Anthropic Certifications

Study kits for the Anthropic certification program that re-verify their own claims against current documentation — and quiz you on what you got wrong, not what you already know.

Certification material goes stale faster than anyone updates it. A guide written in July describes flags that were renamed in August, and you memorise the wrong thing. So this kit keeps a drift log — a running record of where the official documentation has moved since the exam guide was written — and a /refresh-kb command that re-checks it. When the tooling changes, the kit says so instead of quietly teaching you a dead flag.

The rest is what you would hope for: a linked wiki covering every task statement, a bank of original practice items where every wrong answer explains why it is wrong, and a set of Claude Code skills that teach, quiz, drill, and sit you a full timed mock exam.

See it before you install anything

A real 60-item mock exam produces this:

→ examples/mock-exam-report.md — percent correct by domain, which task statements you missed with the command to fix each, the reasoning habits you fell for ranked against their base rate, and a walkthrough of every item you got wrong.

The same report also publishes as a styled HTML page — examples/mock-exam-report.html (open it locally; GitHub shows HTML as source).

The sample has a domain sitting at 55% under a 77% average. That is left in deliberately. A strong average hiding one collapsed domain is the exact failure the report exists to catch, and a sample where everything went well would demonstrate nothing.

Unofficial — and specific about it

Built by a Claude Ambassador, a member of Anthropic's official community program. That is a community role, not an Anthropic one. Concretely:

• Not written, reviewed, endorsed, or sponsored by Anthropic
• Contains no exam content — no live questions, and not even the sample questions printed in the official guide
• Every practice item is original, written from the published objectives

If you have sat the exam, you are under an NDA. Do not contribute anything you saw on it. See DISCLAIMER.md — it is short and it is the honest version.

Certifications

One today. The goal is all of them. The repo is built as a monorepo precisely so that the second certification is content, not plumbing: the skills, the scripts, the schema, and the validator are already shared and certification-aware. Adding one means writing a wiki and a question bank, not rebuilding the machinery.

If you are studying a certification that is not here yet, that is the single most valuable thing you could contribute — and you would be building it with the same tools you study with. See Adding a certification.

Quick start

From the plugin marketplace — the recommended route. Inside any Claude Code session:

/plugin marketplace add alexiocassanifm/anthropic-certifications
/plugin install cert-study-kit@anthropic-certifications
/cert-study-kit:setup

setup asks where to put the kit — defaulting to a folder in the directory you are already in, the same place a hand-clone would land — clones it there, checks the Python scripts can run, and hands you back with a cd and a claude. It warns you first if that directory sits inside another git repository or under a sync folder like iCloud Drive. Nothing to look up and nothing to get wrong, and the marketplace tells you when there is a newer version.

Or clone it yourself, if you would rather see every step:

git clone https://github.com/alexiocassanifm/anthropic-certifications.git
cd anthropic-certifications
pip install -r requirements.txt      # PyYAML, for the scripts
claude                               # open Claude Code at the repo root

Both routes end in the same place: the kit as a git repository you own and study in. That is deliberate — the plugin is the front door, not the kit. See Why the kit is a repo, not a plugin payload.

Then, in the clone:

/study 1.4          # learn one task statement, then get checked on it
/quiz --domain 2    # short targeted quiz, feedback per item
/mock-exam          # full 60-item timed simulation + diagnostic report
/progress           # readiness dashboard
/drill              # spaced repetition on what you are forgetting
/refresh-kb         # re-verify the wiki against current official docs
/author-question    # draft a new practice item, schema-checked

Run them from the repo root. The skills resolve which certification you mean — the only one present, or the one you name.

/mock-exam ends by offering to publish its diagnostic as an HTML or Markdown artifact, so a sixty-item report does not evaporate with your terminal scrollback.

What makes it different

The explanations are the study material. Every option in the bank carries a why. Getting an item right without knowing why teaches you nothing, so the report tells you why the distractor was tempting as well as why the answer was right.

