CCAR-P Advanced Agentic Architecture Practice Test 5 — 15 Questions

Question 1

A Claude agent maintains a long conversation with a user over weeks. The architect notices that the agent gradually forgets early constraints the user stated, even though the conversation is well within the model's context window. The team wants to fix this without re-summarizing the entire history on every turn. Which approach is most appropriate?

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

Accepted Answer

Move persistent user constraints into a structured memory store that is retrieved and injected as a system-level block on each turn.. Persistent constraints belong in a durable memory store that is retrieved and injected as a system-level block each turn, giving them stable salience without repeatedly summarizing the whole history. Larger context windows, repeated reminders, and temperature changes do not fix attention dilution over long conversations, and two of them are actively counterproductive. Structured memory is the right architectural layer for durable user preferences.

Answer

Raise the temperature slightly so the model explores more of the conversation when generating.

Answer

Increase the model's context window by switching to a larger variant so all history fits with room to spare.

Answer

Insert the early constraints again at the very end of every user message as a reminder.

Question 2

A claims-processing agent runs for hours and must survive process restarts without losing in-flight work. You are designing durable execution around Claude's stateless Messages API. Which TWO practices are required to make the agent resumable? (Choose two.)

Accepted Answer

Persist the full message history and pending tool_use ids to durable storage after each turn so the loop can be reconstructed.. Because the Messages API is stateless, resumability is entirely the orchestrator's responsibility. You must persist the conversation, including tool_use and tool_result pairings, and checkpoint in-flight tool dispatches so a restart does not replay side effects. Together these produce a replayable log that reconstructs the loop and reconciles uncertain operations, turning crash recovery into a deterministic continuation.

Answer

Rely on the model's server-side session state to remember prior turns across restarts.

Answer

Cache every model response in a CDN so identical prompts return instantly after a restart.

Answer

Increase the context window by enabling extended thinking so the agent retains more history internally.

Question 3

You are architecting a Claude-based agent for a regulated financial client that performs long-running portfolio rebalancing workflows. A compliance requirement mandates that no single trade instruction may be executed unless it is cryptographically traceable to the exact model-generated intent that produced it. The agent uses the Messages API with tool use, and several downstream services consume tool calls asynchronously. Which architectural mechanism best satisfies this requirement while preserving agent autonomy?

Accepted Answer

Persist every assistant turn containing a tool_use block, its tool_use id, and the matching tool_result, forming an immutable audit chain keyed by the tool_use id.. Binding each tool call to the platform-generated tool_use id and persisting the surrounding assistant turn creates a deterministic chain from model intent to executed instruction. Because the id appears in both the tool_use block and the tool_result, downstream services and auditors can reconcile them without trusting model-authored text. Caching, summaries, or self-hashes do not establish that binding.

Answer

Require the model to emit a SHA-256 hash of its own reasoning text inside the tool input, and verify that hash in the downstream trade service before execution.

Answer

Route all trades through a single orchestrator agent that logs a natural-language summary of each decision to an append-only ledger after the trade is confirmed.

Answer

Enable prompt caching on the system prompt so that every trade instruction inherits a stable cache key that can be presented to auditors as proof of origin.

Question 4

A Claude agent performs a multi-step deployment task. Step 3 calls a `deploy_service` tool that returns success, but the subsequent verification step fails because the service is not yet healthy. The agent currently treats any tool success as completion and ends the workflow. Which change best addresses this?

Accepted Answer

Add a post-tool-use hook that polls the service health endpoint and, if unhealthy, returns a structured error to the model so it can retry or escalate.. The agent conflates tool acceptance with desired end state, so verification must be moved into a deterministic post-tool-use hook that polls health and returns a structured error when the service is unhealthy. That lets the model retry or escalate instead of ending the workflow prematurely. Prompting, return-type changes, and token limits all leave the false-success gap intact.

Answer

Instruct the model in the system prompt to always wait 60 seconds and re-check health after every deploy.

