Choose the route that matches your task
Track A teaches using existing CLI agents; Track B teaches building agents. That distinction is the main reason to use this repository: choosing a tool and writing an agent loop require different learning paths.
The English README directs new programmers to Stage 0 and readers who already know Python, Git and APIs to Stage 1. Both tracks first check the shared foundations. People interested only in everyday AI use can go directly to the role guide.
Track A progresses through tool selection, repeatable processes, the Claude Code ecosystem and integration into work. Track B moves through tool use, workflow graphs, frameworks, memory, retrieval and production engineering. These are curriculum topics, not capabilities our sandbox validated.
The maintainer estimates about 8 to 10 weeks for Track A and 16 to 22 weeks for Track B's main path. These are planning estimates, not measured completion times. Pick a route before committing to the whole catalog.
What our sandbox established
Our run on September 30, 2026, at commit ea24e00 recorded successful installation and build steps; tests were skipped.
Our harness stores each command outcome through success status, exit code, elapsed time and timeout status. The attached summary gives installation as 10 seconds and the build step as 4 seconds. The underlying timings are 9,871 and 4,494 milliseconds respectively. Both commands returned 0 without timing out. The stored build log tail is empty, so this evidence does not identify the artifact built or establish that a documentation site rendered correctly.
The stored parser summaries describe command outcomes and do not establish exhaustive, unique test coverage. Here, the test step explicitly says "skipped" with reason "no script"; it supplies no passing-test count. Successful setup commands do not demonstrate successful model calls, working CLI integrations or a deployed agent.
The recorded pip-audit advisory total is 0, with unknown scan coverage. Treat that as a limited advisory result rather than evidence that the curriculum or its linked projects are secure.
How MrKeyoor tests explains the publisher's command recording and general limitations. The supplied methodology page was fetched on October 31, 2026; it is provenance context, not proof that its current procedure governed our September run. This correction uses the existing record and includes no new execution.
Start with an exercise, then change it
The README recommends running an offline test first, changing one thing and rerunning the test before moving on. The README at the recorded commit describes starter.py files as runnable references. That supports using an existing example as a starting point; it does not establish that solutions must be hidden or that assignments are graded.
A useful first session is therefore small: select your route, check its prerequisites, run the first offline example and make one change you can explain. When an exercise needs a model connection, the README says it explains the cloud or local path. Our run did not validate either path.
When another resource fits better
Datawhale Hello-Agents is a complement when you need full chapters and deeper practical work. The roadmap explicitly points readers there for chapter-length depth. Hello-Agents also describes prerequisites of basic Python and familiarity with calling a language model through an API. Keep the roadmap for sequencing and use the textbook when an explanation needs more room.
Microsoft AI Agents for Beginners is an alternative when you want written lessons, videos and Python examples using Microsoft's agent stack. Its README identifies Microsoft Agent Framework and Microsoft Foundry Agent Service V2, with an Azure account required for Microsoft Foundry. Choose it for that teaching format and stack, rather than assuming it is a shorter or provider-neutral substitute.
How to read the ratings
These editorial ratings assess the learning resource, not production-agent reliability. The scale is 1 for major documented obstacles, 2 for substantial gaps, 3 for a usable resource with material limits, 4 for strong documented support with remaining verification gaps, and 5 for comprehensive evidence within the stated scope. A dimension without enough evidence remains unassessed.
Setup ease is 5 only for inspecting the repository locally: our installation and build steps succeeded, and reading starts in the README. It does not rate exercise credentials, local-model hardware or deployment setup. Documentation is 5 for the documented routes, prerequisites, glossary, exercises and progress guidance; it does not certify every linked resource. Maturity is 4 for the structured curriculum and documented contribution checks, with exercise execution and long-term maintenance reliability unverified. Community is unassessed because the supplied evidence does not establish response times, contributor participation or sustained support.