Banco de Respuestas Correctas: Repaso de 60 Preguntas
Banco de Respuestas Correctas: Repaso de 60 Preguntas
Compendio de las respuestas correctas y su justificación para las 60 preguntas del repaso del examen Claude Certified Architect. Cada bloque incluye el enunciado resumido (tal como aparece en la revisión oficial), la opción acertada y la explicación. Organizado por dominio para facilitar el estudio focalizado.
Los enunciados aparecen truncados porque la herramienta de revisión solo muestra el inicio de cada pregunta. Lo importante para el estudio es el patrón de respuesta correcta y su justificación.
Índice de dominios
- Customer Support Resolution Agent — 15 preguntas
- Claude Code for Continuous Integration — 15 preguntas
- Multi-Agent Research System — 15 preguntas
- Code Generation with Claude Code — 15 preguntas
1. Customer Support Resolution Agent
Pregunta 1.1 — Agent SDK Hook Patterns
Enunciado (extracto): Production logs reveal that the agent misinterprets data from your MCP tools: Un…
Respuesta correcta — opción C:
Use a PostToolUse hook to intercept tool results and apply formatting transformations before agent processing
Justificación:
Using a PostToolUse hook provides a centralized, deterministic point to intercept and normalize all tool outputs—including those from third-party MCP servers—before the agent processes them. This is the most maintainable approach because it applies transformations uniformly via code rather than relying on LLM interpretation or agent behavior.
Pregunta 1.2 — Self-Evaluation Patterns
Enunciado (extracto): Production metrics show that when your agent resolves complex cases involving bi…
Respuesta correcta — opción C:
Add a self-critique step where the agent evaluates its draft response for completeness—ensuring it addresses the customer’s concern, includes relevant context, and anticipates follow-up questions.
Justificación:
A self-critique step (evaluator-optimizer pattern) directly addresses the root cause of inconsistent explanation completeness by having the agent evaluate its own draft against specific criteria—such as policy context, timelines, and next steps—before presenting it. This catches case-specific gaps that vary across different complex scenarios without requiring human review.
Pregunta 1.3 — Conversation Context Management
Enunciado (extracto): Your support agent uses progressive summarization—when context reaches 70% capac…
Respuesta correcta — opción C:
Extract transactional facts (amounts, dates, order numbers) into a persistent “case facts” block included in each prompt, outside the summarized history.
Justificación:
Extracting transactional facts (amounts, dates, order numbers) into a persistent “case facts” block addresses the root cause: summarization is inherently lossy for precise details. By preserving critical information in a structured block outside the summarized history, these facts remain reliably available in every prompt regardless of how many turns are summarized.
Pregunta 1.4 — Multi-step Workflow Orchestration
Enunciado (extracto): Production logs show that for simple requests like “refund order #1234”, your ag…
Respuesta correcta — opción C:
Decompose the request into distinct concerns, then investigate each in parallel using shared customer context before synthesizing a resolution.
Justificación:
Decomposing the request into distinct concerns and investigating them in parallel with shared customer context directly addresses both core issues: it eliminates redundant data fetching by reusing context across concerns and reduces total tool calls by parallelizing investigations before synthesizing a unified resolution.
Pregunta 1.5 — Tool Selection Reliability
Enunciado (extracto): Production logs reveal a consistent pattern: when customers include “account” in…
Respuesta correcta — opción C:
The system prompt contains keyword-sensitive instructions that steer behavior based on terms like “account,” creating unintended tool selection patterns
Justificación:
This is the most likely root cause because the systematic, keyword-triggered pattern (78% vs 93%) strongly suggests explicit routing logic in the system prompt that reacts to the word “account” and directs the agent toward customer-related tools. Since the tool descriptions are already well-written and unambiguous, the discrepancy points to prompt-level instructions creating unintended behavioral steering.
Pregunta 1.6 — Tool Selection Reliability
Enunciado (extracto): Production logs show the agent sometimes selects get_customer when `lookup_ord…
Respuesta correcta — opción D:
Add 4-6 examples targeting ambiguous scenarios, each showing reasoning for why one tool was chosen over plausible alternatives.
