CCAR-F Claude Certified Architect – Foundations Questions and Answers

Option D builds an evidence-based architectural map without indiscriminately loading hundreds of files. Authentication entry points may include route handlers, middleware registration, token-validation functions, session constructors, or identity-provider callbacks. After Grep identifies those anchors, targeted Read operations can establish their responsibilities, while imports, call sites, and configuration references reveal the downstream authorization flow. Anthropic’s large-codebase guidance recommends scoping Claude to the portion of a repository touched by the task because unrelated instructions and file reads consume tokens and degrade performance. Its common workflows guidance similarly recommends beginning broadly and narrowing into specific components. Option A launches agents before the relevant service boundaries are known and may produce overlapping, inconsistent investigations. Option B is an uncontrolled lexical sweep that will include tests, documentation, unrelated tokens, and incidental terminology. Option C appropriately checks project guidance but incorrectly delegates technical file selection to a new engineer who may not know the architecture. Incremental dependency tracing maintains context efficiency while producing a verifiable end-to-end map grounded in actual code paths.

================

Option B identifies escalation through direct operational criteria rather than indirect proxies. A customer’s explicit request for a person is unambiguous. A required policy exception defines an authorization boundary. An inability to make meaningful progress identifies a case where continued autonomous execution is no longer productive.

Anthropic describes agents as systems that act, observe results, adjust, and continue until the task is completed or human input is required. Its trustworthy-agent guidance also emphasizes that agents must recognize when to pause rather than pushing through uncertainty or decisions that only a human can settle. ( https://www.anthropic.com/research/trustworthy-agents )

Option A is too rigid for high-ambiguity support cases. Maintaining exhaustive mappings for every issue, product, and customer segment recreates the brittleness that agentic reasoning is intended to avoid. Option C treats repeated tool use as a substitute for semantic progress; one definitive authorization failure may require immediate escalation, while several legitimate diagnostic calls may not. Option D confuses emotional tone with operational necessity. A calm customer may require a policy exception, while a frustrated customer may still have a straightforward automated resolution.

The escalation criteria should be encoded in the system instructions and supported by evaluations covering explicit requests, authorization limits, unresolved ambiguity, repeated non-progress, and successful self-service cases.

Official references/topics: Human-control checkpoints, escalation criteria, authorization boundaries, progress-aware agents.

In the standard Claude tool-use loop, the application executes lookup_order and sends its output back as a tool_result . That result becomes part of the conversation state available to Claude. Claude then evaluates the purchase date, refund policy, customer request, authorization constraints, and available tools before selecting the next action.

Anthropic describes client-tool orchestration as a repeated loop: Claude emits a tool_use request, the application executes it, returns a tool_result , and Claude continues reasoning from the updated conversation. Claude, rather than the tool implementation, selects when and how to invoke the next available tool unless the application has explicitly implemented a fixed workflow. ( https://platform.claude.com/docs/en/agents-and-tools/tool-use/how-tool-use-works )

Options B and C describe possible custom orchestration architectures, but neither is stated in the scenario. Option D is inconsistent with adaptive agent behavior because later actions depend on information that did not exist before lookup_order completed. A rigid sequence would not respond appropriately to different purchase dates, eligibility states, or order conditions.

The tool result should return high-signal fields such as purchase date, return-window status, refund eligibility, existing refund status, and stable order identifiers so Claude can make the subsequent decision accurately.

Official references/topics: Tool-result continuation, adaptive agent loops, model-directed tool selection, sequential dependency handling.

The task is open-ended because neither the critical modules nor the required testing sequence is known in advance. The agent should first use lightweight discovery tools to map the repository, locate existing tests, identify central modules, and determine which components have high fan-in, business significance, complex branching, or extensive external dependencies. It can then produce an initial risk-based testing plan and refine it as new dependency information appears.

Anthropic distinguishes predefined workflows from agents that dynamically control their processes and tool usage. Agents are appropriate when the required steps cannot be reliably hardcoded and must adapt to environmental evidence. During execution, they should obtain ground truth through tool results and use that feedback to determine subsequent actions. ( https://www.anthropic.com/research/building-effective-agents ) Anthropic also identifies orchestrator-worker designs as suitable for complex coding and search tasks where the necessary subtasks depend on what the investigation reveals. ( https://www.anthropic.com/research/building-effective-agents )

Option A assigns effort using directory boundaries rather than risk. Option C exhausts context before delivering value. Option D uses alphabetical order, which has no relationship to impact or coverage priority.

Option B establishes an evidence-driven decomposition: discover, prioritize, test high-impact paths, measure results, and revise the plan as dependencies and uncovered risks emerge.

Official references/topics: Dynamic Task Decomposition; Adaptive Agent Loops; Orchestrator-Workers; Risk-Based Test Planning.

Project skills are stored at .claude/skills/ < skill-name > /SKILL.md . The directory name becomes the invocable command name, so .claude/skills/migrate-component/SKILL.md creates /migrate-component . Anthropic distinguishes project skills from personal skills: project skills apply only to the repository, while skills under ~/.claude/skills/ apply to one user across all projects. ( https://code.claude.com/docs/en/skills )

Committing the project skill to version control ensures that every developer receives the same workflow and that changes are reviewed, versioned, and synchronized with the codebase. Anthropic’s .claude directory guidance states that project configuration should be committed when it is intended to be shared with the team, whereas files under ~/.claude are personal. ( https://code.claude.com/docs/en/claude-directory )

Option A requires manual duplication across developer machines and allows versions to diverge. Option B loads the migration procedure into every conversation even when no migration is occurring; Anthropic recommends skills for reusable checklists and multi-step procedures because their bodies load only when used. Option C describes no valid mechanism for defining a skill’s content in settings.json .

