Question 590
An architecture review of an engineering knowledge assistant finds two missing guarantees: tool-call logging does not modify a successful tool result; a delegated reviewer has no production write capability. What two-part change closes both gaps?
- rewrite every tool result as a free-form paraphrase; and give the reviewer every administrative tool and ask for restraint
- rewrite every tool result as a free-form paraphrase; and provide only the read-oriented tools required by that reviewer
- use an observation-only hook for telemetry when rewriting is unnecessary; and give the reviewer every administrative tool and ask for restraint
- use an observation-only hook for telemetry when rewriting is unnecessary; and provide only the read-oriented tools required by that reviewer
Explanation
Correct Answer
D:In an engineering knowledge assistant, separating instrumentation from mutation reduces the risk of corrupting evidence; alongside this, agent-scoped tool exposure prevents a write path in a read-only role, satisfying the distinct requirements that tool-call logging does not modify a successful tool result and that a delegated reviewer has no production write capability.
Incorrect Answers
A:In an engineering knowledge assistant, summaries can discard fields that subsequent steps require and prompt restraint is weaker than withholding execution capability; the design would fail both conditions: tool-call logging does not modify a successful tool result and a delegated reviewer has no production write capability.
B:In an engineering knowledge assistant, agent-scoped tool exposure prevents a write path in a read-only role, but summaries can discard fields that subsequent steps require; therefore the separate requirement that tool-call logging does not modify a successful tool result would remain unsatisfied.
C:In an engineering knowledge assistant, separating instrumentation from mutation reduces the risk of corrupting evidence, but prompt restraint is weaker than withholding execution capability; therefore the separate requirement that a delegated reviewer has no production write capability would remain unsatisfied.
Question 591
For a cross-service developer portal, acceptance depends on demonstrating that agents exchange outputs without losing important structure. Which approach is the strongest fit?
- use one focused tool-assisted agent for the bounded task
- define artifact schemas and handoff contracts between roles
- add a synthesis stage with explicit conflict-resolution criteria
- pass raw chat logs between every stage
Explanation
Correct Answer
B:For a cross-service developer portal, bounded interfaces make missing fields and incompatible assumptions visible; this directly establishes that agents exchange outputs without losing important structure instead of relying on an unrelated workflow control.
Incorrect Answers
A:Although orchestration overhead is not justified when there is no useful specialization or concurrency, that only supports the distinct goal that a straightforward lookup avoids unnecessary agent coordination, not the requirement that agents exchange outputs without losing important structure in a cross-service developer portal.
C:Although merging divergent work requires an authoritative step rather than choosing by arrival time, that only supports the distinct goal that two teams' analyses are reconciled before publication, not the requirement that agents exchange outputs without losing important structure in a cross-service developer portal.
D:In a cross-service developer portal, unstructured context is noisy and can obscure the decision fields; as a result, the stated requirement that agents exchange outputs without losing important structure would still not be met.
Question 592
An audit of a static analysis automation identifies two unmet requirements: an exploratory branch can be compared with a baseline later and conflicting model assessments of a disputed claim are reconciled against evidence. Select TWO actions that remedy those particular deficiencies.
- overwrite the baseline notes with the trial branch's results
- select whichever answer appears in the most generated responses
- preserve both branch identifiers and outcome summaries
- resolve the dispute against cited source evidence rather than majority vote
Explanation
Correct Answers
C:For a static analysis automation, tracking lineage makes result comparison possible without confusing which session produced an artifact, which directly establishes that an exploratory branch can be compared with a baseline later; the separate requirement that conflicting model assessments of a disputed claim are reconciled against evidence is handled by a different action.
D:For a static analysis automation, independent verification distinguishes a well-grounded minority view from shared guessing, which directly establishes that conflicting model assessments of a disputed claim are reconciled against evidence; the separate requirement that an exploratory branch can be compared with a baseline later is handled by a different action.
Incorrect Answers
A:For a static analysis automation, losing provenance prevents trustworthy evaluation, so this option cannot establish that an exploratory branch can be compared with a baseline later; correctly addressing conflicting model assessments of a disputed claim are reconciled against evidence elsewhere does not repair that deficiency.
B:For a static analysis automation, agreement among models is not a substitute for factual grounding, so this option cannot establish that conflicting model assessments of a disputed claim are reconciled against evidence; correctly addressing an exploratory branch can be compared with a baseline later elsewhere does not repair that deficiency.
Question 593
Before expanding a pre-merge code inspection workflow, the owner must demonstrate that an automation expects a defined JSON result schema, while also demonstrating that a patch that fails regression tests is diagnosed and corrected. Which design provides both controls?