Justificación:
Targeting few-shot examples at the specific ambiguous scenarios where errors occur, with explicit reasoning about why one tool is preferred over another, directly teaches the model the comparative decision-making process it needs for edge cases. This approach is the most effective because worked examples demonstrating reasoning are better than declarative rules for nuanced tool selection.
Pregunta 1.7 — Escalation Decisions
Enunciado (extracto): Your agent achieves 55% first-contact resolution, well below the 80% target. Log…
Respuesta correcta — opción B:
Add explicit escalation criteria to your system prompt with few-shot examples demonstrating when to escalate versus resolve autonomously.
Justificación:
Adding explicit escalation criteria with few-shot examples directly addresses the root cause—unclear decision boundaries between straightforward and complex cases. This is the most proportionate and effective first intervention, as it teaches the agent precisely when to escalate versus resolve autonomously without requiring additional infrastructure.
Pregunta 1.8 — Multi-step Workflow Enforcement
Enunciado (extracto): Production data shows that in 12% of cases, your agent skips get_customer enti…
Respuesta correcta — opción B:
Add a programmatic prerequisite that blocks
lookup_orderandprocess_refundcalls untilget_customerhas returned a verified customer ID.
Justificación:
Adding a programmatic prerequisite that blocks downstream tools until get_customer returns a verified customer ID provides a deterministic guarantee that the required sequence is followed. This is the most effective approach because it removes the possibility of the agent skipping verification, regardless of LLM behavior.
Pregunta 1.9 — Ambiguous Result Handling
Enunciado (extracto): Your get_customer tool returns all matches when searching by name. Claude curr…
Respuesta correcta — opción C:
Instruct Claude to ask for an additional identifier (email, phone, or order number) when
get_customerreturns multiple matches, before taking any customer-specific action.
Justificación:
Asking the user for an additional identifier (such as email, phone, or order number) is the most reliable way to disambiguate multiple matches, since the user has definitive knowledge of their own identity. One extra conversational turn is a small cost to eliminate the 15% error rate caused by incorrect customer selection.
Pregunta 1.10 — Tool Selection Reliability
Enunciado (extracto): In testing, you notice the agent frequently calls get_customer when users ask…
Respuesta correcta — opción B:
Review tool descriptions to ensure they clearly distinguish each tool’s purpose
Justificación:
Tool descriptions are the primary input the model uses to decide which tool to call. When an agent consistently selects the wrong tool, the first diagnostic step is to examine whether the tool descriptions clearly distinguish each tool’s purpose and specify when each should be used.
Pregunta 1.11 — Tool Interface Design
Enunciado (extracto): Production logs show the agent frequently calls get_customer when users ask ab…
Respuesta correcta — opción C:
Expand each tool’s description to include input formats it handles, example queries, edge cases, and boundaries explaining when to use it versus similar tools.
Justificación:
Expanding tool descriptions to include input formats, example queries, edge cases, and boundaries directly addresses the root cause—minimal descriptions that leave the LLM unable to distinguish between similar tools. This is a low-effort, high-leverage first step that improves the primary mechanism LLMs use for tool selection.
Pregunta 1.12 — Parallel Tool Execution
Enunciado (extracto): Production metrics show your agent averages 4+ API round-trips per resolution. A…
Respuesta correcta — opción C:
Prompt Claude to batch tool requests per turn, and return all tool results together before the next API call.
Justificación:
Prompting Claude to batch related tool requests in a single turn, and returning all results together before the next API call, leverages Claude’s native ability to request multiple tools simultaneously. This is the most effective approach because it directly addresses the sequential calling pattern with minimal architectural changes.
Pregunta 1.13 — Tool Selection Reliability
Enunciado (extracto): Your agent handles single-concern requests with 94% accuracy (e.g., “I need a re…
Respuesta correcta — opción C:
Add few-shot examples to your prompt demonstrating the correct reasoning and tool sequence for multi-concern requests.
Justificación:
Adding few-shot examples demonstrating correct reasoning and tool sequencing for multi-concern requests is the most effective approach because the agent already handles individual concerns well at 94% accuracy—it simply needs pattern guidance for handling multiple concerns in one message. This is a low-cost, proven technique that directly addresses the root cause of the agent failing to decompose and properly route parameters across multiple requests.