Option D provides the correct scope, command name, on-demand loading behavior, and team distribution model.

Official references/topics: Project skills, SKILL.md structure, command invocation, version-controlled team configuration.

Claude selects tools principally from their names, descriptions, parameter schemas, and the task context. The current description—“Analyzes dependency graph”—does not explain why the MCP tool is superior to a familiar text search. It omits the tool’s scope, the circumstances in which it should be selected, and the structured information it returns.

Anthropic identifies detailed descriptions as the most important factor in tool-use performance. A strong description should state what the tool does, when it should and should not be used, what its parameters mean, and any limitations. Anthropic recommends several sentences for complex tools rather than a short generic label. ( https://docs.anthropic.com/en/docs/agents-and-tools/tool-use/implement-tool-use ) The MCP connector guidance likewise states that Claude selects among available tools using their names and descriptions and that clear, specific descriptions improve selection accuracy. ( https://docs.anthropic.com/en/docs/agents-and-tools/mcp-connector )

Option B makes the functional distinction explicit: Grep locates textual import statements, whereas analyze_dependencies constructs a semantic graph containing direct and transitive dependencies, cycles, and potentially unresolved references. Option A is a brittle global override. Option C removes a generally useful tool. Option D increases the number of tools and selection ambiguity, contrary to Anthropic’s recommendation to consolidate related operations where practical.

Official references/topics: MCP Tool Discovery; Tool Descriptions; Tool Selection Accuracy; Tool-Surface Design.

Resuming preserves the prior architectural analysis, dependency mapping, and conversation history, avoiding the cost of repeating an hour of valid work. Anthropic states that resuming by session ID restores the agent’s full prior context, including files read, analysis performed, and decisions made. ( https://code.claude.com/docs/en/agent-sdk/sessions )

However, the session represents historical knowledge rather than a repository snapshot. Because three files changed externally, the agent must be told which files are affected and directed to re-read them before relying on earlier conclusions. Anthropic clarifies that sessions persist the conversation, not the filesystem. ( https://code.claude.com/docs/en/agent-sdk/sessions ) Targeted re-analysis updates the stale portion of the agent’s model while retaining the nine unchanged files’ established context.

Option A discards valuable work even though most of the analyzed code remains unchanged. Option C is accurate but inefficient because it reprocesses all twelve files regardless of whether they changed. Option D is unsafe: the agent may continue reasoning from obsolete function signatures, dependencies, or control flows.

The continuation prompt should identify the changed files, summarize the merge’s purpose where known, and request a focused comparison against the earlier understanding. Any architectural conclusions affected by those changes should then be revised explicitly.

Official references/topics: Session Resume; Filesystem Versus Conversation State; Targeted Re-Analysis; Repository Change Awareness.

Option B combines lossless persistence with selective restoration. Each agent should checkpoint structured records containing completed document IDs, extracted findings, source metadata, unresolved work, validation status, and schema version. A coordinator-level manifest records where those artifacts are stored and which pipeline stages completed. On resumption, the coordinator can skip the 12 completed documents and inject only the artifacts relevant to each restarted agent. Anthropic’s effective context-engineering guidance recommends structured note-taking outside the context window so agents can preserve project state across resets while loading only the information needed later. A vector store, option A, is useful for relevance retrieval but can omit exact details through similarity-based selection and is unsuitable as the sole recovery record. Option C fragments ownership of recovery state and makes cross-agent consistency difficult to verify. Option D preserves an extensive trace, but replaying a complete conversation wastes context and mixes obsolete instructions, intermediate reasoning, and redundant outputs. Structured checkpoints plus a manifest provide deterministic recovery, auditability, exact provenance, and efficient prompt reconstruction. Checkpoints should be written atomically and validated before the corresponding stage is marked complete.

================

Option C allows orchestration effort to scale according to the actual evidence and processing requirements of each request. A factual lookup might require one search operation and a direct coordinator response, while comparative research may justify parallel search, document analysis, synthesis, and formal report generation.

Anthropic’s orchestrator-workers guidance describes a central model dynamically decomposing a task, choosing workers, and combining their results. This pattern is particularly appropriate when required subtasks cannot be predicted reliably in advance. Anthropic’s multi-agent research architecture similarly uses a lead agent to analyze the query and determine an appropriate research strategy.

Option A creates an inflexible two-route system and still sends every non-factual query through the most expensive workflow. Option B requires labeled training data, an externally maintained classifier, and a stable definition of the optimal agent combination. Option D is less expensive to implement but becomes brittle as query types evolve beyond predetermined categories. Dynamic coordination uses the model’s existing reasoning ability to assess source requirements, analytical depth, expected output, and uncertainty before invoking only the specialists that materially improve the result.

================

Option A parallelizes independent precedent analysis while preserving centralized control. The coordinator can partition the 12 precedents into balanced groups, provide each worker with identical extraction and citation requirements, monitor completion or failure, and aggregate the structured results before invoking synthesis.

Anthropic’s multi-agent research architecture uses an orchestrator-worker pattern in which a lead agent coordinates specialized subagents operating in parallel. Parallel execution is valuable when tasks are substantially independent, as each precedent can be analyzed without waiting for the preceding case. Maintaining the fan-out at the coordinator also produces a clear execution trace showing each assignment, status, and returned result.

Option B introduces nested delegation and makes tool usage, permissions, failures, and costs harder for the coordinator to observe. Option C compounds those problems through recursive spawning and risks excessive agent and token consumption. Option D may be appropriate for a large distributed processing platform, but it adds infrastructure without inherently improving the coordinator’s reasoning-level observability or defining how results are associated with the case. Coordinator-managed parallel workers provide the required latency reduction with the simplest debuggable architecture.