- use the CLI structured-output schema option for the response; and rewrite large unrelated modules before re-running tests
- ask for valid JSON in prose but accept any text result; and inspect failure evidence and revise the smallest relevant change
- ask for valid JSON in prose but accept any text result; and rewrite large unrelated modules before re-running tests
- use the CLI structured-output schema option for the response; and inspect failure evidence and revise the smallest relevant change
Explanation
Correct Answer
D:In a pre-merge code inspection workflow, schema-constrained output is more robust for downstream integrations than prompt-only formatting; alongside this, test-driven feedback pinpoints a specific unmet behavior for the next iteration, satisfying the distinct requirements that an automation expects a defined JSON result schema and that a patch that fails regression tests is diagnosed and corrected.
Incorrect Answers
A:In a pre-merge code inspection workflow, schema-constrained output is more robust for downstream integrations than prompt-only formatting, but broad edits create new regression surfaces while the cause remains unclear; therefore the separate requirement that a patch that fails regression tests is diagnosed and corrected would remain unsatisfied.
B:In a pre-merge code inspection workflow, test-driven feedback pinpoints a specific unmet behavior for the next iteration, but a natural-language instruction does not guarantee parseable or complete fields; therefore the separate requirement that an automation expects a defined JSON result schema would remain unsatisfied.
C:In a pre-merge code inspection workflow, a natural-language instruction does not guarantee parseable or complete fields and broad edits create new regression surfaces while the cause remains unclear; the design would fail both conditions: an automation expects a defined JSON result schema and a patch that fails regression tests is diagnosed and corrected.
Question 594
While preparing a symbol-search assistant for deployment, engineers must ensure that the assistant locates files containing a particular symbol or string; they must also ensure that the model can use a tool result in its next decision. Which combination best satisfies those requirements?
- scan only filenames with Glob; and put the tool output only in application logs
- use Grep to search file contents; and append the assistant tool-use response and matching user tool_result blocks to history
- scan only filenames with Glob; and append the assistant tool-use response and matching user tool_result blocks to history
- use Grep to search file contents; and put the tool output only in application logs
Explanation
Correct Answer
B:In a symbol-search assistant, text search locates occurrences efficiently before focused reading; alongside this, Claude needs the prior request and the execution output in the conversation for the next inference, satisfying the distinct requirements that the assistant locates files containing a particular symbol or string and that the model can use a tool result in its next decision.
Incorrect Answers
A:In a symbol-search assistant, path matching does not inspect symbol occurrences inside files and unseen logs cannot inform the model's next decision; the design would fail both conditions: the assistant locates files containing a particular symbol or string and the model can use a tool result in its next decision.
C:In a symbol-search assistant, Claude needs the prior request and the execution output in the conversation for the next inference, but path matching does not inspect symbol occurrences inside files; therefore the separate requirement that the assistant locates files containing a particular symbol or string would remain unsatisfied.
D:In a symbol-search assistant, text search locates occurrences efficiently before focused reading, but unseen logs cannot inform the model's next decision; therefore the separate requirement that the model can use a tool result in its next decision would remain unsatisfied.
Question 595
Before expanding a service refactoring initiative, the owner must demonstrate that a risky refactor is evaluated before any files change, while also demonstrating that conflicting guidance from different files is detected before rollout. Which design provides both controls?
- use plan mode to inspect dependencies and propose steps before execution; and audit overlapping CLAUDE.md files and resolve instruction contradictions
- use plan mode to inspect dependencies and propose steps before execution; and assume the model will infer which contradictory rule to follow
- implement only use plan mode to inspect dependencies and propose steps before execution; make no corresponding change for the second requirement
- let the agent immediately edit every matching file; and assume the model will infer which contradictory rule to follow
- implement only audit overlapping CLAUDE.md files and resolve instruction contradictions; make no corresponding change for the first requirement
- let the agent immediately edit every matching file; and audit overlapping CLAUDE.md files and resolve instruction contradictions
Explanation
Correct Answer
A:In a service refactoring initiative, planning separates investigation and approval from code mutation; alongside this, clear ownership and scope prevent two directions from competing in the same coding task, satisfying the distinct requirements that a risky refactor is evaluated before any files change and that conflicting guidance from different files is detected before rollout.
Incorrect Answers
B:In a service refactoring initiative, planning separates investigation and approval from code mutation, but unresolved conflicts create inconsistent output across directories; therefore the separate requirement that conflicting guidance from different files is detected before rollout would remain unsatisfied.
