratch course. It is for learners who want guided practice building and securing their own agent skills, and for teams who want to see how skills are discovered, invoked, permissioned and packaged. After working through this guide you will know how to install the skill, start or resume the route, and what the tutor will ask you to do.
Learn Agent Skills Tutorrohitg00/ai-engineering-from-scratch · MIT Download the skill Quick start
- 1Download the ZIP. Download the learn-agent-skills ZIP from the skill's page on our directory.
- 2Install the skill. Unzip it into ~/.claude/skills/ for your user account, or into .claude/skills/ in a project. In the Claude apps, upload it under Settings → Capabilities → Skills.
- 3Start the tutor. Invoke learn-agent-skills using your host's form, such as /learn-agent-skills in Claude Code.
- 4Pass the preflight. Confirm node, npx and python3 respond, choose a skill-capable host, and choose an install scope. Answer only the questions the tutor cannot infer.
- 5Complete one lesson. Create the files, run the lab, pass the checkpoint, answer the quiz, and let the tutor mark the row Done in AGENT-SKILLS-LEARNING.md.
What does the Learn Agent Skills tutor make the agent do?
Each invocation covers one lesson. The agent explains the idea, has you create files and run a lab, asks quiz questions, and records your progress in a local AGENT-SKILLS-LEARNING.md file.
The tutor reads its lesson plan from the course repository, either from a local clone or from GitHub. The route is the Agent Skills Engineering path, with five required lessons, and the lessons are followed in manifest order.
- Lesson 22: portable contract and runtime boundary
- Lesson 24: discovery and progressive disclosure
- Lesson 25: invocation and routing
- Lesson 26: permissions, sandboxes and trust
- Lesson 27: evals, packaging and portability
- Lesson 23 is optional and follows the manifest's entry rule
How do you install the learn-agent-skills skill?
Download the ZIP from the skill's page on our directory. Then either unzip it into your user skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill limited to one project.
In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.
- User scope: ~/.claude/skills/
- Project scope: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills
How do you start or resume the tutor?
The command depends on your host. The skill's portable name is learn-agent-skills, so do not assume one command works everywhere.
If the route is already in progress, the tutor resumes at the first row marked Next or In progress rather than restarting.
- Codex: type learn-agent-skills, or choose it from /skills
- Claude Code: type /learn-agent-skills
- Other SKILL.md-compatible hosts: ask in plain words, for example 'Use learn-agent-skills to start or resume the Agent Skills Engineering path.'
/learn-agent-skillsWhat does the tutor need before it runs a lab?
Before the first host checkpoint, the tutor checks that node, npx and python3 respond to their version commands, that you have chosen one skill-capable host, and that you have chosen a writable project or user install scope. It also confirms which working directory will be treated as your workspace.
If any of these is missing, the tutor continues with the conceptual lesson and marks real-host observations as Pending. It does not describe that fallback as a real host pass. Lab checkpoints stay Pending without a real skill-capable host.
- Node.js and npx
- Python 3
- A skill-capable host such as Claude Code, Codex, or another SKILL.md-compatible agent
- A writable project or user directory for progress notes and lab files
- Optional: a local clone of the repository, or network access to raw.githubusercontent.com to read lesson files
How is your progress tracked?
The tutor creates AGENT-SKILLS-LEARNING.md in your current working directory the first time you start. It holds a route summary, a prerequisite check and a progress table with Order, Lesson, Status, Evidence and Completed columns. If the file already exists, the tutor keeps your notes and evidence.
A lesson is marked Done only after its checkpoint and quiz are complete. Each lesson has a checkpoint, for example a trace of discovery, invocation, or script loading in a real host.
- Lesson 22 checkpoint: create a minimal skill, install the reviewer bundle into a real host, invoke it, verify the report, and remove it cleanly
- Lesson 24 checkpoint: trace discovery, catalog metadata, body activation and reference or script loading
- Lesson 25 checkpoint: record explicit, implicit, negative and near-miss routing results
- Lesson 26 checkpoint: label each control as instruction, permission, sandbox or verification, and prove the boundary
- Lesson 27 checkpoint: exercise discovery, references, scripts, approvals, upgrade and uninstall in one host, then repeat in a second host
What safety points should you know before you use it?
The skill does not run scripts, but it does use the network. It can read lesson files from GitHub, and it can ask for a clone of the repository. The skill says that install, clone and update actions need your confirmation, but they still change local files and systems, so read each action before you approve it.
Skill instructions do not bypass your host's permissions or sandbox. When the tutor cannot observe a host behavior, it records that behavior as unverified rather than assuming support.
- Review every install, clone or update action before confirming it
- Keep the tutor's working directory separate from the installed skill folder
- Expect checkpoints to stay Pending if no real host is available
What people use it for
- Learning how a skill is discovered, invoked, permissioned and packaged, one lesson at a time
- Practising building a minimal skill and checking that it installs and uninstalls cleanly in a real host
- Testing whether a skill routes correctly with explicit, implicit, negative and near-miss requests
- Reviewing the trust boundaries between instructions, permissions, sandboxes and verification
Frequently asked questions
Is the learn-agent-skills skill free?
The source material lists the skill under the MIT licence. It does not list prices, so check the licence file in the repository for the terms that apply to you.
What do I need to run learn-agent-skills?
You need Node.js with npx and Python 3, a skill-capable host such as Claude Code, Codex, or another SKILL.md-compatible agent, and a writable directory for progress notes and lab files. A local clone of the repository is optional if the network can reach raw.githubusercontent.com.
Where do I install the learn-agent-skills skill?
Unzip the downloaded ZIP into ~/.claude/skills/ for all your projects, or into .claude/skills/ inside one project. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
Why does my lab checkpoint say Pending?
Lab checkpoints stay Pending when no real skill-capable host is installed, or when the required repository files or runtime are unavailable. Install a supported host and confirm the preflight, and the tutor can re-check the lab.
Does the tutor run scripts on my computer?
The source says this skill runs no scripts, but it uses the network to read lesson files. Any install, clone or update action changes local files, so approve each one knowingly.
Is this a good alternative to a finished skill for daily work?
No. This skill is a course tutor, not a productivity tool. If you only want a finished skill to use right away, install a ready-made skill instead of this one.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use learn-agent-skills if you are following the Agent Skills Engineering path or want checkpointed, hands-on practice building and securing your own skills, and you have a skill-capable host installed. Pick something else if you only want a finished skill for everyday work, or if you would rather not let an agent make local changes during a lesson.
Download the skill