Answer

Increase the agent's max_tokens so it has more room to notice the verification failure in its reasoning.

Answer

Change the `deploy_service` tool to return a boolean instead of a status string so the agent can parse it more easily.

Question 5

A production agent uses a ReAct loop and frequently reaches its maximum step budget while still mid-task, then returns a partial answer that looks complete. Telemetry shows the agent often re-reads the same file and re-queries the same database row across consecutive steps. Which change most directly reduces wasted steps while preserving the agent's ability to finish?

Accepted Answer

Add a step-level deduplication cache that returns the prior result when the agent issues an identical tool call with identical arguments, and inject a reminder of remaining budget into the loop.. The wasted steps come from repeating identical tool calls, so the most direct fix is to detect and short-circuit duplicates by caching results keyed on the call and its arguments. Pairing that with visibility into remaining budget helps the model spend its steps on novel actions and wrap up cleanly. Adjusting the budget up or down changes when the agent stops but not how much of its work is redundant, and collapsing the loop removes the observation-driven adaptation the agent depends on.

Answer

Raise the maximum step budget from 15 to 60 so the agent always has enough room to finish any task.

Answer

Replace the ReAct loop with a single-shot prompt that asks the model to plan every step up front and then answer without further tool calls.

Answer

Lower the maximum step budget to 8 so the agent is forced to be more decisive and avoid redundant work.

Question 6

A team wants an agent to answer questions about a large internal corpus. They notice the agent invents details when the retrieved chunks are only loosely related to the question. Which change most directly reduces fabricated answers grounded in weak evidence?

Accepted Answer

Instruct the agent in the system prompt to answer only from provided documents and to state when evidence is insufficient.. The reported failure is generation overreach under weak retrieval, so the highest-leverage fix is a grounding instruction that scopes answers to supplied documents and explicitly permits declining. This directly constrains what the model may assert and gives it a safe alternative to invention, independent of any retrieval tuning that might follow.

Answer

Raise the temperature setting so the model explores more diverse phrasings of the answer.

Answer

Increase the number of retrieved chunks returned by the vector search to fifty per query.

Answer

Switch the embedding model to one with a larger vector dimension for finer similarity scoring.

Question 7

Your team operates a Claude agent that triages inbound customer support tickets. After several weeks in production, the agent begins approving refunds above the policy ceiling and citing outdated policy text. Investigation shows the system prompt embeds a policy document that was updated three weeks ago, but the deployed prompt was never regenerated. Which architectural practice most directly prevents this class of failure?

Accepted Answer

Move policy content out of the system prompt and retrieve the current policy at runtime via a versioned tool or retrieval step, recording the version used in each decision.. Stale behavior traced to a frozen prompt is best solved by externalizing the mutable knowledge and fetching it at decision time from a versioned source. That way policy updates propagate immediately, and logging the retrieved version makes each action auditable. Compression, randomness, or peer review all leave the embedded copy as the de facto source of truth.

Answer

Add a second Claude agent that reviews the first agent's refund decisions and overrides any that appear inconsistent with its own recollection of company policy.

Answer

Increase the model's temperature slightly so the agent explores alternative interpretations of the policy text it has memorized.

Answer

Shorten the system prompt by summarizing the policy into a compact bullet list, which reduces the chance that outdated sections remain embedded.

Question 8

Your agent orchestrates a research task by spawning several subagents, each with its own Claude conversation, and merging their outputs. You observe that the final synthesis contradicts the subagents' findings. Which architectural change most reliably preserves fidelity when merging?

Accepted Answer

Pass each subagent's structured findings, including source citations and confidence, into the synthesizer as explicit quoted evidence.. Fidelity through a merge depends on provenance. When the synthesizer receives each subagent's findings as attributable, cited evidence, it can align claims to sources, surface real disagreements, and avoid fabricating a consensus. Structured handoff preserves the trail from subagent output to final answer, which is precisely what a paraphrased or mutable handoff destroys.

Answer

Raise the synthesizer subagent's temperature so it can creatively reconcile conflicting viewpoints.

