CCDV-F Prompt and Context Engineering Practice Test 3 — 15 Questions

Question 1

You are integrating Claude 3.5 Sonnet into a customer support application that maintains long, multi-turn conversations. After about 40 turns, you notice Claude begins contradicting policy details it stated earlier in the same session, even though the policy text is still included in the system prompt. The conversation history is passed in full on each request. What is the most effective structural change to preserve instruction adherence across the session?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Restate the critical policy instructions in the system prompt and periodically re-inject a condensed reminder of them in recent turns.. Instruction adherence degrades in long conversations because relevant constraints become diluted across many tokens. Keeping durable rules in the system prompt maintains their authority, and re-injecting condensed reminders near the current turn restores salience so the model honors them. Together these preserve consistency without abandoning the conversational history the application relies on.

Answer

Increase the max_tokens parameter so Claude has more room to generate responses that include the policy details.

Answer

Move the policy text from the system prompt into the first user message of the conversation.

Answer

Enable extended thinking so Claude reasons about the policy each time before responding.

Question 2

A developer is building a pipeline that summarizes a 200-page technical manual with Claude Sonnet. The manual is too long for a single request, so the developer splits it into 40 chunks and summarizes each chunk independently. The final summaries must remain factually consistent with one another and with the source, and the developer has a fixed token budget. Which two techniques should the developer apply to keep the chunk summaries consistent and grounded? (Choose two.)

Accepted Answer

Carry a running 'state' summary of key entities, definitions, and decisions forward into each subsequent chunk request.. Cross-chunk consistency requires either shared context or a reconciliation step. Carrying a running state forward gives each request the key facts established earlier, while a verification pass that reconciles adjacent summaries catches contradictions before they propagate. Together they maintain grounding and consistency within a bounded token budget, unlike duplicating output or resending all prior text.

Answer

Include the full text of all previously processed chunks in every new request.

Answer

Ask the model to summarize each chunk twice and concatenate both outputs to increase coverage.

Answer

Lower the temperature to zero and rely on deterministic sampling to guarantee identical facts across chunks.

Question 3

A developer is building a customer-support agent on the Claude Messages API. The agent receives a long conversation history plus a retrieved knowledge-base article, and the developer wants Claude to answer only from the retrieved article while still seeing the full chat for tone. The developer wants the retrieved article treated as the most authoritative source, even if earlier turns contain outdated policy statements. Where should the retrieved article be placed, and how should it be marked?

Accepted Answer

Place the retrieved article inside the system prompt, wrapped in tags, and instruct Claude that content overrides any conflicting statements in the conversation.. The system prompt is the operator-controlled channel and carries more weight than user or assistant turns, making it the right place to establish that the retrieved article is authoritative. Tagging the article with XML-style delimiters gives Claude an unambiguous boundary for the trusted content, so outdated policy in earlier conversation turns does not win.

Answer

Insert the retrieved article into a tool_result block from a search tool and instruct Claude to prefer tool results over all other content.

Answer

Place the retrieved article as the first user turn, followed by the conversation history, and rely on Claude's recency bias to prefer the article.

Answer

Append the retrieved article as the final assistant turn so Claude treats it as its own prior statement and repeats it verbatim.

Question 4

A developer is building a customer-support assistant that must answer only from a provided knowledge base article and must refuse to answer when the article does not contain the information. The assistant currently invents plausible answers. Which prompt change best enforces the refusal behavior?

Accepted Answer

Instruct the model to answer only using information found in the article and to respond with a fixed phrase such as 'I don't have that information' when the article is insufficient.. Grounding requires two explicit constraints: define the permitted source and define what to do when that source is inadequate. Naming the article as the only allowed source, plus a fixed refusal phrase for missing information, gives the model an unambiguous rule and produces a consistent, detectable abstention. Tone, token limits, and document placement do not establish either constraint.

Answer

Increase the max_tokens parameter so the model has more room to explain its reasoning.

Answer

Move the knowledge base article to the end of the prompt so it is the last thing the model reads.

Answer

Add an instruction to answer in a polite and professional tone at all times.

Question 5

A developer is building a pipeline that uses Claude to convert free-text incident reports into a strict JSON object with fields severity, services, and summary. During testing, roughly 8 percent of outputs include a friendly preamble such as 'Sure, here is the JSON:' or wrap the object in markdown code fences, which breaks the downstream parser. The developer has already described the schema precisely in the prompt. What is the most reliable next step to eliminate the malformed outputs?