Wrong answers are taxonomised. Seven named distractor families — prompt-instead-of-enforcement, blames-wrong-component, unreliable-proxy, and four more. Reports rank the ones you fall for by rate against how often each appeared, not by raw count, because the most common family collects the most misses on any form and tells you nothing. The rate is what names a real reasoning habit.

It knows when it is out of date. A SessionStart hook checks how long ago each wiki was verified and offers /refresh-kb. No network calls, prompts at most once a day, prints nothing when fresh, and never fails a session.

It admits what it cannot know. The exam passes at a scaled 720 out of 1,000, and the raw percentage that maps to 720 varies by form. So this kit never emits a scaled score — a fabricated "you scored 743" invites you to stop studying at exactly the wrong moment. You get percent-correct by domain, which is what the real score report gives you anyway.

Contributing

Practice items, corrections, and new certifications are all welcome — see CONTRIBUTING.md.

The fastest way in is to let the kit write with you:

/author-question 3.4

It interviews you, drafts an item against the schema and style guide, validates it, and appends it to the right bank file. You supply the production experience; it handles the shape.

The one hard rule: never contribute real exam content. Write from the published objectives and from what you have actually built.

Layout

├── certifications/
│   ├── README.md            conventions + how to add one
│   └── ccar-f/
│       ├── wiki/            98 linked notes — tasks, concepts, scenarios, heuristics
│       ├── questions/       schema, style guide, item bank
│       └── progress/        your local state (gitignored)
├── examples/                sample mock exam report
├── .claude/skills/          seven shared, certification-aware skills
├── scripts/                 validator + mock exam builder
├── .claude-plugin/          marketplace catalogue for cert-study-kit
├── plugin/cert-study-kit/   the front-door plugin — clones this repo, holds nothing
├── DISCLAIMER.md            applies to every kit here
├── CONTRIBUTING.md
└── LICENSE                  MIT (code) + CC BY-SA 4.0 (content)

Shared at the root: skills, scripts, licence, disclaimer, contributing guide. Per certification: wiki, questions, progress.

Why the kit is a repo, not a plugin payload

Six of the seven skills write. /study, /quiz, /drill, and /mock-exam record your progress; /author-question appends to the question bank; /refresh-kb rewrites wiki notes and the drift log. Progress is personal state and could live anywhere. The bank and the wiki are versioned content — correcting a note that has drifted is worth something only if the correction lands in a working tree you can commit, and open a pull request from.

Installing a plugin copies it into ~/.claude/plugins/cache/. Anything written there is discarded by the next /plugin update and can never reach upstream, so the two things this kit leads with — noticing when it is out of date, and letting you contribute — would both be quietly broken. A read-only study kit would make a fine plugin payload. This one is not read-only by design.

So the plugin ships the front door and the repository holds the kit: cert-study-kit contains exactly one skill, /cert-study-kit:setup, which clones this repo where you want it. The seven study skills stay project-scoped in the clone, which is why they are /study and not /cert-study-kit:study.

Hooks require workspace trust, so the first claude in a fresh clone asks you to approve the folder. Declining is fine — everything still works, you just run /refresh-kb yourself.

Scripts

python scripts/validate_questions.py --cert ccar-f     # schema + style checks
python scripts/build_exam.py --cert ccar-f --seed 1    # assemble a form
python scripts/check_verification_age.py --force       # see what the hook would say

With one certification present it is the default; with several, --cert is required — guessing would silently validate the wrong bank. build_exam.py always reports the seed it used, with or without --seed, so any form can be rebuilt afterwards to re-grade an exam or check a report against the form it came from.

Reading the wiki without Claude

Everything under certifications/<slug>/wiki/ is plain Markdown with relative links, so it renders on GitHub and navigates in Obsidian. To use an existing vault:

ln -s "$(pwd)/certifications/ccar-f/wiki" /path/to/vault/ccar-f

Licence

Code is MIT. Wiki and question content are CC BY-SA 4.0. See LICENSE.

_{If this saved you time, a ⭐ helps other people studying the same exam find it.}