================

The prompt explicitly instructs Claude to suppress uncertain and lower-severity findings. Increasing reasoning depth cannot override that reporting policy: Claude may identify a real defect during analysis and still omit it from the final response. Option B separates comprehensive detection from acceptance filtering, removing the source of the false negatives while retaining metadata needed for automated decisions.

Anthropic’s current code-review prompting guidance recommends reporting every issue, including uncertain or lower-severity candidates, and attaching confidence and estimated severity so a separate verification or filtering stage can rank them. A structured schema can additionally require fields such as file path, line number, category, confidence, severity, evidence, and recommended action.

Option A may improve investigation quality but leaves the suppression instruction intact. Option C removes the problematic threshold but also discards the explicit structured classification needed by downstream processing. Option D repeats the same review policy, so the second pass can suppress the same findings as the first. Option B preserves recall and machine-readable output while allowing deterministic, adjustable downstream thresholds. This is more reliable than embedding an overly restrictive acceptance decision inside the model’s initial detection task.

================

The failure is caused by missing evidence, not insufficient reasoning over the supplied evidence. A static prompt containing only changed files cannot reliably identify callers, configuration dependencies, generated interfaces, or indirect relationships located elsewhere in the repository. Asking Claude to reason more deeply cannot recover code that it was never given.

Option A converts the review into a bounded agentic loop. Claude can search for symbol references, read relevant callers, inspect type definitions, and follow newly discovered dependencies before validating a potential defect. Anthropic describes the Claude Code agentic loop as gathering context, taking action, verifying results, and repeating based on tool feedback. A turn limit preserves predictable cost and execution time.

Option C improves coverage but imposes an arbitrary two-hop boundary and depends on the accuracy of a precomputed graph, which may omit dynamic imports, reflection, generated code, configuration references, or language-specific call relationships. Option D creates duplicated context and aggregation complexity while still limiting each reviewer to predetermined dependents. Option B changes the reasoning instructions but supplies no mechanism for checking unchanged files. Tool-enabled, just-in-time retrieval is therefore the most adaptable and reliable architecture for cross-file review.

================

The agent must preserve durable state for several concurrently open issues without keeping every conversational turn in the active context window. Option D creates a structured issue ledger containing identifiers, issue type, requested action, current status, completed steps, unresolved questions, and relevant tool outputs.

Anthropic recommends structured note-taking for long-horizon agents. In this pattern, critical state is written outside the context window and retrieved later, allowing the agent to preserve dependencies and progress across dozens of tool calls. Anthropic also describes compaction as retaining critical details while discarding redundant raw messages and historical tool output. ( https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents )

Option A is better than retaining the entire transcript, but a narrative summary can merge separate issues or obscure exact identifiers and statuses. Option B will discard the refund conversation because it occurred outside the most recent 30 turns. Option C may recover current backend facts, but it cannot reconstruct conversational commitments, prior explanations, or why a particular action remains pending.

A structured context layer should maintain one record per issue and be updated after every material action. The active prompt can include compact summaries of all open issues and expanded details only for the issue currently being discussed.

Official references/topics: Structured note-taking, external agent memory, context compaction, multi-issue state management.

Claude Code’s --bare option is specifically designed for faster scripted execution. It skips automatic discovery of CLAUDE.md files, hooks, skills, plugins, MCP servers, and auto-memory while retaining essential built-in capabilities such as Bash, file reading, and file editing. Because --bare also prevents automatic loading of the root CLAUDE.md, the required standards must be added explicitly.

Option C accomplishes both objectives. --append-system-prompt-file ./CLAUDE.md loads the project standards into the current invocation while preserving Claude Code’s default coding-agent system prompt and tool-use guidance. According to the official Claude Code CLI reference , append flags add file contents to the default system prompt, whereas replacement flags discard that default guidance.

Option A could work functionally but duplicates the standards in every command and creates configuration drift. Option B replaces the complete default prompt, removing useful coding, safety, and tool-use instructions. Option D improves prompt-cache reuse across different machines, but it does not disable hooks, plugins, skills, MCP servers, or CLAUDE.md discovery and therefore does not directly address the measured initialization delay.

================

Option C defines the reviewer’s actual objective instead of relying on an undefined notion of confidence. The prompt should identify reportable categories such as correctness defects, security vulnerabilities, data-loss risks, broken requirements, and meaningful performance regressions. It should also specify that cosmetic preferences, lint-enforced formatting, speculative refactoring, and intentional repository conventions must not be reported unless they cause concrete incorrect behavior.

Anthropic’s Claude Code best-practices guidance recommends telling a reviewer to report gaps affecting correctness or stated requirements and to treat other observations as optional. Without such criteria, a reviewer asked to “find issues” will predictably return technically defensible but operationally low-value observations. Option A adds cost and another probabilistic decision layer before correcting the primary prompt. Option B relies on self-reported confidence, which does not necessarily align with developer value or severity. Option D can suppress genuine defects that require inference or cross-file verification because “highly confident” is undefined. Explicit scope, evidence requirements, and examples of both reportable and excluded findings directly calibrate the review toward what developers will act upon. Claude Code best practices

================

Path-scoped rules directly satisfy the requirement that instructions load only when Claude works with matching files. Each infrastructure domain can have a separate Markdown file under .claude/rules/ , with YAML paths frontmatter targeting Terraform files, Kubernetes manifests, or pipeline configuration.

