Claude Certification Blog
Prompt and context engineering: the two skill names only CCDV-F has
The developer exam is the only one that names context engineering and output handling as skills, and it weights them at 3.8% and 2.6% — smaller than the discourse around them suggests.
Prompt and Context Engineering is 11% of CCDV-F — about 6 of 53 items — across three published skills: Prompt Engineering at 4.6%, Context Engineering at 3.8% and Output Handling at 2.6%. Two of those three names appear on no other blueprint in the programme, which makes this the only exam that treats what you put in and what you get back as separate examinable things.
Three skills, eleven percent
| Skill | Weight | What it covers |
|---|---|---|
| Prompt Engineering | 4.6% | Instruction, examples and structure, applied one at a time |
| Context Engineering | 3.8% | What survives compaction and what has to live outside it |
| Output Handling | 2.6% | Structure against prose, and what a schema guarantees |
Two names that exist only here
Search all four objective lists for context engineering and output handling and they appear once each, both on this track. CCAR-F names structured output inside a domain title; the associate and professional blueprints name neither.
That is worth more than a trivia point. It tells you the developer exam has decomposed the per-call problem further than the others — into shaping the instruction, shaping what surrounds it, and handling what comes back — and that a candidate arriving from another track will have studied all three as one subject.
Context engineering is 3.8%
Roughly two items, for the subject that has probably generated more writing in the last year than anything else on this blueprint. The examinable content is narrow and durable: what survives when a long history is compressed, and what has to live somewhere else.
Compression is lossy in a known direction. The topic survives; the particulars — an amount, a date, an identifier, an exact commitment — are what disappear, and nothing announces it. So the fix is structural rather than linguistic: anything that must stay exact belongs in a field something reads, not in a narrative something summarises. A better summarisation instruction is the recurring distractor, and the working version of this is in context management for long sessions.
Attention is uneven, so layout is a decision
What sits at the two ends of a long input gets read more dependably than whatever is buried in the middle of it. That makes the order you assemble a request in a design choice with a measurable effect, and it is also why caching rewards putting the stable part first — the same arrangement, paying twice.
What a schema does not promise
Output Handling is the smallest skill in the domain at 2.6%, and it carries a distinction that decides items well beyond it. A schema constrains the shape of a response. It says nothing about whether the contents are true, complete, or the result of a generation that actually finished.
So a response can validate perfectly and still be wrong in two different ways, and the objective pairs structure with stop reasons for exactly that reason: what a response says about why it ended is available before you parse a character of it. The other half — enforcing shape through tool use rather than asking for a format in prose — is in tool use and structured output.
The prompt skill is the largest
At 4.6% Prompt Engineering is the biggest of the three, and CCDV-F is the only track publishing that figure in isolation — everywhere else prompting sits inside a domain that covers more. Its examinable shape is the same across the programme: a named failure, and the single technique that addresses that failure specifically.
Which means the phrase “write a better prompt” answers nothing here. The cross-track version, including which techniques each guide names, is in prompt engineering on the Claude exams.
How to spend six items
Two evenings at most, and the second one shorter. Prompt engineering and output handling together first, because they are the two halves of a single call and the schema distinction is the highest-value thing in the domain. Context engineering after, aimed narrowly at what compression loses.
Then leave it. Eleven percent is real and it is a third of what Applications and Integration carries alone, and this is the domain where reading feels most productive because there is so much written about it. The published figures are the correction.
Key takeaways
- Eleven percent, six items, three skills. Prompt engineering 4.6%, context engineering 3.8%, output handling 2.6%.
- Two of the names exist only here. No other blueprint names context engineering or output handling as skills.
- Compression loses particulars. Amounts, dates and identifiers go; the topic stays, and nothing raises an error.
- A schema promises shape, not truth. A perfectly valid response can be truncated or invented.
- Check why generation ended first. The response says so before you parse any of its content.
- The reading-to-weight ratio is worst here. More has been written about context engineering than about anything else worth two items.
The most-discussed subject on the paper is worth two items
That is a hard correction to apply from a blueprint alone, because the reading is genuinely interesting and the weight is not. A timed paper on the published 53-item allocation makes the argument in a way a percentage table cannot. Our claude certification study guide covers how to sequence them.
See the CCDV-F blueprintQuestions
Frequently asked
The follow-up questions people search next.
How much of CCDV-F is prompt and context engineering?
Domain 6 is 11% of the exam — about 6 of 53 items — split into Prompt Engineering at 4.6%, Context Engineering at 3.8% and Output Handling at 2.6%.
Do the other Claude exams name context engineering?
Not as a skill. CCDV-F is the only track where context engineering and output handling appear as named units; CCAR-F names structured output inside a domain title, and neither the associate nor the professional blueprint names either one.
What does a JSON schema actually guarantee?
The shape of a response, and nothing about whether its contents are true or complete. A response can validate perfectly against a schema and still be a truncated answer or a confident invention, which is why the objective pairs structure with checking why generation ended.
What is examined about context engineering?
What survives a compaction and what does not. Compression keeps the topic and loses the particulars — amounts, identifiers, exact commitments — so anything that must stay exact belongs somewhere the compression cannot reach rather than in the narrative.
Is output handling worth studying for one or two items?
Yes, because it is short and the judgment transfers. The distinction between a shape guarantee and a truth guarantee decides items in this domain and appears again in application design, which is the heaviest domain on the paper.
Keep reading
Related posts
Not affiliated with, or endorsed by, Anthropic or Pearson VUE. Details are summarised from publicly published program information and can change — always confirm against the official exam guide before booking.