Answer

Have the synthesizer re-run each subagent's task itself to verify the results before merging.

Answer

Give all subagents access to a shared mutable scratchpad so they can overwrite each other's conclusions during execution.

Question 9

A customer-support agent must sometimes escalate to a human and sometimes resolve autonomously. The compliance team requires that any action touching billing be reviewed by a human before execution, while password resets may proceed automatically. The architect wants the model to decide routing without hardcoding every rule in the prompt. Which design best satisfies the requirement?

Accepted Answer

Classify each proposed tool call by risk tier in the orchestration layer, auto-approve low-risk actions, and require human approval for billing-tier actions regardless of model output.. The compliance requirement is a hard gate, so the decision to require human approval must live in deterministic orchestration code rather than in model judgment. Tiering tool calls by risk lets low-risk actions like password resets flow automatically while billing actions are held for approval before execution. The model can still propose actions, but the authoritative approval decision is enforced outside it, which makes the control auditable and immune to prompt drift or probabilistic misrouting.

Answer

Fine-tune the model on historical escalations so it learns which actions compliance typically requires humans to approve.

Answer

Give the model a single escalate tool and instruct it in the system prompt to use judgment about when human review is required.

Answer

Log every tool call the agent makes and have the compliance team review the logs weekly to catch any unauthorized billing actions.

Question 10

You are architecting a customer-support agent for a SaaS platform. The agent must answer billing questions using a live invoice API and general policy questions using a static knowledge base. The invoice API is fast but occasionally returns stale data; the knowledge base is large and slow to search. Your design gives the agent a dedicated 'billing' sub-agent and a 'policy' sub-agent, coordinated by a router agent. After deployment, you observe the router sending nearly every query to both sub-agents in parallel, causing high latency. Which architectural change best addresses this while preserving answer quality?

Accepted Answer

Have the router emit a structured routing decision with a confidence score, and only fan out to both sub-agents when the confidence falls below a configured threshold.. The router is being overly cautious and fanning out on every query, which multiplies latency. Adding an explicit confidence signal to the routing decision lets the orchestrator fan out only when the router is genuinely unsure. High-confidence routing keeps the common case single-hop, while the fallback preserves correctness on ambiguous queries. This is the only option that changes the router's decision policy rather than replacing it with a brittle rule or an unrelated caching layer.

Answer

Merge both sub-agents into a single agent with access to both the invoice API tool and the knowledge base search tool, and let it choose tools freely.

Answer

Cache the router's routing decisions for 24 hours so repeated identical queries skip the router on subsequent requests.

Answer

Replace the router with a fixed decision tree that maps every query containing the word 'invoice' to the billing sub-agent and all other queries to the policy sub-agent.

Question 11

You are designing a Claude agent that must complete a multi-hour research task spanning dozens of tool calls, and the transcript will eventually exceed the model's context window. You want the agent to keep making correct decisions without losing critical earlier findings. Which TWO architectural strategies best preserve decision quality across the compaction boundary? (Choose two.)

Accepted Answer

Maintain an external structured scratchpad of confirmed findings, open questions, and decisions, and inject a curated summary of it into each new context window.. Durable continuity comes from moving authoritative state outside the transcript. A structured scratchpad preserves confirmed findings and open questions, while a persisted task state object lets the agent rehydrate deterministically at each boundary. Together they let compaction discard raw turns safely. Token limits and internal reasoning do not expand context or survive eviction.

Answer

Summarize and discard the oldest transcript turns once a threshold is reached, retaining only the most recent turns verbatim.

Answer

Enable extended thinking on every turn so the model can reason through the full history internally without needing external state.

Answer

Raise the max_tokens parameter on every request so the model can hold more of the transcript in a single response.

Question 12

An agent must summarize a 400-page contract that exceeds the context window. The team wants a summary that references specific clauses without losing cross-references between distant sections. Which approach best preserves cross-reference integrity?

