How to use the Build Project Tutor agent skill (build-project)

What is an agent skill, and what does build-project make the agent do?

An agent skill is a folder containing a SKILL.md file that gives a compatible AI agent specific instructions for a task. When the skill is installed and triggered, the agent follows those instructions instead of its general behaviour. This one is named build-project and is version-tagged in its file, but the important part is its job.

The skill turns the agent into a project tutor for the AI Engineering from Scratch Projects section. Per session it reads the lesson for one stage, asks you a prediction question, points you to starter files, runs the stage's grader with you, and gives hints when tests fail. It records your progress in a PROJECTS-LEARNING.md file in your working folder so you can resume later.

• Reads the stage lesson from docs/en.md and the starter stubs in the project folder
• Asks one prediction question before any code is written
• Runs the stage grader and reports results only from real grader output
• Gives up to three levels of hints, starting with a question
• Tracks each stage as Next, In progress or Done in PROJECTS-LEARNING.md

Install the build-project skill

Download the ZIP for the build-project skill from this page. You have two routes, depending on how you use your agent.

For Claude Code, unzip the download into your user skills folder ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder to limit it to one repository. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

The source lists Codex and other SKILL.md-aware hosts as compatible. It does not give a separate install path for them, so check your host's documentation for where skills are loaded from.

• Claude Code, all projects: unzip into ~/.claude/skills/
• Claude Code, one project: unzip into .claude/skills/ inside the project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check the requirements before you start

The skill needs Python 3 available as the python3 command, because the grader and workspace setup run through it. You also need a local clone of the rohitg00/ai-engineering-from-scratch repository, or network access so the skill can fetch raw files from GitHub.

You can confirm Python is ready with the version check the skill itself uses. If Python or the repository is missing, the skill falls back to conceptual mode, where grader results are marked Pending rather than Pass. You get less verification in that mode.

• Python 3, invoked as python3
• A local clone of the repository, or network access to GitHub raw files
• A compatible agent host such as Claude Code or Codex

Trigger the skill with a request

How you start a session depends on the host. In Claude Code you type /build-project, optionally followed by a project ID. In Codex you type build-project or pick it from /skills. In other compatible hosts, ask in plain language.

The skill also responds to natural requests. Its listed trigger phrases are "build a project", "next project stage", "continue my project" and "start the research report agent". If you do not name a project, the skill lists the ready projects with their level and a one-line description and asks which one to open.

• Claude Code: /build-project or /build-project research-report-agent
• Codex: build-project, or select it from /skills
• Other hosts: Use build-project to start or resume my project.

Work through one stage at a time

Each session covers a single stage of a single project. The skill first sets up a workspace folder for the first stage, defaulting to a folder named my-<project-id> next to the repository. It then goes through the lesson, asks a prediction question, and points you to the starter file and the function signatures from the lesson's task section.

You write the code in your workspace. When you are ready, the skill runs the grader for the current stage. Because the grader runs stages 1 to N, a failure in an earlier stage means your new code broke older behaviour, and the skill will say so. After a pass, it asks the lesson's Check yourself questions one at a time.

• Stage 1 setup runs the project test script with --init into your workspace
• Hints go from a question, to the concept, to the specific line or edge case
• The skill stops after recording progress for the stage

Track progress and finish a project

After each stage the skill updates the stage row in PROJECTS-LEARNING.md with the status, grader summary, date and a one-line note in your own words. The file can hold several projects, and the skill never overwrites existing notes. Next session, it resumes at the first row marked Next or In progress.

When the last stage passes, the skill runs all stages in strict mode and writes a completion report. The source says you can import that report on the project page for a local completion certificate, and that original projects can be submitted through projects/SUBMITTING.md.

• Resume automatically from PROJECTS-LEARNING.md
• Reports are imported on the website's project page
• The projects page at projects.html lets you tick stages as done