The prompting docs got consolidated
Here is the thing that will bite you before any exam question does: the prompting documentation was restructured, and most of what people quote at you online describes the old layout.
There used to be roughly ten separate pages, one per technique. Be clear and direct had a page. Use examples had a page. XML tags had a page. Chain of thought had a page. People bookmarked them, cited them, built courses on them.
They are gone. Not deleted, but not pages either. Twelve legacy slugs now 308-redirect into anchors on a single consolidated page: claude-prompting-best-practices.
What the docs actually say now
The docs are blunt about it. Their own words: “All prompting techniques are covered in Prompting best practices. That’s the living reference; start there.”
Think of it as a monorepo migration. Ten packages, each with its own README and its own release cadence, collapsed into one package with sections. Nothing was deleted, every old import path still resolves, and the thing you actually have to update is your mental index of where code lives. That is the whole change, and it is why bookmarks still work while advice built on them has gone subtly stale.
A 308 is a permanent redirect, which is the strongest signal the web has for “this thing moved and is not coming back”. Twelve of them pointing at one page is not a tidy-up. It is a decision that the technique-per-page taxonomy was the wrong shape.
The order of the page is the lesson
This part is worth memorising, because the order tells you what the docs think matters most.
Model-specific first. Four pages: Fable 5, Sonnet 5, Opus 5, Opus 4.8. Before any general principle, they want you looking at advice for the specific model you are calling. That is a real change in emphasis and we spend a whole lesson on it.
Then General principles, which is where the classic material lives: be clear and direct, add context, use examples effectively, structure with XML tags, give Claude a role, long-context prompting, and model self-knowledge.
Then, in order: Output and formatting (which contains the migration guidance for prefilled responses, and that one is a trap we cover in Module 3), Tool use, Thinking and reasoning, Agentic systems, Capability-specific tips, and Migration.
Learn the section names rather than the sentences inside them. Anything you can find in twenty seconds does not need to be memorised, and a page you can navigate is worth more than a page you have read once.
The generator you cannot use any more
While you are updating your map: the Prompt Tools API is gone. generate, improve and templatize are fully delisted, and the URL now 307-redirects to a stub.
If you learned prompting from material written eighteen months ago, you may have “call the improve endpoint” filed as a real workflow. It is not. The only surviving generator is a Colab metaprompt notebook, which is a notebook, not an API you can call from a service.
This is exactly the kind of thing an exam likes, because it separates people who read the docs this quarter from people who read them once and cached the result. The trap answer is always “deprecated but working”, because that is the polite decay everyone expects. Delisted is not deprecated.
Try it yourself
Where the ten technique pages went
This card is about the shape of the documentation rather than any one technique. It is the fact that most study material written before the restructure gets wrong by omission.
What happened to the individual prompting-technique documentation pages, and where did they go?
Reveal answer
They no longer exist as separate pages. Twelve legacy slugs now 308-redirect into anchors on a single consolidated page, claude-prompting-best-practices, which the docs describe as the living reference and the place to start. A study guide built on the old one-page-per-technique taxonomy is navigating a map that no longer matches the territory.
The Prompt Tools API
A colleague suggests you use the Prompt Tools API to generate and improve your prompts programmatically. What is your response?
Show answer
Correct answer: B — It is fully delisted; the URL 307s to a stub and the only surviving generator is a Colab metaprompt notebook
The Prompt Tools API (generate, improve, templatize) is fully delisted and its URL now 307-redirects to a stub. Option 3 is the tempting one because deprecated-but-working is the normal shape of API decay and it is what you would assume. Here the endpoint is gone, not softened, so any plan that routes through it is dead on arrival. The metaprompt Colab notebook is what survives.
Reading the redirects
Two requests against the documentation site. Paths are shortened for readability, and the status codes are the point.
GET .../use-xml-tags
308 location: .../claude-prompting-best-practices#xml-tags
GET .../prompt-tools-generate
307 location: .../a-stub-pageWhat do these two responses tell you about the state of each resource?
Show answer
Correct answer: B — The XML page moved permanently into an anchor on the living reference; the 307 lands on a stub, not a replacement
A 308 is a permanent redirect, the strongest signal the web has for content that moved and is not coming back, and twelve of them point at the one consolidated page. The 307 on Prompt Tools is a different situation: the destination is a stub, not a working endpoint at a new address. Option 4 is the tempting one because a temporary redirect usually implies the thing still exists somewhere, and that inference is exactly what makes people plan around an API that is gone. Read where a redirect lands, not only what code it returns.
The shape of the living reference
Worth knowing cold because the ordering is deliberate, and the section sitting at the very top is the one most people scroll straight past.
What is the top-level structure of the consolidated prompting best-practices page?
Reveal answer
Four model-specific pages come first (Fable 5, Sonnet 5, Opus 5, Opus 4.8). Then General principles, then Output and formatting, then Tool use, then Thinking and reasoning, then Agentic systems, then Capability-specific tips, then Migration.
Find three things in two minutes
A navigation drill, not a reading exercise. Open the consolidated best-practices page and race the clock. Knowing which section holds a fact is worth more than half-remembering the fact.
- Question one: your tool descriptions are causing the model to call tools it should not. Name the section you would scroll to, before you scroll.
- Question two: how many examples should a few-shot prompt carry. Name the section.
- Question three: you need the migration guidance for prefilled responses. Name the section.
- Now check all three, and note which of the three lives under a model-specific page rather than under General principles.