Pregunta 1.14 — Escalation Decisions
Enunciado (extracto): After calling get_customer and lookup_order, the agent has retrieved all ava…
Respuesta correcta — opción D:
The customer requests a price match against a competitor. Your policies allow adjustments for price drops on your own site within 14 days but are silent on competitor pricing. The agent should escalate for policy interpretation.
Justificación:
This represents a genuine policy gap where the company’s guidelines cover own-site price drops but are silent on competitor price matching, meaning the agent cannot fabricate a policy and must escalate for human judgment on how to interpret or extend existing rules.
Pregunta 1.15 — Agentic Loop Fundamentals
Enunciado (extracto): You’re implementing the agentic loop for your support agent. After each API call…
Respuesta correcta — opción B:
Check the
stop_reasonfield in Claude’s response—continue when it equals"tool_use"and stop when it equals"end_turn".
Justificación:
This is correct. The stop_reason field is Claude’s explicit, structured signal for loop control: "tool_use" indicates Claude wants to execute a tool and receive the results back, while "end_turn" indicates Claude has completed its response and the loop should terminate.
2. Claude Code for Continuous Integration
Pregunta 2.1 — False Positive Reduction
Enunciado (extracto): Your automated code review averages 15 findings per pull request, with developer…
Respuesta correcta — opción B:
Require Claude to include its reasoning and confidence assessment inline with each finding
Justificación:
Including reasoning and confidence assessments inline with each finding directly addresses the investigation time bottleneck by allowing developers to quickly evaluate findings without clicking into each one separately. This approach respects the constraint against filtering, since all findings remain visible while making triage significantly faster.
Pregunta 2.2 — CI/CD Integration
Enunciado (extracto): Your pipeline script runs `claude “Analyze this pull request for security issues…
Respuesta correcta — opción A:
Add the
-pflag:claude -p "Analyze this pull request for security issues"
Justificación:
The -p (or --print) flag is the documented way to run Claude Code in non-interactive mode. It processes the given prompt, outputs the result to stdout, and exits without waiting for user input, making it ideal for CI/CD pipelines.
Pregunta 2.3 — CI/CD Integration
Enunciado (extracto): After an initial automated review generates 12 findings, a developer pushes new…
Respuesta correcta — opción D:
Include prior review findings in context, instructing Claude to only report new or still-unaddressed issues.
Justificación:
Including prior review findings in context allows Claude to intelligently distinguish between new issues and those already addressed by recent commits. This approach maintains thorough analysis while leveraging Claude’s reasoning ability to avoid redundant feedback on fixed code.
Pregunta 2.4 — Classification Consistency
Enunciado (extracto): Your automated code review system shows inconsistent severity ratings—similar is…
Respuesta correcta — opción B:
Include explicit severity criteria in your prompt with concrete code examples for each severity level
Justificación:
Including explicit severity criteria with concrete code examples directly addresses the root cause of inconsistency by removing ambiguity about what each severity level means. This is a proven prompt engineering technique that gives the model clear reference points for classification, leading to more reliable and predictable severity assignments.
Pregunta 2.5 — Context Provision Methods
Enunciado (extracto): Your automated review generates test case suggestions for each PR. When reviewin…
Respuesta correcta — opción A:
Include the existing test file in the context so Claude can identify what scenarios are already covered
Justificación:
Including the existing test file in the context directly addresses the root cause of duplication: Claude can only avoid suggesting already-covered scenarios if it knows what tests already exist. This gives Claude the information needed to reason about which suggestions would be genuinely new and valuable.
Pregunta 2.6 — False Positive Reduction
Enunciado (extracto): Analysis of your automated code review shows significant variation in false posi…
Respuesta correcta — opción D:
Temporarily disable high false positive categories (style, naming, documentation) and run only high-precision categories while improving prompts.
Justificación:
Temporarily disabling the high false positive categories (style, naming, documentation) immediately stops trust erosion by removing the noise that causes developers to dismiss all findings, while preserving the value of high-precision categories like security and correctness. This approach allows time to improve prompts for the problematic categories before re-enabling them, rebuilding trust through demonstrated accuracy.