Anthropic documents that .claude/rules/ keeps project guidance modular and supports conditional loading based on glob patterns. Rules with a paths field apply only when Claude works with matching files, reducing irrelevant context and token consumption. Rules without path frontmatter load unconditionally. ( https://code.claude.com/docs/en/memory )

Option B improves readability but leaves all 500-plus lines in every session. Option C can provide directory-specific CLAUDE.md guidance, but path-scoped rules are more precise when conventions depend on extensions or patterns spanning multiple directories. Option D still requires manual imports and does not provide the documented conditional rule-loading mechanism.

A suitable design would use rules such as terraform.md scoped to terraform/**/*.tf , kubernetes.md scoped to kubernetes/**/*.{yaml,yml} , and pipelines.md scoped to the pipeline directory. Broad project-wide commands and repository etiquette can remain in the root CLAUDE.md.

Official references/topics: .claude/rules/ ; Path-Specific Rules; YAML Frontmatter; Context-Efficient Instruction Loading.

Option A removes the semantic ambiguity created by one generic tool performing three substantially different operations. Each replacement tool can have a distinct name, focused description, appropriate required parameters, and an output schema designed for its task. For example, extract_data_points can return typed records, while verify_claim_against_source can return a verdict, supporting excerpt, and source location.

Anthropic’s tool-design guidance recommends thoughtful tools with clear, distinct purposes and natural task boundaries. Ambiguous or overlapping capabilities make it harder for agents to select the correct operation and interpret its results consistently.

Option B improves the existing interface, but the implementation and result contract remain overloaded, increasing the possibility that the model or tool will mix behavior across modes. Option C adds another probabilistic classification stage without repairing the underlying interface. Option D may reduce some errors, but free-text interpretation remains the principal control mechanism. Purpose-specific tools encode the intended operation directly into the callable interface, permit strict schemas tailored to each result type, and make evaluation failures easier to diagnose. This produces a stronger contract between the agent and deterministic document-processing system.

================

The Messages API exposes stop_reason specifically so the application can determine why Claude stopped generating and what action is required next. A value of tool_use means Claude is requesting one or more tools and expects the application to execute them and return corresponding tool_result blocks. A value of end_turn means Claude has naturally completed the response. ( https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons )

Anthropic’s agent-loop tutorial implements this as a loop that continues executing tools and returning results while stop_reason remains tool_use . Once the value changes, the response is treated according to the new terminal or continuation condition. ( https://platform.claude.com/docs/en/agents-and-tools/tool-use/build-a-tool-using-agent )

Option A is unreliable because an assistant response can contain both text and tool-use blocks. Text alone does not prove that the workflow is complete. Option B requires the application to predict that no further tool call is needed, defeating Claude’s ability to adapt from tool results. Option D is an important safety limit but is not the primary completion signal; an arbitrary maximum may terminate a valid workflow prematurely.

The harness should check stop_reason , process every requested tool call, append results to the conversation, and maintain a separate maximum-turn safeguard against runaway loops.

Official references/topics: Messages API stop reasons, tool-use loops, tool-result continuation, bounded orchestration.

Option A separates competing objectives so each review call can devote its attention and examples to a coherent concern. Anthropic’s Building Effective AI Agents describes sectioning as a parallelization pattern in which independent aspects are handled by separate model calls and then aggregated. It specifically notes that models generally perform better on complex tasks when each consideration receives focused attention. The evaluation results demonstrate prompt interference: optimizing business-logic detection reduces API-design recall.

A security and API-design reviewer can use examples, terminology, and evidence criteria appropriate to interfaces and vulnerabilities. A business-logic reviewer can focus on state transitions, arithmetic boundaries, invariants, and domain-specific edge cases. Their structured findings can then be deduplicated and ranked before posting. Option B may improve only the weak category and can overfit to known quiz-scoring failures. Option C assumes that model capacity alone removes prompt interference without evaluation evidence. Option D adds substantial unrelated context and does not isolate the competing review objectives. The split prompts should be evaluated independently and together so consolidation does not reintroduce omissions or duplicate findings.

================

Option A uses the dedicated control surface for managed Claude Code Review. Anthropic’s Code Review documentation states that a root-level REVIEW.md is injected into every review agent as the highest-priority instruction block. It can define skip paths, suppress categories already enforced by CI, recalibrate severity, cap nit volume, and require source evidence before reporting particular findings. The documentation explicitly identifies generated code, linting, and verification requirements as appropriate uses.

Option B configures a self-hosted GitHub Actions workflow, but the scenario concerns the separate managed Code Review service running on Anthropic’s infrastructure. Instructions in that workflow do not control the managed reviewer. Option C provides useful general project context, but CLAUDE.md has lower review-specific authority: managed Code Review treats violations of it primarily as nit-level findings. REVIEW.md is the stronger and more precise mechanism for changing what the managed service reports. The file should skip src/gen/**, suppress style issues already enforced by CI, and require concrete behavioral evidence for rendering-helper warnings while continuing to report verified correctness and security defects.

================

Retry-with-error-feedback is effective when the required information is available and the defect concerns representation. Options A, C, and D are correctable formatting failures: the model can flatten an object into an array, remove thousands separators and emit an integer, or convert a datetime into the required date-only format. The validation message supplies enough information to revise the output.

Option B is fundamentally different. The complete co-author list is absent from the model’s input and exists only in an external document. No number of retries can recover evidence that was never supplied. Repeated attempts may instead increase the probability of fabrication. Anthropic’s hallucination guidance recommends restricting responses to available documents, allowing the model to acknowledge missing information, and withholding or retracting claims that cannot be grounded in the source. ( https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations )

The correct response is therefore to return an explicit missing-data state, retrieve the external document through a tool, or route the record for enrichment. Retrying without changing the information boundary cannot solve the problem.

