Anthropic CCAR-F 問題集

質問 1：
You are integrating Claude Code into your Continuous Integration/Continuous Deployment (CI/CD) pipeline. The system runs automated code reviews, generates test cases, and provides feedback on pull requests. You need to design prompts that provide actionable feedback and minimize false positives.
Your automated review calls the Claude API for each pull request, using tool_use with a report_findings tool that returns a JSON array of finding objects. Each object contains file_path, line_number, severity, category, and description. During testing on a large pull request touching more than 30 files, the response reaches the max_tokens limit and is truncated in the middle of the JSON, causing your pipeline's parser to fail.
What is the most effective way to handle this?
A. Increase max_tokens to the model's maximum and instruct Claude to keep each finding description under 50 words.
B. Switch from tool_use to prompting Claude to return findings as a Markdown list.
C. Add retry logic that detects truncated JSON and resends the request with instructions to report only critical and high-severity findings.
D. Split the review into multiple API calls that each analyze a subset of the changed files, and then merge the resulting findings arrays.
正解：D
解説: (Topexam メンバーにのみ表示されます)

質問 2：
You are building a structured data extraction system using Claude. The system extracts information from unstructured documents, validates the output using JavaScript Object Notation (JSON) schemas, and maintains high accuracy. It must handle edge cases gracefully and integrate with downstream systems.
Testing reveals that when source documents are missing certain specifications, the model fabricates plausible-sounding values to satisfy your schema's required fields. For example, a document mentioning only dimensions receives a fabricated "weight: 2.3 kg" in the extraction output.
What schema design change most effectively addresses this hallucination behavior?
A. Implement semantic validation that verifies each extracted value appears in or can be inferred from the source document text.
B. Add explicit instructions to the prompt stating "only extract information explicitly stated in the document; use placeholder text for missing values."
C. Add a "confidence" field alongside each specification where the model self-reports its certainty, then filter out low-confidence extractions.
D. Change fields that may not exist in source documents from required to optional, allowing the model to omit them.
正解：D
解説: (Topexam メンバーにのみ表示されます)

質問 3：
Production reviews reveal inconsistent handling of uncertainty in final reports. Sometimes conflicting subagent findings are synthesized into a single confident statement (losing nuance), while other times reports over-hedge with excessive qualifications (becoming unhelpful). When the web search agent returns "industry analysts estimate $50B market size (methodology varies)" and the document analysis agent returns "peer-reviewed study estimates $35B (±$7B, 95% CI)," the coordinator either picks one arbitrarily or produces vague statements like "the market may be
$35B-$50B depending on factors." What systematic approach best addresses this?
A. Implement a confidence calibration layer that normalizes subagent uncertainty expressions to standardized probability scores (0.0-1.0), then weight-average findings by their calibrated confidence.
B. Instruct the synthesis agent to structure reports with explicit sections distinguishing well- established findings from contested ones, preserving original source characterizations and methodological context.
C. Add a verification subagent that cross-references findings across sources, only passing claims to synthesis that are corroborated by at least two independent sources.
D. Configure subagents to only report findings meeting a high-confidence threshold, filtering uncertain information before it reaches the coordinator.
正解：B
解説: (Topexam メンバーにのみ表示されます)

質問 4：
Your pipeline runs:
PROMPT="You are a code reviewer."
PROMPT="$PROMPT Analyze the provided diff"
PROMPT="$PROMPT for bugs, security issues,"
PROMPT="$PROMPT and style violations."
claude -p \
--dangerously-skip-permissions \
--system-prompt "$PROMPT" < diff.txt
The reviews complete and return feedback, but Claude comments only on the piped diff--it never reads surrounding files in the checked-out repository to understand broader context, even when the diff modifies a function called by many other modules. Which change to the invocation will cause Claude to read related repository files while still applying your custom review instructions?
A. Keep --system-prompt and add --allowedTools "Read, Glob, Grep" because non-interactive -p mode otherwise disables filesystem tools.
B. Replace --system-prompt with --append-system-prompt so the review instructions are added to Claude Code's default prompt instead of overwriting its built-in file-reading and code-navigation guidance.
C. Remove --system-prompt entirely and place the review instructions in a root-level CLAUDE.md because --system-prompt is incompatible with tool use under -p.
D. Stop piping the diff through standard input and embed it in the prompt string so Claude Code treats the invocation as an agentic session rather than a stream-processing operation.
正解：B
解説: (Topexam メンバーにのみ表示されます)

質問 5：
You are integrating Claude Code into your Continuous Integration/Continuous Deployment (CI/CD) pipeline. The system runs automated code reviews, generates test cases, and provides feedback on pull requests. You need to design prompts that provide actionable feedback and minimize false positives.
Your automated review generates many findings per pull request, but developer feedback shows that approximately half are dismissed as "not worth addressing." Analysis reveals that these findings are often technically accurate but involve minor style preferences or patterns that are acceptable in the project.
Before adding infrastructure complexity, what prompt-design change would most effectively reduce dismissals while maintaining detection of genuine issues?
A. Ask Claude to rate each finding's confidence from 1 to 10 and include only findings rated 8 or higher.
B. Add a secondary classification model that filters findings according to predicted developer acceptance.
C. Define explicit reporting criteria that distinguish reportable bugs and security issues from minor style preferences and accepted local patterns.
D. Add the instruction: "Only report findings you are highly confident are genuine problems."
正解：C
解説: (Topexam メンバーにのみ表示されます)

質問 6：
You are building a structured data extraction system using Claude. The system extracts information from unstructured documents, validates the output using JSON schemas, and maintains high accuracy. It must handle edge cases gracefully and integrate with downstream systems.
After deployment, you find that 12% of extractions contain semantic errors that pass JSON Schema validation--for example, a duration such as "30 minutes" is incorrectly placed in an ingredient-quantity field. Human reviewers have the capacity to check only 20% of extractions.
Which approach most effectively allocates reviewer attention?
A. Randomly sample 20% of extractions for review, using corrections to track accuracy and identify error patterns.
B. Have the model output field-level confidence scores, and then calibrate review thresholds using a labeled validation set.
C. Prioritize the review of all extractions where required fields are empty or explicitly marked as not found.
D. Review all extractions from documents with formatting anomalies, such as unusual layouts or mixed content types.
正解：B
解説: (Topexam メンバーにのみ表示されます)