C:In a service refactoring initiative, planning separates investigation and approval from code mutation, but the independent mechanism for the other criterion is absent: clear ownership and scope prevent two directions from competing in the same coding task; as written, the design does not ensure that conflicting guidance from different files is detected before rollout.
D:In a service refactoring initiative, early writes can spread an unapproved assumption through the repository and unresolved conflicts create inconsistent output across directories; the design would fail both conditions: a risky refactor is evaluated before any files change and conflicting guidance from different files is detected before rollout.
E:In a service refactoring initiative, clear ownership and scope prevent two directions from competing in the same coding task, but the independent mechanism for the other criterion is absent: planning separates investigation and approval from code mutation; as written, the design does not ensure that a risky refactor is evaluated before any files change.
F:In a service refactoring initiative, clear ownership and scope prevent two directions from competing in the same coding task, but early writes can spread an unapproved assumption through the repository; therefore the separate requirement that a risky refactor is evaluated before any files change would remain unsatisfied.
Question 596
An incident postmortem for a warranty claims assistant requires controls so that out-of-policy exceptions receive an authorized decision and that near-overlapping retrieval tools are selected by actual boundary. Which paired remedy covers both failure modes?
- route the exception to a human approval step before execution; and rely exclusively on short tool names for routing
- let the agent approve an exception whenever its wording sounds confident; and document examples and conditions where each tool must not be used
- let the agent approve an exception whenever its wording sounds confident; and rely exclusively on short tool names for routing
- route the exception to a human approval step before execution; and document examples and conditions where each tool must not be used
Explanation
Correct Answer
D:In a warranty claims assistant, approval is a workflow boundary separate from language-model confidence; alongside this, contrastive guidance helps tool choice when resource names look similar, satisfying the distinct requirements that out-of-policy exceptions receive an authorized decision and that near-overlapping retrieval tools are selected by actual boundary.
Incorrect Answers
A:In a warranty claims assistant, approval is a workflow boundary separate from language-model confidence, but names alone may not capture the distinction in scope or authorization; therefore the separate requirement that near-overlapping retrieval tools are selected by actual boundary would remain unsatisfied.
B:In a warranty claims assistant, contrastive guidance helps tool choice when resource names look similar, but model confidence is not delegated business authority; therefore the separate requirement that out-of-policy exceptions receive an authorized decision would remain unsatisfied.
C:In a warranty claims assistant, model confidence is not delegated business authority and names alone may not capture the distinction in scope or authorization; the design would fail both conditions: out-of-policy exceptions receive an authorized decision and near-overlapping retrieval tools are selected by actual boundary.
Question 597
For a merchant support assistant, a reviewer identifies two independent obligations: the retry process does not create duplicate records and a long workflow preserves confirmed facts separately from speculation. Which proposal should be approved?
- insert a fresh record on every model retry; and mark verified, user-reported, and unknown state explicitly
- insert a fresh record on every model retry; and combine guesses and verified attributes in one unqualified summary
- assign stable source identifiers and deduplicate writes; and mark verified, user-reported, and unknown state explicitly
- assign stable source identifiers and deduplicate writes; and combine guesses and verified attributes in one unqualified summary
Explanation
Correct Answer
C:In a merchant support assistant, idempotent processing separates repeated model attempts from committed business mutations; alongside this, epistemic labels stop a tentative suggestion becoming a confirmed action parameter, satisfying the distinct requirements that the retry process does not create duplicate records and that a long workflow preserves confirmed facts separately from speculation.
Incorrect Answers
A:In a merchant support assistant, epistemic labels stop a tentative suggestion becoming a confirmed action parameter, but repeated extraction attempts could create duplicate downstream state; therefore the separate requirement that the retry process does not create duplicate records would remain unsatisfied.
B:In a merchant support assistant, repeated extraction attempts could create duplicate downstream state and downstream operations may rely on an assumption as though it were checked; the design would fail both conditions: the retry process does not create duplicate records and a long workflow preserves confirmed facts separately from speculation.
D:In a merchant support assistant, idempotent processing separates repeated model attempts from committed business mutations, but downstream operations may rely on an assumption as though it were checked; therefore the separate requirement that a long workflow preserves confirmed facts separately from speculation would remain unsatisfied.
Question 598
In a procurement document processing system, the design must ensure that the assistant can use external service capabilities via MCP and that a destructive tool receives human authorization at execution time. Choose TWO independent changes that directly meet these two requirements.