Pregunta 2.7 — Few-Shot Prompting
Enunciado (extracto): Your automated reviews identify valid issues but developers report the feedback…
Respuesta correcta — opción C:
Add 3-4 few-shot examples showing the exact format you want: issue identified, code location, specific fix suggestion
Justificación:
Few-shot examples are the most effective technique for achieving consistent output format when instructions alone produce variable results. Providing 3-4 examples showing the exact desired format (issue, location, specific fix) gives the model a concrete pattern to follow, which is more reliable than abstract instructions.
Pregunta 2.8 — Prompt Specificity
Enunciado (extracto): Your automated review analyzes comments and docstrings. The current prompt instr…
Respuesta correcta — opción B:
Specify explicit criteria: flag comments only when their claimed behavior contradicts actual code behavior
Justificación:
Specifying explicit criteria—flag comments only when their claimed behavior contradicts actual code behavior—directly addresses the root cause by replacing the vague instruction with a precise definition of what constitutes a problem. This eliminates both false positives on acceptable patterns and false negatives on genuinely misleading comments.
Pregunta 2.9 — Multi-Instance Verification
Enunciado (extracto): Your team uses Claude Code to generate code suggestions, but you notice a patter…
Respuesta correcta — opción B:
Have a second, independent Claude Code instance review the changes without seeing the generator’s reasoning.
Justificación:
Using a second, independent Claude Code instance without access to the generator’s reasoning directly addresses the root cause by eliminating confirmation bias. This fresh perspective mirrors the benefit of human peer review, where a different team member catches issues the original author rationalized away.
Pregunta 2.10 — Task Decomposition
Enunciado (extracto): A pull request modifies 14 files across the stock tracking module. Your single-p…
Respuesta correcta — opción C:
Split into focused passes: analyze each file individually for local issues, then run a separate integration-focused pass examining cross-file data flow.
Justificación:
Splitting the review into focused per-file passes directly addresses the root cause of attention dilution, ensuring consistent depth and catching local issues reliably. A separate integration-focused pass then handles cross-file concerns like data flow dependencies, covering both dimensions of review quality.
Pregunta 2.11 — Batch Processing
Enunciado (extracto): Your CI pipeline includes two Claude-powered code review modes: a `pre-merge-com…
Respuesta correcta — opción B:
Deep analysis only
Justificación:
Deep analysis is the ideal candidate for batch processing because it already runs overnight, tolerates latency, and uses a polling model to check for completion before posting results—perfectly matching the Message Batches API’s asynchronous, poll-based design while capturing the 50% cost savings.
Pregunta 2.12 — Batch Processing
Enunciado (extracto): Your team wants to reduce API costs for automated analysis. Currently, real-time…
Respuesta correcta — opción C:
Use batch processing for the technical debt reports only; keep real-time calls for pre-merge checks.
Justificación:
This is the correct approach because the Message Batches API’s up to 24-hour processing time with no guaranteed latency SLA makes it ideal for overnight technical debt reports but unsuitable for blocking pre-merge checks where developers are waiting. This matches each workflow to the appropriate API based on its latency requirements.
Pregunta 2.13 — Batch Processing
Enunciado (extracto): The code review component works iteratively: Claude analyzes a changed file, the…
Respuesta correcta — opción C:
The asynchronous model prevents executing tools mid-request and returning results for Claude to continue analysis.
Justificación:
This is correct. The batch API’s asynchronous fire-and-forget model means there is no mechanism to intercept a tool call mid-request, execute the tool, and return results for Claude to continue its analysis. This fundamentally breaks iterative tool-calling workflows that require multiple rounds of tool invocation and response within a single logical interaction.
Pregunta 2.14 — Batch Processing
Enunciado (extracto): Your CI/CD system performs three types of Claude-powered analysis: (1) quick sty…
Respuesta correcta — opción A:
Use synchronous calls for PR style checks; use the Message Batches API for weekly security audits and nightly test generation.
Justificación:
This is the correct approach. PR style checks block developers and require immediate responses via synchronous calls, while weekly security audits and nightly test generation are scheduled tasks with flexible timelines that can easily tolerate the up-to-24-hour batch processing window, capturing the 50% cost savings on both.
Pregunta 2.15 — Structured Output
Enunciado (extracto): Your CI pipeline runs the Claude Code CLI (with --print mode) using CLAUDE.md…
Respuesta correcta — opción D:
Use CLI flags
--output-format jsonand--json-schemato enforce structured findings, then parse output to post inline comments via the GitHub API.