Current Anthropic Structured Outputs can eliminate many shape-related failures by guaranteeing JSON Schema conformance, but they still cannot create unavailable source facts. ( https://platform.claude.com/docs/en/build-with-claude/structured-outputs ) This distinction—format failure versus evidence failure—is central to reliable extraction architecture.

Official references/topics: Retry Boundaries; Missing Context; Grounded Extraction; Structured Outputs; External Data Retrieval.

Option C is the strongest choice among the listed approaches because it moves structured data generation from unconstrained response text into a schema-governed interface. Prompt-only JSON instructions, regular-expression extraction, and corrective retries are probabilistic recovery mechanisms: they may reduce failures, but none guarantees that the first response satisfies the contract. Anthropic now provides two schema-constrained mechanisms: JSON Structured Outputs through output_config.format and strict tool use through a tool definition with strict: true. Strict tool use uses grammar-constrained sampling so the tool input conforms to the declared JSON Schema, including required properties and data types. Therefore, the production implementation represented by option C should define the extraction tool’s input_schema and enable strict mode. The application reads the tool_use block as the extracted record; it does not need to scrape prose or search for braces. Option A remains vulnerable to preambles or malformed output. Option B can select a JSON-looking substring but cannot establish schema validity. Option D adds latency and cost and can still fail repeatedly. The schema-constrained tool contract is consequently the most reliable architecture available in these options.

================

The failure is caused by an underspecified extraction policy rather than an inadequate JSON structure. “Extract all skills mentioned” does not define whether coordinated phrases should be split, whether inferred capabilities qualify, or what level of specificity constitutes one skill. Few-shot examples can demonstrate all three decisions using concrete inputs and expected outputs.

Anthropic identifies examples as one of the most effective ways to improve output accuracy and consistency because they communicate behavior more precisely than abstract instructions alone. Representative examples can show that “Python and SQL” becomes two entries, that an unstated skill must not be inferred from a job title, and that broad descriptions should not be decomposed into dozens of speculative micro-skills. ( https://docs.anthropic.com/en/docs/about-claude/use-case-guides/ticket-routing )

Option A adds provenance and confidence metadata but does not define the correct extraction granularity. Option C addresses explicitness and splitting, but the arbitrary 10–20 limit can truncate documents containing more than 20 legitimate skills or force padding when fewer than 10 appear. Option D is useful for canonicalization and deduplication, but post-processing cannot reliably remove plausible implied skills or determine how the original phrase should have been interpreted.

The strongest prompt should combine explicit rules with diverse few-shot examples covering normal cases, compound phrases, implied information, duplicates, and unusually long or short skill lists.

Official references/topics: Few-Shot Prompting, Output Consistency, Explicit Extraction Criteria, Representative Examples.

The --system-prompt flag replaces Claude Code’s complete default system prompt. Although this does not technically remove the available tools, it discards the default coding-agent guidance that tells Claude how to gather repository context, navigate code, use tools, and verify findings. The replacement prompt says only to analyze the supplied diff, so the observed diff-only behaviour is consistent with the configured instructions.

Option B preserves the standard Claude Code identity and tool guidance while adding the custom review criteria. The Claude Code CLI reference states that --append-system-prompt appends instructions to the default prompt, whereas --system-prompt replaces it. Anthropic specifically recommends appending when Claude should remain a coding assistant that follows additional per-invocation rules.

Option A is incorrect because print mode does not inherently disable Read, Glob, or Grep; --allowedTools restricts or pre-authorizes tools but does not restore overwritten behavioural guidance. Option C is incorrect because replacement system prompts remain compatible with tools, although the replacement must supply suitable tool instructions. Option D changes where the diff appears but does not restore Claude Code’s default code-navigation behaviour. Appending the review instructions directly addresses the configuration error.

================

The evaluation results demonstrate prompt interference: improving attention to business-logic examples reduces performance on API design. Option C isolates the competing objectives so each model call can apply focused terminology, examples, evidence requirements, and evaluation criteria. The resulting structured findings can then be merged, deduplicated, and ranked before being posted.

Anthropic’s Building Effective AI Agents describes parallelization through sectioning, where independent aspects of a task are handled by separate model calls and subsequently aggregated. This pattern is appropriate when one prompt must evaluate several distinct considerations and a single call does not handle all of them reliably.

Option A increases context volume without resolving competition between the review objectives; irrelevant repository content may further dilute attention. Option B may improve known quiz-scoring cases but can overfit the reviewer to an incomplete checklist and does not restore API-design recall. Option D assumes that additional model capability will eliminate prompt interference without supporting evaluation evidence. Separate specialized review passes directly address the measured failure. Each pass should be evaluated independently, followed by an end-to-end evaluation confirming that consolidation preserves recall and does not introduce duplicated findings.

================

Option C is the only approach that guarantees the required terminal condition independently of the model’s behavior. When the agent loop stops, the orchestration layer evaluates the recorded outcome. If neither a successful resolution nor a confirmed human escalation occurred, deterministic application code invokes escalate_to_human using the information accumulated before termination.

The max_turns limit is a host-controlled safety boundary. Once that boundary is reached, the model may no longer have an opportunity to follow a prompt instruction or make another tool call. Therefore, the required fallback cannot depend entirely on Claude deciding to escalate before its remaining turns are exhausted.

Option A escalates based on an arbitrary 80% threshold and may prematurely terminate cases that could have been resolved successfully within the remaining budget. Option B provides additional turn capacity but does not guarantee that the second agent will complete its work or escalate before its own limit is reached. Option D remains probabilistic and cannot operate after an unexpected termination has already occurred.