- describe the external API in a prompt without connecting any tool
- configure a compatible MCP server and expose its declared tools to the client
- interrupt or deny the tool until an authorization callback approves it
- collect a general approval at the start of the conversation
Explanation
Correct Answers
B:For a procurement document processing system, MCP bridges a client to provider-defined capabilities rather than inventing a Claude built-in, which directly establishes that the assistant can use external service capabilities via MCP; the separate requirement that a destructive tool receives human authorization at execution time is handled by a different action.
C:For a procurement document processing system, execution gates make the approval binding on the selected action and inputs, which directly establishes that a destructive tool receives human authorization at execution time; the separate requirement that the assistant can use external service capabilities via MCP is handled by a different action.
Incorrect Answers
A:For a procurement document processing system, instructions cannot create an executable integration endpoint, so this option cannot establish that the assistant can use external service capabilities via MCP; correctly addressing a destructive tool receives human authorization at execution time elsewhere does not repair that deficiency.
D:For a procurement document processing system, blanket consent does not verify the later high-risk operation, so this option cannot establish that a destructive tool receives human authorization at execution time; correctly addressing the assistant can use external service capabilities via MCP elsewhere does not repair that deficiency.
Question 599
Before expanding a mobile client bug-fix task, the owner must demonstrate that a developer returns to an interrupted session with prior reasoning, while also demonstrating that sensitive tool calls are denied before they reach an executor. Which design provides both controls?
- resume the stored session by its identifier; and write a stronger instruction in the system prompt
- resume the stored session by its identifier; and register a PreToolUse hook that checks the operation against policy
- start an unrelated new session and assume it knows past actions; and write a stronger instruction in the system prompt
- start an unrelated new session and assume it knows past actions; and register a PreToolUse hook that checks the operation against policy
Explanation
Correct Answer
B:In a mobile client bug-fix task, session resumption retains the recorded conversation and avoids rebuilding the entire task from scratch; alongside this, pre-execution interception can reject a disallowed call rather than relying on post-hoc advice, satisfying the distinct requirements that a developer returns to an interrupted session with prior reasoning and that sensitive tool calls are denied before they reach an executor.
Incorrect Answers
A:In a mobile client bug-fix task, session resumption retains the recorded conversation and avoids rebuilding the entire task from scratch, but an instruction alone does not provide deterministic execution-time authorization; therefore the separate requirement that sensitive tool calls are denied before they reach an executor would remain unsatisfied.
C:In a mobile client bug-fix task, a fresh context lacks the prior interaction unless the facts are deliberately supplied and an instruction alone does not provide deterministic execution-time authorization; the design would fail both conditions: a developer returns to an interrupted session with prior reasoning and sensitive tool calls are denied before they reach an executor.
D:In a mobile client bug-fix task, pre-execution interception can reject a disallowed call rather than relying on post-hoc advice, but a fresh context lacks the prior interaction unless the facts are deliberately supplied; therefore the separate requirement that a developer returns to an interrupted session with prior reasoning would remain unsatisfied.
Question 600
The release checklist for a billing-dispute support agent explicitly checks that important workflow state survives context compression or process restarts; it also checks that a broad tool catalog does not overwhelm a small task. What coordinated solution meets the checklist?
- persist authoritative task facts and artifact IDs outside ephemeral conversation text; and supply only context-relevant tools for the current agent
- depend on the model remembering a prior value forever; and load all available MCP tools into every specialist turn
- depend on the model remembering a prior value forever; and supply only context-relevant tools for the current agent
- persist authoritative task facts and artifact IDs outside ephemeral conversation text; and load all available MCP tools into every specialist turn
Explanation
Correct Answer
A:In a billing-dispute support agent, external durable state can be reloaded and checked when the session window changes; alongside this, reducing irrelevant tool definitions improves selection and saves context, satisfying the distinct requirements that important workflow state survives context compression or process restarts and that a broad tool catalog does not overwhelm a small task.
Incorrect Answers
B:In a billing-dispute support agent, context limits and summarization can lose exact identifiers and tool bloat makes the available actions harder to discriminate; the design would fail both conditions: important workflow state survives context compression or process restarts and a broad tool catalog does not overwhelm a small task.
C:In a billing-dispute support agent, reducing irrelevant tool definitions improves selection and saves context, but context limits and summarization can lose exact identifiers; therefore the separate requirement that important workflow state survives context compression or process restarts would remain unsatisfied.
D:In a billing-dispute support agent, external durable state can be reloaded and checked when the session window changes, but tool bloat makes the available actions harder to discriminate; therefore the separate requirement that a broad tool catalog does not overwhelm a small task would remain unsatisfied.