Justificación:
Using --output-format json with --json-schema enforces structured output at the CLI level, guaranteeing well-formed JSON with the required fields (file path, line number, severity, suggested fix) that can be reliably parsed and posted as inline PR comments via the GitHub API. This is the most effective approach because it leverages native CLI capabilities designed specifically for structured output enforcement.
3. Multi-Agent Research System
Pregunta 3.1 — Error Propagation
Enunciado (extracto): During a materials research task, the web search subagent queries three source c…
Respuesta correcta — opción D:
Distinguish access failures (timeout) needing retry decisions from valid empty results (“0 results”) representing successful queries.
Justificación:
This is correct because a timeout (access failure) and ‘0 results’ (valid empty result) are semantically distinct outcomes requiring different responses. Distinguishing them enables the coordinator to retry the timed-out patent database while accepting the empty industry report results as a valid and informative finding.
Pregunta 3.2 — Error Propagation
Enunciado (extracto): The web search subagent returns results for only 3 of 5 requested source categor…
Respuesta correcta — opción D:
Structure the synthesis output with coverage annotations indicating which findings are well-supported versus which topic areas have gaps due to unavailable sources.
Justificación:
Structuring the synthesis output with coverage annotations embodies graceful degradation with transparency, allowing downstream consumers and end users to understand which findings are well-supported and which topic areas have gaps. This approach preserves the value of completed work while propagating uncertainty information so informed decisions can be made about confidence levels.
Pregunta 3.3 — Multi-Agent Orchestration
Enunciado (extracto): When researching a broad topic, you observe that the web search agent and docume…
Respuesta correcta — opción B:
Have the coordinator explicitly partition the research space before delegation, assigning distinct subtopics or source types to each agent
Justificación:
Having the coordinator explicitly partition the research space before delegation is the most effective approach because it addresses the root cause—unclear task boundaries—before any work begins. This preserves the benefits of parallel execution while preventing duplicated effort and wasted tokens.
Pregunta 3.4 — Error Propagation
Enunciado (extracto): The document analysis subagent frequently encounters failures when processing PD…
Respuesta correcta — opción D:
Have the subagent implement local recovery for transient failures and only propagate errors it cannot resolve to the coordinator, including what was attempted and any partial results obtained.
Justificación:
Implementing local recovery for transient failures within the subagent follows the principle of handling errors at the lowest level capable of resolving them. This reduces excessive coordinator involvement while still escalating truly unresolvable issues with full context, including what recovery was attempted and any partial results obtained.
Pregunta 3.5 — Multi-Agent Orchestration
Enunciado (extracto): A colleague suggests having the document analysis agent send its output directly…
Respuesta correcta — opción B:
The coordinator can observe all interactions, handle errors consistently, and decide what information each subagent should receive
Justificación:
This is correct. The coordinator pattern provides centralized visibility into all interactions, consistent error handling across the system, and fine-grained control over what information each subagent receives, which are the primary advantages of hub-and-spoke communication.
Pregunta 3.6 — Tool Distribution
Enunciado (extracto): During testing, you observe that the synthesis agent frequently needs to verify…
Respuesta correcta — opción C:
Give the synthesis agent a scoped
verify_facttool for simple lookups, while complex verifications continue delegating to the web search agent through the coordinator.
Justificación:
Providing a scoped fact-verification tool handles the 85% of simple lookups directly, eliminating most round-trips while preserving the coordinator-based delegation path for the 15% of complex verifications. This applies the principle of least privilege, keeping the synthesis agent focused on its primary task while still reducing latency significantly.
Pregunta 3.7 — Tool Distribution
Enunciado (extracto): Production logs reveal a consistent pattern: requests to “analyze the quarterly…
Respuesta correcta — opción D:
Rename the web search tool to
extract_web_resultsand update its description to “processes and returns information retrieved from web searches and URLs.”
Justificación:
Renaming the web search tool to extract_web_results and updating its description to clearly reference web searches and URLs directly addresses the root cause by eliminating the semantic overlap between the two tools’ names and descriptions. This makes each tool’s purpose unambiguous, allowing the coordinator to correctly distinguish between document analysis and web search tasks.
