AI-Assisted Development Certification
One paper across the whole course: what an agent is and how to keep control of one, the daily craft of context, prompting and verification, then configuration with CLAUDE.md, skills, hooks and MCP, and finally scaling across sessions and CI. Sat in one two-hour sitting.
- Sitting
- 2 hours
- Questions
- 60
- Pass mark
- 70%
- Price
- ₹199$9.99
Sit this when you can do these without looking them up
This is not a certificate of attendance. It is 60 questions in 2 hours at 70% with no partial credit and one attempt, and roughly a third of the paper is foundations — the certificate is for the whole language, not the parts you enjoyed.
- Say what a hook guarantees that the same rule written in CLAUDE.md does not
- Name what loads into context at the start of a session, and what only loads on demand
- Explain why a long session with three corrections in it often does worse than a fresh one
- Choose between a skill, a subagent and a hook for a given job, and justify the choice
- Describe two ways to make Claude verify its own work while you are not watching
- Say what a checkpoint will not restore, and what you still need git for
- Explain what auto mode's classifier is looking for, and what happens when it keeps blocking
If you'd rather find out than guess, the free 5-question readiness check reports a predicted band in two minutes. If several readiness points are unfamiliar, the free course covers everything on the syllabus.
Syllabus
Every paper is drawn to this mix, so the weights are a promise rather than a guide. The bars show each area's share of the 60 questions.
Agents and the loop
5 Q·8%What separates an agent from a chatbot, the gather-act-verify loop, the built-in tool categories, and where the human sits in the loop.
Setup and surfaces
5 Q·8%Installing and updating, starting a session in a project, the terminal against the IDE, desktop and web surfaces, and /init and /doctor.
Permissions and control
6 Q·10%The permission modes and what each allows, auto mode and its classifier, allowlisting, checkpoints and rewinding, and what a checkpoint cannot restore.
Context management
6 Q·10%What occupies the context window, why performance degrades as it fills, compaction and what survives it, /context and /compact, and clearing between tasks.
Prompting
5 Q·8%Scoping a task, pointing at sources and existing patterns, describing symptoms rather than diagnoses, file references and images, and when a vague prompt is right.
Verification
5 Q·8%Why looks-done is the default stopping condition, giving Claude a check it can run, goals and Stop hooks, adversarial review, and asking for evidence.
Planning and correction
4 Q·7%Plan mode and the explore-plan-code sequence, when planning is overhead, interrupting and steering, and when to abandon a session rather than correct it again.
CLAUDE.md and memory
5 Q·8%What belongs in CLAUDE.md and what does not, why a bloated file is ignored, file locations and imports, and how auto memory differs from what you wrote.
Skills and hooks
6 Q·10%Skill structure and on-demand loading, manual invocation, hook events and configuration, and the advisory-against-deterministic distinction between the two.
MCP and subagents
5 Q·8%What MCP connects and what it costs in context, adding and inspecting servers, subagent context isolation, restricting a subagent's tools, and delegating investigation.
Automation and CI
4 Q·7%Non-interactive runs and output formats, scoping tools for unattended work, fanning out across files, parallel sessions and worktrees, and review in CI.
Cost and security
4 Q·7%What drives token spend and the levers that reduce it, the threat model including prompt injection, sandboxing, and what to keep an unattended agent away from.
Three questions from the bank
Representative of the mix and difficulty — answer visible, shown here to calibrate expectations rather than test you.
- Sample 1Agents and the loop
A colleague argues that Claude Code is just a chat window with the same model behind it. Which observation is the strongest evidence that it is something different?
- It runs on your machine rather than in a browser tab
- It has been trained on your repository, so its answers are specific to your code
- It ran the test suite, read the failure and fixed it without anyone pasting the output back in
- It produces longer answers, because it is not limited to the length of a chat message
Why: An agent is a model plus tools plus the ability to read what those tools returned, and that last part is what lets it act, check and correct without you couriering results between the answer and your machine. Where it runs is packaging, and no model is trained on your repository — it reads the repository with tools, which is exactly the point.
- Sample 2Setup and surfaces
You installed Claude Code with Homebrew three months ago, and a feature the documentation describes is simply not there. What should you suspect first?
- The feature needs a Max plan and yours is Pro
- Homebrew installs stay on the version you installed, so you are months behind and need to upgrade the cask
- The background updater failed, and `claude doctor` will repair the installation
- A settings file is quietly disabling newer features
Why: Only the native install keeps itself current in the background; Homebrew, WinGet and the Linux package repositories leave you exactly where you were until you run their upgrade command. Assuming an auto-updater is running is how people end up months behind on a tool that ships constantly, describing behaviour no documentation matches.
- Sample 3Permissions and control
A hook's configuration refers to a permission mode called `default`, but the status bar in your session says Manual. What is going on?
- `default` means whichever mode your settings choose, which happens to be Manual
- `default` is a separate mode that only hooks and integrations can select
- Your version is too old to display the mode you are actually in
- They are the same mode: Manual is the label in the interface and `default` is the configuration value
Why: The mode that reviews every action is written `default` in configuration and shown as Manual on screen, and you will meet both spellings in the same week. Reading `default` as "whatever your settings say" is the trap, because it makes the value sound dynamic when it names one specific mode.
What this does NOT replace
An honest comparison, because overclaiming cert value helps nobody and gets caught by anyone who hires.
- This is not an AWS / Google / Microsoft vendor certification — those cover a proprietary platform, carry employer discounts, and some recruiters filter resumes against them. If a role explicitly names those, this doesn't substitute.
- This is not a replacement for a degree, a bootcamp with job placement, or shipping real work. Those produce different evidence. A certificate is a focused signal, not a career.
- This is a focused, honest signal that you can hold the full syllabus in your head under a clock. Pair it with a résumé that shows what you've built — that's the combination that lands interviews.
Everything on this syllabus is taught free
The Claude Code Masterclass course covers the whole syllabus above across 34 lessons, with a playground in each one. Work through it first — the exam is the same material under a clock.
Open the Claude Code Masterclass courseWhat sitting it is like
The exam opens full screen with a countdown, a question at a time, and a grid showing what you have answered and what you have flagged to come back to. Nothing is submitted until you say so, or the clock reaches zero.
Every answer saves to the server as you give it. If your laptop dies at question forty, sign back in and you will find question forty exactly as you left it — with however much time was left still ticking, because the clock does not pause.
Leaving full screen or switching tabs is noted but does not end your exam. It is a record, not a trap: nobody is auto-failed for a notification stealing focus.
When it ends you get your mark, a topic-by-topic breakdown, and every question with the right answer and an explanation — whether you passed or not. Passing also produces a certificate with a link anyone can verify.