The orchestration fallback should be idempotent, ensuring that repeated completion checks cannot create duplicate escalation records. It should also include verified customer information, relevant order data, completed actions, failure reasons, and the unresolved request in the handoff payload.

Official references/topics: Agent-loop termination, deterministic fallback orchestration, terminal-state validation, human handoff guarantees.

The current prompt asks for testing analysis at too high a level. Claude recognizes the obvious absence of an entire test but has not been instructed to construct a path-level inventory. Option B turns the desired behavior into an explicit verification procedure: enumerate each condition, alternative branch, early return, exception handler, and failure path, then locate a test assertion that exercises its behavior.

Anthropic’s prompting best practices emphasize clear, specific instructions and explicit sequential steps when a task requires a defined analysis process. This change keeps the existing single review call while making the missing evaluation criterion unambiguous.

Option A improves proximity between implementation and tests but does not tell Claude what coverage relationship to inspect. Option C could work, but it adds orchestration, latency, cost, and another handoff before testing whether a direct instruction solves the observed failure. Option D may improve recognition of examples resembling the demonstration, but a few cases cannot enumerate every branch structure. Explicit path enumeration generalizes across unfamiliar code and creates auditable output: each reported gap can name the uncovered condition, expected behavior, and missing assertion.

================

Option A reduces the maximum output required from any single response while preserving the structured schema and complete severity range. The pipeline can partition files into coherent groups, execute bounded reviews, validate each returned array, and merge and deduplicate findings using stable fields such as file path, line number, category, and description.

Anthropic’s stop-reason documentation confirms that max_tokens means generation reached the configured output limit and the response must be treated as incomplete. Structured output constraints can guarantee schema-valid generation when completion succeeds, but they cannot create unlimited output capacity. A large findings array can still exceed the available token budget.

Option B may postpone the failure but provides no durable guarantee for still-larger pull requests, and aggressively shortening descriptions may eliminate necessary evidence. Option C abandons machine-validated structure without reducing the amount of generated content. Option D deliberately suppresses medium- or low-severity findings and repeats an oversized request rather than addressing its scope. Partitioning establishes predictable output bounds, supports targeted retries, retains every required finding category, and prevents a single truncated response from invalidating the complete review.

================

Option B creates a targeted correction loop using information the first extraction did not have: the validator’s precise failure report. Anthropic’s prompting guidance describes prompt chaining as appropriate when an application must inspect intermediate output or enforce a specific pipeline, with self-correction following the pattern generate, review against criteria, and refine. Here, deterministic validators identify arithmetic mismatches and invalid vendor-ID formats. Returning the source document, original extraction, and explicit errors lets Claude revise only the defective fields while preserving valid data. The corrected record must then be validated again before acceptance, with a bounded retry count and human-review fallback. Option A is unsafe because automatically recalculating a total may overwrite the amount actually printed on the invoice; a mismatch can originate in an OCR or line-item extraction error. Option C repeats the same extraction without explaining what failed, wasting calls and relying on chance. Option D may prevent some format violations, but JSON Schema cannot express every cross-field arithmetic relationship or external vendor-registry rule. A validator-guided second pass therefore reduces manual review without silently changing source facts, while keeping deterministic code—not the model—as the final acceptance authority.

================

The root cause is an incomplete tool contract. Claude sees that process_refund requires an order_id , but the tool description does not explain the parameter’s trusted source or the prerequisite lookup sequence. Option A makes the dependency explicit and prohibits fabricated values.

Anthropic states that tool descriptions should explain what a tool does, when it should be used, how it behaves, and any important limitations. Its prompting guidance also states that dependent tool calls must be executed sequentially and that Claude must never use placeholders or guess missing tool parameters. ( https://platform.claude.com/docs/en/agents-and-tools/tool-use/define-tools )

Option B merely forces some tool use; it does not force the correct tool or prevent a fabricated identifier. Option C is an essential defense-in-depth control, but it detects the invalid ID after the model has already made the faulty call. It does not correct the selection logic that caused the fabrication. Option D works only when the customer actually provides an identifier and introduces an unnecessary preprocessing dependency.

The process_refund schema should describe order_id as a verified identifier returned by lookup_order , and the tool description should state that the tool is unavailable until the corresponding order has been retrieved and eligibility established.

Official references/topics: Tool descriptions, prerequisite tool calls, parameter provenance, sequential tool orchestration.

The reference to generic rendering patterns indicates that the current context has become saturated or that important earlier details are no longer receiving sufficient attention. Anthropic documents that performance can degrade as a Claude Code context window fills with conversation history, file contents, and command output; symptoms include forgetting earlier instructions and making less precise responses. ( https://code.claude.com/docs/en/best-practices )

Option C preserves the high-value rendering discoveries while isolating the physics investigation. The summary should identify the relevant rendering classes, synchronization boundaries, debug-overlay interfaces, and unresolved integration questions. A physics-focused subagent can then inspect collision and simulation code in its own context without adding dozens of additional files to the already crowded main conversation.

This explicit summary matters because an ordinary named subagent starts with a fresh, isolated context and does not automatically receive the parent’s full conversation history or previously read files. Claude constructs a delegation message containing the task context, so the essential rendering findings must be supplied deliberately. ( https://docs.anthropic.com/en/docs/claude-code/sub-agents )

Option A omits the integration context needed to guide the physics search. Option B destroys useful prior analysis. Option D keeps expanding an already degraded context and treats the symptom rather than controlling context growth.

Official references/topics: Context-Window Management; Subagent Isolation; Delegation Summaries; Codebase Exploration.

