CertForge

1 / 10 single answer hard

Your team created a `/analyze-codebase` skill that performs deep code analysis—dependency scanning, test coverage counts, and code quality metrics. After running the command, team members report Claude becomes less responsive in the session and loses the context of the original task. How do you most effectively fix this while keeping full analysis capabilities?

A. Split the skill into three smaller skills, each producing less output. B. Add instructions to the skill to compress all results into a short summary before displaying them. C. Add `model: haiku` in frontmatter to use a faster, cheaper model for analysis. D. Add `context: fork` in the skill frontmatter to run the analysis in an isolated subagent context.

2 / 10 single answer medium

A retail POS team commits project settings that allow read-only package inspection and deny access to production credential files. A developer adds a local allow rule for one of those credential files to troubleshoot an issue, but Claude Code still blocks it. Why?

A. An applicable deny remains authoritative; a local allow should not bypass the repository's protected-file policy. B. Project settings only govern Bash commands, not file access. C. The file can be opened after the developer restarts the conversation twice. D. Local settings are ignored unless CLAUDE.md explicitly imports them.

3 / 10 select ALL that apply hard

A marketing agency needs a system that takes a single product brief and generates 50 distinct variations of ad copy for different demographics. They want this to execute as quickly as possible. Which workflow patterns are appropriate for this architecture? (Select all that apply)

A. A parallelization pattern where the system maps the brief to 50 independent model calls simultaneously. B. A routing pattern where the initial brief is sent exclusively to the single model most specialized in the primary demographic. C. An evaluator-optimizer pattern where a single model drafts all 50 variations in one pass and iteratively rewrites them for diversity. D. An orchestrator-workers pattern where a lead model segments the demographics and dispatches sub-tasks to worker models concurrently. E. A prompt chaining pattern where the model writes the first ad, passes it to the next prompt to write the second ad, and repeats.

4 / 10 single answer medium

A project's `CLAUDE.md` has grown so large that it consumes 10,000 tokens per request, significantly increasing costs. What is the best way to optimize this without losing essential project context?

A. Summarize the instructions into a single paragraph. B. Use a slash command to load the instructions only when needed. C. Move the detailed documentation to an MCP resource. D. Remove all comments and extra whitespace.

5 / 10 single answer hard

A compliance assistant's request contains a 12-page system prompt, full tool manuals, the last 80 conversation turns, and every policy document in the company. Important instructions are often ignored near the end of a session. What is the best context redesign?

A. Repeat the most important policy paragraph after every tool result. B. Keep stable non-negotiable rules concise in the system prompt, expose tool details through definitions, retrieve only relevant policy passages, and compact old history into structured state. C. Remove tool schemas and describe all tools in prose inside the system prompt. D. Move all content into the latest user message so it appears closest to the model's response.

6 / 10 single answer hard

A ticket-classification prompt includes three examples — all of them straightforward billing tickets, added when billing misclassifications were the complaint. Since then, unusual tickets (multi-issue, sarcastic, non-English) get shoehorned into billing categories more than before. What went wrong?

A. Examples belong in the system prompt; placing them in the user turn caused the bias. B. Three examples is too many — more than one example causes overfitting. C. The example set is homogeneous, so it teaches 'most things look like billing' — examples should be diverse and include the edge cases you want handled. D. The examples lack XML tags, so the model is memorizing them as content.

7 / 10 select ALL that apply hard

A forestry-permit assistant uses a third-party OCR tool before submitting structured harvest declarations to an internal system. Which controls should surround the OCR output? Select all that apply.

A. Treat all OCR text as untrusted evidence rather than executable model instructions. B. Place OCR text in system so later user messages cannot override it. C. Preserve page references so reviewers can trace submitted values back to source images. D. Skip validation when the OCR provider returns a high-confidence score for every field. E. Allow the OCR provider to call the submission tool directly to reduce transformation errors. F. Validate all declaration fields against schema and forestry business rules before submission.

8 / 10 single answer hard

A telecom provisioning agent checks an activation, sees status PENDING, calls submit_activation again, and later creates duplicate service orders. The API has separate states for queued, validating, awaiting_carrier, failed, and complete, but the tool collapses them all into PENDING. What is the best fix?

A. Tell the model that PENDING usually means it should wait. B. Retry submit_activation automatically with exponential backoff for every PENDING response. C. Expose the precise provisioning state, add an idempotency key to submissions, and encode which states permit retry. D. Add more examples of successful activations to the prompt.

9 / 10 single answer medium

A platform team wants to ensure that any code modified by Claude Code passes an internal proprietary static analysis tool before the changes are committed to git. They want this to happen automatically without relying on the model remembering to run the tool. Which hook is appropriate?

A. Add a rule to the `CLAUDE.md` file demanding the model manually execute the static analysis script before calling git. B. Implement a `PostToolUse` hook targeting the `edit_file` tool to run the analyzer immediately after every file modification. C. Implement a `PreToolUse` hook targeting the `edit_file` tool to ensure the model only writes valid syntax. D. Implement a `PreToolUse` hook targeting the `git_commit` tool that executes the static analyzer and aborts if it fails.

10 / 10 select ALL that apply hard

A medical device manufacturer uses an agent to manage its quality management system (QMS). The agent can execute several operations based on incoming deviation reports. Select all operations that inherently demand a human-in-the-loop checkpoint due to blast radius or irreversibility.

A. Approving a formal root-cause analysis document and closing the associated regulatory deviation ticket. B. Drafting a preliminary summary of the deviation report based on the raw technician notes. C. Searching the historical QMS database for previous occurrences of similar manufacturing defects. D. Modifying the acceptable tolerance ranges for physical part dimensions in the master production specification. E. Initiating a mandatory factory floor recall for a specific batch of sterilized surgical instruments.

Try Claude Certified Architect – Foundations — no account needed