Pregunta 3.8 — Tool Distribution
Enunciado (extracto): When designing the system, you gave the document analysis agent access to a gene…
Respuesta correcta — opción A:
Replace
fetch_urlwith aload_documenttool that validates URLs point to document formats.
Justificación:
Replacing the general-purpose tool with a document-specific tool that validates URLs point to document formats addresses the root cause by constraining capability at the interface level. This follows the principle of least privilege, making the undesired search behavior impossible rather than merely discouraged.
Pregunta 3.9 — Multi-Agent Orchestration
Enunciado (extracto): After running the system on the topic “impact of AI on creative industries,” you…
Respuesta correcta — opción B:
The coordinator agent’s task decomposition is too narrow, resulting in subagent assignments that don’t cover all relevant domains of the topic.
Justificación:
The coordinator’s logs directly reveal it decomposed the broad topic into only three visual arts subtasks (digital art, graphic design, photography), completely omitting music, writing, and film. Since the subagents all executed their assigned tasks correctly, the narrow decomposition by the coordinator is clearly the root cause of the missing coverage.
Pregunta 3.10 — Error Propagation
Enunciado (extracto): The document analysis subagent encounters a corrupted PDF file it cannot parse.…
Respuesta correcta — opción B:
Return the error with context to the coordinator agent, letting it decide how to proceed.
Justificación:
Returning the error with context to the coordinator agent is the most effective approach because it enables the coordinator to make an informed decision—such as skipping the file, trying an alternative parsing method, or notifying the user—while maintaining visibility into the failure.
Pregunta 3.11 — Multi-Agent Orchestration
Enunciado (extracto): The document analysis agent discovers that two credible sources contain directly…
Respuesta correcta — opción D:
Complete the document analysis with both figures included, explicitly annotate the conflict with source attribution, and let the coordinator decide how to reconcile before passing to synthesis.
Justificación:
This is the most effective approach because it respects separation of concerns: the document analysis agent completes its primary task without blocking, preserves both conflicting data points with explicit source attribution, and appropriately defers the reconciliation decision to the coordinator, which has the broader context needed to resolve the conflict.
Pregunta 3.12 — Long Context Position Effects
Enunciado (extracto): Production monitoring reveals inconsistent synthesis quality. When aggregated re…
Respuesta correcta — opción A:
Place a key findings summary at the beginning of the aggregated input and organize detailed results with explicit section headers for easier navigation.
Justificación:
Placing a key findings summary at the beginning leverages the primacy effect, ensuring critical information occupies the most reliably attended position. Adding explicit section headers throughout the aggregated input helps the model navigate and attend to middle-section content, directly mitigating the ‘lost in the middle’ phenomenon.
Pregunta 3.13 — Multi-Agent Orchestration
Enunciado (extracto): During testing, combined outputs from the web search agent (85K tokens including…
Respuesta correcta — opción B:
Modify upstream agents to return structured data (key facts, citations, relevance scores) instead of verbose content and reasoning
Justificación:
Modifying upstream agents to return structured data (key facts, citations, relevance scores) addresses the root cause by reducing token volume at the source while preserving essential information. This eliminates verbose page content and reasoning chains that inflate token counts without adding value for the synthesis step.
Pregunta 3.14 — Multi-Agent Orchestration
Enunciado (extracto): The web search and document analysis agents have both completed their tasks and…
Respuesta correcta — opción B:
The coordinator passes both sets of findings to the synthesis agent for unified integration
Justificación:
This is correct because the orchestrator-workers pattern requires the coordinator to maintain centralized control by collecting results from subagents and routing them to the appropriate next component—in this case, the synthesis agent, which is specifically designed to unify and integrate findings into a coherent output.
Pregunta 3.15 — Error Propagation
Enunciado (extracto): The web search subagent times out while researching a complex topic. You need to…
Respuesta correcta — opción D:
Return structured error context to the coordinator including the failure type, the attempted query, any partial results, and potential alternative approaches.
Justificación:
Returning structured error context—including the failure type, attempted query, partial results, and alternative approaches—gives the coordinator all the information it needs to make intelligent recovery decisions, such as retrying with a modified query or proceeding with partial results. This is the best approach because it preserves maximum context for informed decision-making at the coordination level.