Accepted Answer

Add a few-shot example showing the exact JSON output with no preamble, and pre-fill the assistant turn with an opening curly brace so Claude must continue the object.. Pre-filling the assistant turn with an opening curly brace forces the model to begin inside the object, making a preamble or code fence structurally impossible. Pairing that constraint with a few-shot example that shows the exact no-preamble form teaches the desired surface pattern and covers cases where the schema itself is ambiguous.

Answer

Lower the temperature to zero and add the sentence 'Respond only with JSON' to the end of the user message.

Answer

Wrap the schema description in <json_schema> tags and instruct Claude to output only what matches the schema.

Answer

Increase max_tokens so the model has room to finish the JSON object without truncation.

Question 6

A developer is building a customer-support bot with the Claude Messages API. The system prompt instructs Claude to answer concisely, but the bot also needs to return a stable JSON object with the fields 'intent', 'sentiment', and 'reply' on every turn. Where should the instruction to produce this JSON object be placed to most reliably control the response format?

Accepted Answer

In the system prompt, placed after the conciseness instruction.. The system prompt is the correct place because it is sent with every Messages API request and reliably conditions all assistant turns. Interface requirements such as JSON keys should live there so they are not diluted by user content. Putting format rules in the user turn or in a lone assistant example makes them fragile and easy to override.

Answer

Only in the first assistant message as an example.

Answer

Appended to each user message as a trailing reminder.

Answer

In a separate follow-up call after Claude replies in prose.

Question 7

A developer is using the Messages API to have Claude return a JSON object describing a product. The response sometimes includes a conversational preamble such as 'Here is the JSON you requested:' before the object, which breaks the downstream parser. What is the most reliable way to eliminate the preamble?

Accepted Answer

Use the assistant turn prefill with an opening curly brace so the response must continue from that character.. Prefilling the assistant turn with an opening brace removes the space where a preamble could be generated, forcing the response to begin mid-JSON. Instructions and temperature settings only bias behavior probabilistically, and token limits address length rather than structure. Prefill is the only option that structurally guarantees the response starts with the expected character.

Answer

Increase max_tokens so the model has enough room to output the full JSON without truncation.

Answer

Set the temperature to zero so the model always produces the same output.

Answer

Add 'Do not include any preamble' to the system prompt and hope the model complies.

Question 8

A developer is writing a support-triage prompt for Claude. The prompt contains a 12-page product manual followed by the customer's question. Testing shows Claude sometimes answers using general knowledge about competing products instead of the manual. The developer wants Claude to ground every answer in the manual and to say 'not covered' when the manual lacks the answer. Which change best achieves this?

Accepted Answer

Instruct Claude to quote the relevant manual passage before answering, and to reply 'not covered' if no passage supports the answer.. Grounding improves when the model must produce evidence before its conclusion, because the quoted passage anchors the answer in the manual. Giving an explicit 'not covered' response removes the pressure to invent an answer, which is what allows general knowledge to leak in. Together these two elements address both the sourcing and the fallback behavior.

Answer

Shorten the manual to its first two pages so the context is easier for Claude to process.

Answer

Move the customer question to the top of the prompt so Claude reads the question before the manual.

Answer

Add the sentence 'Be accurate and do not hallucinate' to the end of the prompt.

Question 9

A developer is using Claude to review a 180-page contract. The model must cite the exact page number for each risk it flags. Placing the entire contract in the prompt causes the model to cite vaguely or omit page numbers. Which change best improves citation accuracy without exceeding the context window?

Accepted Answer

Insert explicit page-marker tags such as '[[PAGE 42]]' before each page's text and instruct the model to cite the nearest marker.. Embedding page markers makes citation a local pattern-matching task instead of a global spatial inference. The model can still read the whole contract in one pass, preserving cross-reference reasoning, while the nearest marker gives an unambiguous page label. Temperature changes and summary-then-cite pipelines do not supply the missing positional signal.

Answer

Ask the model to summarize each page first, then cite the summary.

Answer

Lower the temperature to 0 so the model copies page numbers verbatim.

Answer

Split the contract into 180 separate API calls, one per page.

Question 10

A developer maintains a Claude-powered assistant that answers questions over a 60,000-token internal policy corpus. The corpus is stable and reused across every request, and the developer wants to cut cost and latency while preserving answer quality. Which two changes will most directly reduce per-request token processing for this workload? (Choose two.)

Accepted Answer

Retrieve only the policy sections relevant to each question and place those sections in the prompt instead of the full corpus.. Prompt caching avoids reprocessing the unchanged corpus prefix, and retrieval shrinks the context to only the passages a given question needs. Both act directly on the number of tokens the model must process per request, and both preserve quality because the relevant policy content remains available to the model.