The problem is an undefined relevance threshold. Claude is finding technically valid observations, but the prompt does not clearly distinguish actionable defects from acceptable conventions and low-value style preferences. Option A establishes explicit positive and negative reporting criteria, allowing the model to apply the team’s actual definition of a useful finding during generation.

Anthropic’s prompt-engineering guidance emphasizes clear, specific instructions and well-defined success criteria. The managed Code Review documentation similarly recommends defining skipped categories, generated paths, severity rules, and evidence requirements. Reporting “bugs affecting correctness or security” while excluding “formatting handled by CI and documented local conventions” is substantially more precise than asking the model to be generally conservative.

Option B adds cost, latency, another probabilistic decision, and a new evaluation surface before improving the original prompt. Option C confuses confidence with importance: Claude may be highly confident about a trivial style observation. Option D is dangerously broad because it can suppress legitimate but uncertain bugs, reducing recall. Explicit relevance criteria address the demonstrated failure directly while preserving the model’s ability to investigate and report genuine correctness or security problems.

================

Formatting is deterministic and should be enforced by deterministic tooling rather than stronger natural-language instructions. A PostToolUse hook executes after a successful tool operation. Matching Edit|Write restricts the hook to file modifications, allowing the edited file path to be passed directly to Prettier.

Anthropic provides this exact pattern in its official hooks guidance: configure a PostToolUse event with an Edit|Write matcher and run Prettier against the modified file. This ensures formatting occurs automatically after every relevant edit instead of depending on Claude remembering to execute a formatter. ( https://code.claude.com/docs/en/hooks-guide )

Option A still relies on instruction-following and unnecessarily loads procedural material. Option B asks another probabilistic model evaluation to determine whether formatting is correct, even though Prettier can enforce the result directly. A Stop hook also runs later than necessary. Option C improves contextual relevance but does not guarantee compliance; path-scoped rules remain instructions rather than enforcement controls.

The hook should be stored in project settings so the workflow is shared by the team. The formatter’s exit status should also be monitored so configuration or syntax failures are visible instead of silently ignored.

Official references/topics: Claude Code Hooks; PostToolUse; Edit|Write Matchers; Deterministic Formatting Enforcement.

Option D allows orchestration effort to scale with the actual request. A simple factual query may require only the web-search agent and a direct coordinator response, whereas a comparative investigation may require web research, document analysis, synthesis, and report generation. Anthropic’s Building Effective AI Agents describes the orchestrator-workers pattern as a central model dynamically identifying subtasks, delegating them, and combining the results. It is specifically appropriate when the required subtasks cannot be predicted reliably in advance. Anthropic’s multi-agent research architecture likewise emphasizes varying the number of agents and tool calls according to task complexity. Option A introduces an inflexible binary decision and still sends every non-factual request through the full pipeline. Option B requires labeled data, ongoing retraining, and reliable definitions of the “optimal” agent combination. Option C is easier to implement but will become brittle as new query types appear. A capable coordinator can examine the requested output, necessary evidence, source requirements, and analytical depth at runtime, then invoke only the specialists that materially contribute to the answer.

================

Option B targets the remaining failure mode: semantic recognition across heterogeneous document structures. Strict schemas eliminate malformed JSON and can guarantee that tool inputs conform to declared types, but they cannot force Claude to locate evidence that appears under unfamiliar headings or in atypical sections. Anthropic’s prompting guidance says that a few relevant, diverse, structured examples are among the most reliable ways to improve accuracy and consistency. Examples should therefore show inline citations, reference lists, numbered bibliographies, methodology sections, and methods embedded in introductions, each paired with the correct extracted structure. This teaches the intended evidence-location and granularity rules rather than merely repeating the same request. Option A retries an unchanged prompt and can reproduce the same omission. Option C introduces brittle regex rules that may miss nonstandard citations and mistake keyword mentions for methodology content. Option D suppresses validation failures by weakening the contract, but it does not improve extraction and would convert recoverable omissions into incomplete records. The examples should be drawn from real failure cases, evaluated on a held-out set, and expanded when monitoring reveals new layouts. Schema constraints and few-shot coverage solve different layers of reliability and should be used together.

================

Option A uses the limited review capacity as a risk-ranking mechanism rather than spending it uniformly. JSON Schema validation can prove that an output has the expected types and structure, but it cannot detect a semantically misplaced value such as a duration stored as an ingredient quantity. Field-level confidence can identify the exact parts of a record that warrant inspection, provided those scores are calibrated against labeled examples rather than trusted at face value. The labeled validation set must mirror the production distribution and include edge cases; Anthropic’s evaluation guidance explicitly recommends task-specific evaluations, representative distributions, measurable criteria, and automated grading where possible. Calibration then converts raw confidence into an empirically tested estimate of error risk, allowing reviewers to inspect the lowest-confidence records or fields first. Option C, random sampling, is useful for unbiased quality measurement and discovering new failure modes, but it does not maximize error detection within a fixed 20 percent review budget. Options B and D target narrow proxies—layout anomalies and missing required values—while the observed errors pass structural validation and may occur in ordinary-looking documents. The operational design should combine calibrated risk-based review with a smaller random audit sample to detect drift and overconfidence.

================

Option B preserves provenance as structured data throughout every handoff. Each finding should include the summarized claim and an associated source record containing its URL or document identifier, page or section location, relevant excerpt, and possibly retrieval date and confidence. The synthesis agent can combine claims without destroying this relationship, and the report generator can produce citations deterministically.

Anthropic’s multi-agent research system retains findings until a dedicated citation stage identifies the precise source locations supporting report claims. This demonstrates that citation provenance must remain available beyond initial retrieval and summarization.