4. Code Generation with Claude Code
Pregunta 4.1 — Custom Slash Commands
Enunciado (extracto): Your team created an /analyze-codebase skill that performs comprehensive code…
Respuesta correcta — opción D:
Add
context: forkto the skill’s frontmatter to run the analysis in an isolated sub-agent context
Justificación:
Using context: fork in the skill’s frontmatter runs the analysis in an isolated sub-agent context, which prevents the verbose output from polluting the main conversation’s context window and causing Claude to lose track of the original task. This preserves full analysis capability while keeping the main session responsive.
Pregunta 4.2 — Custom Slash Commands
Enunciado (extracto): Your team has created a /migration skill that generates database migration fil…
Respuesta correcta — opción B:
Add
argument-hintfrontmatter to prompt for required parameters, usecontext: forkto isolate execution, and restrictallowed-toolsto file write operations.
Justificación:
This approach correctly uses three distinct skill configuration features to address each issue: argument-hint frontmatter shows expected parameters during autocomplete (addressing missing arguments), context: fork isolates execution in a subagent context separate from conversation history (preventing context bleeding from earlier conversations), and allowed-tools restricts tool access to only file write operations (preventing destructive actions).
Pregunta 4.3 — CLAUDE.md Modular Organization
Enunciado (extracto): Your team’s CLAUDE.md file has grown to over 500 lines, mixing TypeScript conven…
Respuesta correcta — opción A:
Create separate markdown files in
.claude/rules/, each covering one topic (e.g.,testing.md,api-conventions.md)
Justificación:
This is correct. Claude Code supports a .claude/rules/ directory where you can create separate markdown files for topic-specific guidelines (e.g., testing.md, api-conventions.md), allowing teams to organize large instruction sets into focused, maintainable modules.
Pregunta 4.4 — Custom Slash Commands
Enunciado (extracto): You want to create a custom /review slash command that runs your team’s standa…
Respuesta correcta — opción D:
In the
.claude/commands/directory in the project repository
Justificación:
Placing custom slash commands in the .claude/commands/ directory within the project repository is correct because these files are version-controlled and automatically available to every developer who clones or pulls the repo. This is the designated location for project-scoped custom commands in Claude Code.
Pregunta 4.5 — Path-Specific Rule Configuration
Enunciado (extracto): Your codebase has distinct areas with different coding conventions: React compon…
Respuesta correcta — opción A:
Create rule files in
.claude/rules/with YAML frontmatter specifying glob patterns to conditionally apply conventions based on file paths
Justificación:
Using rule files in .claude/rules/ with YAML frontmatter and glob patterns (e.g., **/*.test.tsx, src/api/**/*.ts) allows conventions to be automatically and deterministically applied based on file paths, regardless of where those files are located in the directory structure. This is the most maintainable approach because it handles cross-cutting concerns like test files spread throughout the codebase without requiring duplication or manual intervention.
Pregunta 4.6 — Custom Slash Commands
Enunciado (extracto): You’re creating a custom /explore-alternatives skill that your team uses to br…
Respuesta correcta — opción B:
Add
context: forkto the skill’s frontmatter.
Justificación:
The context: fork frontmatter option runs the skill in an isolated sub-agent context, so the exploration discussion does not pollute the main conversation history. This prevents abandoned approaches and exploratory context from influencing subsequent implementation work.
Pregunta 4.7 — Iterative Refinement
Enunciado (extracto): You’ve asked Claude Code to implement a function that transforms API responses i…
Respuesta correcta — opción D:
Provide 2-3 concrete input-output examples showing the expected transformation for representative API responses.
Justificación:
Providing concrete input-output examples is the most effective approach because it eliminates the ambiguity inherent in prose descriptions by showing Claude exactly what the expected transformation looks like. This directly addresses the root cause—misinterpretation of prose requirements—by giving unambiguous, concrete targets for field nesting and timestamp formatting.
Pregunta 4.8 — Skills vs CLAUDE.md Scope
Enunciado (extracto): Your CLAUDE.md has grown to over 400 lines containing coding standards, testing…
Respuesta correcta — opción C:
Keep universal standards in CLAUDE.md and create Skills for task-specific workflows (PR reviews, deployments, migrations) with trigger keywords
Justificación:
This is the most effective approach because CLAUDE.md content is loaded for every conversation, ensuring coding standards and testing conventions are always applied, while Skills are invoked on-demand when Claude detects relevant trigger keywords, making them ideal for task-specific workflows like PR reviews, deployments, and migrations.