Answer

Raise the temperature so Claude explores a wider range of possible answers and settles faster.

Answer

Add a system instruction telling Claude to read the corpus more quickly to save tokens.

Answer

Convert the corpus to a single long paragraph with no headings to reduce whitespace tokens.

Question 11

A developer is designing a prompt that asks Claude to extract action items from meeting transcripts. The transcripts are noisy, with overlapping speakers and side conversations. Which TWO prompt-engineering practices will most improve the reliability of the extracted action items? (Choose two.)

Accepted Answer

Instruct the model to think step by step internally before emitting the final list, then output only the list.. Few-shot examples anchor the noisy-input-to-clean-output mapping, and asking for internal reasoning before the final list lets the model resolve attribution and filter side talk. Together they address the two hard parts of this task: deciding what counts as an action item and deciding who owns it. Output limits and label removal do not improve either decision.

Answer

Remove all speaker labels to reduce token count before sending the transcript.

Answer

Set the max_tokens parameter as high as possible so the model never truncates an action item.

Answer

Ask the model to include every sentence that mentions a task, regardless of who said it.

Question 12

You are building a support-triage assistant on the Anthropic API. Each request must return a JSON object with keys 'category' and 'priority'. During testing, Claude wraps its output in markdown fences and adds a friendly sentence before the JSON. You want the raw, parseable object every time without changing the model or adding a second call. What is the most reliable prompt-level change?

Accepted Answer

Pre-fill the assistant turn with an opening brace so the response must continue as the JSON object.. Constraining the assistant turn with a leading brace forces continuation as the JSON object, since the model must complete the already-started structure. This removes markdown fences and conversational preamble in a single call, with no model change. Instruction-only or explanation-first approaches leave room for stray text, so prefilling is the most reliable prompt-level fix for guaranteed parseable output.

Answer

Ask Claude to explain its reasoning first, then append the JSON at the end of the response.

Answer

Add the sentence 'Please output only JSON' to the system prompt and rely on that instruction alone.

Answer

Increase the temperature setting so the model has more freedom to choose a clean format.

Question 13

A developer is building a document-review assistant that must extract every monetary figure from contracts and cite the page where each figure appears. The contracts average 40 pages. Early tests show Claude misses figures that appear in tables and occasionally cites the wrong page. The developer wants to improve recall and citation accuracy without changing the model. Which approach is most effective?

Accepted Answer

Ask Claude to first list all candidate monetary figures with their page numbers, then verify each candidate against the source text before producing the final extraction.. Separating candidate generation from verification gives the model an explicit intermediate list it can check against the source, which catches table figures that a single extraction pass tends to skip. Verifying each page citation against the text before finalizing corrects misattributed page numbers, and the whole workflow stays within one model.

Answer

Instruct Claude to be thorough and to pay special attention to tables and page numbers in the contract.

Answer

Increase max_tokens so Claude has room to output more figures and longer citations.

Answer

Split each contract into 40 separate single-page prompts and merge the extracted figures afterward.

Question 14

A developer is iterating on a classification prompt for support tickets. Each test run uses a different random sample of 200 tickets, and accuracy swings by 8 points between runs. The prompt itself is unchanged. What is the best first step to get trustworthy signal about whether a prompt edit actually helped?

Accepted Answer

Freeze a held-out evaluation set and run every prompt version against the same fixed examples.. A fixed held-out evaluation set is the foundation of reliable prompt iteration. By scoring every prompt version against identical examples, the developer isolates the effect of the prompt change from the effect of sampling. Temperature and example count affect generation, but neither controls which inputs are being compared.

Answer

Increase max_tokens so the model has more room to reason before classifying.

Answer

Add more few-shot examples until accuracy stops changing between runs.

Answer

Lower the temperature to 0 and rerun the same random sample twice.

Question 15

A developer needs Claude to transform a list of product descriptions into a fixed XML schema that a downstream parser expects. The model sometimes adds a friendly introductory sentence before the XML. Which change most directly eliminates the extra prose?

Accepted Answer

Add a clear instruction that the response must begin immediately with the opening XML tag and contain nothing else.. Explicit output-format constraints are the direct fix for unwanted preamble. Telling the model exactly where the response must begin, and that nothing may precede it, removes the ambiguity that lets conversational framing leak in. Temperature, input length, and added reasoning all fail to target the specific behavior.

Answer

Shorten the input descriptions so the model has less to respond to.

Answer

Increase temperature so the model varies its phrasing and may omit the introduction.

Answer

Ask the model to explain each transformation step before producing the XML.