Option A attempts to reconstruct provenance after it has been discarded, which is expensive and may associate a claim with a different or weaker source. Option C preserves evidence but wastes context by transferring approximately all raw search and document content. Option D is better than losing references entirely, but embedding citations only in prose makes programmatic validation, deduplication, and claim-to-source mapping less reliable. Structured content and source fields maintain both context efficiency and auditable attribution while allowing different agents to transform the narrative safely.

================

Option B uses the tool designed to search file contents. Anthropic’s Claude Code tools reference distinguishes Grep from Glob: Grep searches lines inside files, whereas Glob matches filenames and paths. Grep is built on ripgrep and accepts regular-expression patterns, so the opening parenthesis should be escaped as eval\( when the intention is to match the literal function call. The search can return matching files, line numbers, and surrounding context without loading every file into the model’s context window. Option A searches filenames resembling eval , not source files whose contents invoke the function. Option C follows only code reachable from a selected entry point and can miss test utilities, dynamically loaded modules, scripts, and dormant vulnerable code. Option D combines a recursive filename listing with text filtering; it still searches names rather than file contents. After Grep identifies matches, the agent should use Read on the relevant ranges to distinguish actual executable calls from comments, strings, or safe wrappers. Grep therefore provides the most complete and context-efficient initial security scan.

================

An aggregate accuracy value can conceal severe performance disparities. A system may achieve 97% overall accuracy while performing poorly on a low-volume document type, a critical financial field, or a specific edge case. Automating outputs solely from the aggregate figure could therefore expose downstream systems to concentrated, high-impact errors.

Anthropic’s evaluation guidance states that evaluations should be task-specific, reflect the real-world task distribution, and explicitly include edge cases. It also emphasizes multidimensional success criteria rather than reliance on a single global metric. ( https://docs.anthropic.com/en/docs/build-with-claude/develop-tests ) Option A applies those principles by stratifying performance according to document type and field. This reveals whether confidence is calibrated consistently and whether the proposed automation threshold remains safe for every operationally significant segment.

Option B is useful only after segment-level performance has been understood. Selecting a global threshold cannot correct a subgroup where confidence is systematically overstated. Option C is necessary governance work, but it treats the overall 97% result as though errors were uniformly distributed. Option D places unvalidated outputs into downstream systems and depends on passive error reporting, which may fail to detect silent corruption.

The correct deployment gate is therefore segmented validation, followed by threshold selection, downstream acceptance criteria, and a controlled pilot.

Official references/topics: Define Success Criteria; Task-Specific Evaluations; Edge-Case Coverage; Reliability Segmentation.

Option A gives the report generator the finished narrative context plus the evidence required to verify and cite its claims. The 3,000-token synthesis draft supplies organization and conclusions, while the source index preserves URLs, supporting excerpts, and claim-to-source mappings without transferring 120,000 tokens of raw search material. Anthropic’s effective context-engineering guidance recommends curating the smallest high-signal context and having subagents return condensed, distilled results rather than flooding later stages with their complete working history. Anthropic’s citation guidance likewise depends on retaining specific supporting passages and source locations. Option B is complete but inefficient and introduces context pollution that can reduce attention to relevant evidence. Option C separates citation insertion from the reasoning that produced the claims, making incorrect or unsupported matches more likely. Option D retains source names but discards the passages needed to demonstrate that each source actually supports the associated claim. The structured index should use stable source IDs and include only the excerpts necessary for the report’s factual claims, with validation that every cited claim maps to at least one evidence entry.

================

Option B removes the identified startup work while deliberately restoring the one source of project context the reviews require. Anthropic documents that --bare skips automatic discovery of hooks, skills, plugins, MCP servers, auto memory, and CLAUDE.md files. It is specifically intended for CI and scripted execution where fast, reproducible startup behavior is more important than loading every locally configured extension.

Because bare mode also skips the root CLAUDE.md, the pipeline must supply that content explicitly. --append-system-prompt-file ./CLAUDE.md loads the standards while retaining Claude Code’s default coding-agent behavior and tool guidance. Option A replaces the default system prompt but does not, by itself, establish the same minimal startup path as --bare; replacement also discards valuable default coding instructions. Option C can work technically but duplicates repository policy inside every pipeline invocation and creates configuration drift. Option D improves prompt-cache reuse by relocating machine-specific prompt sections, but it does not eliminate discovery of hooks, MCP servers, plugins, skills, and nested instructions. Bare mode plus an explicitly appended standards file directly addresses both performance and policy requirements. Claude Code bare-mode documentation

================

The agent is behaving inconsistently because every failure is represented identically. “Operation failed” contains no information about permanence, user correctability, authorization, or whether another attempt is likely to succeed. Adding structured fields converts the error into an actionable interface contract.

Anthropic instructs tool implementations to mark failures using an error flag and return information Claude can use to retry with corrected input, ask for clarification, or explain a limitation. Tool design should expose sufficient detail for the model to choose the correct next action rather than forcing it to infer the cause from a generic message. ( https://platform.claude.com/docs/en/agents-and-tools/tool-use/build-a-tool-using-agent?utm_source=chatgpt.com )

With error_category , isRetryable , and a causal description, the agent can retry transient network failures, request corrected identifiers for validation errors, and escalate or report permission failures without pointless repetitions. Option B retries permanent validation and permission failures unnecessarily. Option C adds another tool call and failure point merely to classify information the original tool already possesses. Option D relies on parsing variable text and examples rather than providing explicit machine-readable semantics.

The tool should preserve isError: true , return a stable structured error object, include a safe customer-facing message where appropriate, and avoid exposing internal secrets or stack traces.

Official references/topics: MCP error contracts, actionable tool results, retryability metadata, agent-computer interface design.