Pregunta 4.9 — Plan Mode vs Direct Execution
Enunciado (extracto): You need to add Slack as a new notification channel. The existing codebase has c…
Respuesta correcta — opción B:
Enter plan mode to explore the integration options and their architectural implications, then present a recommendation before implementing.
Justificación:
This is correct because the Slack integration involves multiple valid approaches with significantly different architectural implications, and the requirements are ambiguous. Using plan mode to explore trade-offs between webhooks, bot tokens, and Slack Apps allows for an informed recommendation and team alignment before committing to an implementation path.
Pregunta 4.10 — Custom Slash Commands
Enunciado (extracto): You’ve created a /commit skill in .claude/skills/commit/SKILL.md that your t…
Respuesta correcta — opción B:
Create a personal version in
~/.claude/skills/with a different name like/my-commit
Justificación:
This is correct. Since project skills take precedence over personal skills with the same name, the developer must use a different skill name (like /my-commit) in their personal ~/.claude/skills/ directory to ensure their custom version is accessible alongside the team’s project skill.
Pregunta 4.11 — CLAUDE.md Configuration Hierarchy
Enunciado (extracto): Your team has been using Claude Code for several months. Recently, three develop…
Respuesta correcta — opción D:
The guideline exists in the original developers’ ~/.claude/CLAUDE.md files (user-level) instead of the project’s .claude/CLAUDE.md. Move the instruction to the project-level file so all team members receive it.
Justificación:
This is the most likely cause: if the error handling guideline was added to each original developer’s user-level ~/.claude/CLAUDE.md rather than the project’s .claude/CLAUDE.md, new team members would not receive it. Moving the instruction to the project-level configuration file ensures all current and future team members automatically receive the guideline.
Pregunta 4.12 — Custom Slash Commands
Enunciado (extracto): You’ve found that including 2-3 full exemplar endpoint implementations as contex…
Respuesta correcta — opción C:
Create a skill that references the exemplar endpoints and includes pattern-following instructions, invoked on-demand via slash command.
Justificación:
Creating a skill with the exemplar endpoints and pattern-following instructions allows on-demand invocation via a slash command, ensuring the context is loaded only when generating new endpoints and not during unrelated tasks like bug fixes or code reviews.
Pregunta 4.13 — Subagent Delegation Strategy
Enunciado (extracto): You’re adding error handling wrappers to external API calls across a 120-file co…
Respuesta correcta — opción C:
Use the Explore subagent for Phase 1 to isolate verbose output and return a summary, then continue Phases 2-3 in the main conversation.
Justificación:
Using the Explore subagent for Phase 1 is ideal because it isolates the verbose discovery output in a separate context, returning only a concise summary to the main conversation. This preserves the main context window for the collaborative design and consistent implementation phases where retained context is most valuable.
Pregunta 4.14 — Plan Mode vs Direct Execution
Enunciado (extracto): You’ve been assigned to restructure the team’s monolithic application into micro…
Respuesta correcta — opción B:
Enter plan mode to explore the codebase, understand dependencies, and design an implementation approach before making changes.
Justificación:
Using plan mode to explore the codebase, understand dependencies, and design an approach before making changes is the correct strategy for a complex architectural restructuring like breaking apart a monolith. This allows safe exploration and informed decision-making about service boundaries before committing to potentially costly changes across dozens of files.
Pregunta 4.15 — MCP Server Integration
Enunciado (extracto): Your team wants to add a GitHub MCP server to enable PR lookups and CI status ch…
Respuesta correcta — opción A:
Add the server to a project-scoped
.mcp.jsonwith environment variable expansion (${GITHUB_TOKEN}) for authentication, and document the required environment variable in your project README.
Justificación:
Using a project-scoped .mcp.json with environment variable expansion (${GITHUB_TOKEN}) is the idiomatic approach—it provides a single, version-controlled source of truth for the team’s MCP configuration while allowing each developer to supply their own credentials through environment variables. Documenting the required variable in the README ensures easy onboarding without ever committing secrets.