Accepted Answer

Map each section to a structured node with its clause identifiers and cross-references, summarize nodes, then synthesize using the reference graph.. Cross-reference integrity survives summarization only if the relational structure is modeled explicitly. Converting sections into nodes with clause identifiers and reference edges lets the synthesis stage traverse those links, resolving 'see Section X' pointers instead of severing them at chunk boundaries. This preserves meaning that flat chunking or larger output budgets cannot recover.

Answer

Increase max_tokens on each summarization call so the model can produce longer per-chunk summaries.

Answer

Use a smaller model with a longer context window to process the whole contract in a single pass.

Answer

Split the contract into fixed-size chunks, summarize each independently, and concatenate the summaries in order.

Question 13

An agentic system uses a supervisor agent that delegates to specialized worker agents. During a long incident, the supervisor's own context fills with worker transcripts, and it begins losing track of which workers have completed and which are still running. Which TWO architectural changes best preserve the supervisor's ability to coordinate correctly? (Choose two.)

Accepted Answer

Have the supervisor maintain an external task ledger that records each delegated task, its assigned worker, and its status, and read from that ledger instead of retaining full worker transcripts.. The supervisor's difficulty comes from mixing coordination state with raw worker output. Externalizing task state into a ledger and having workers return compact structured envelopes keeps the authoritative record of what is delegated and what has finished separate from bulky transcripts. The supervisor can then coordinate from small, reliable signals and pull detail only when needed, rather than trying to hold the entire incident in its context window.

Answer

Restart the supervisor with a fresh context whenever its window fills, and rely on workers to resend their current status on request.

Answer

Allow workers to message each other directly to resolve dependencies, removing the supervisor from the coordination path entirely.

Answer

Increase the supervisor's context window to the largest available model so it can retain all worker transcripts for the duration of the incident.

Question 14

You operate a long-running research agent that maintains a scratchpad of findings across many turns. You notice that as the scratchpad grows, the agent begins ignoring recent tool results and repeating earlier conclusions. You cannot increase the context window. Which intervention most directly addresses the root cause of the recency failure?

Accepted Answer

Restructure the scratchpad into a fixed-size rolling summary that is regenerated each turn, with the most recent tool results kept verbatim in a dedicated, clearly delimited section.. The agent is not forgetting recent results so much as losing them in an ever-growing block of text where early material dominates attention. Bounding the scratchpad with a regenerated rolling summary keeps total size stable, and reserving a clearly delimited section for the newest tool results puts current evidence in a position the model reliably attends to. This fixes the structural cause rather than the sampling or storage symptoms.

Answer

Move the scratchpad out of the prompt entirely and store it in an external vector database, retrieving relevant entries only when the agent explicitly asks.

Answer

Append every tool result verbatim to the scratchpad so no information is lost, and instruct the agent to re-read the entire scratchpad before each decision.

Answer

Lower the model's temperature to zero so the agent becomes more deterministic and less likely to drift from the current evidence.

Question 15

A Claude agent orchestrates three specialized subagents: one for data extraction, one for validation, and one for report generation. In production, the validation subagent sometimes receives malformed input from the extraction subagent and silently produces a passing result. You need the orchestrator to detect and contain these failures without halting the entire pipeline. Which design change is most appropriate?

Accepted Answer

Have each subagent return a structured result envelope containing a status, a schema-validated payload, and an error field, and make the orchestrator validate the envelope before routing onward.. Reliable multi-agent pipelines depend on explicit contracts at every handoff. A structured envelope with status, validated payload, and error field lets the orchestrator verify upstream output before trusting it downstream. That check enables targeted containment, such as retrying or quarantining one branch, rather than trusting a failing agent's self-assessment or retrying blindly.

Answer

Increase the orchestration layer's retry count so that any subagent failure is retried several times before the pipeline reports an error to the caller.

Answer

Merge the extraction and validation subagents into a single agent so that malformed intermediate data never crosses a component boundary.

Answer

Instruct the validation subagent to always return a confidence score between zero and one, and have the orchestrator discard results below a fixed threshold.