Claude Certified Architect · ES Professional Foundations Glosario FAQ Simulador

CCAR-F · Examen de práctica

60 preguntas · 120 minutos · corte 720/1000

Material no oficial. Preguntas originales escritas contra los objetivos publicados en la guía del examen. No reproducen contenido del examen real. Ver aviso completo.

Las respuestas están en respuestas.md. Para practicar con corrección automática y cronómetro, usá el simulador.


1. Selecciona 2

D2 · Claude Code: configuración y workflows

A platform team is enabling AI-assisted development across twelve squads. Which TWO measures scale the practice without scaling the risk?

2. Selecciona UNA

D2 · Claude Code: configuración y workflows

Reviews of AI-authored pull requests almost never raise findings, yet defects keep reaching production. What is the most likely cause?

3. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

An agent in production starts calling an unexpected tool on a minority of requests. What does the on-call engineer do first?

4. Clasifica cada fila

D2 · Claude Code: configuración y workflows

For each situation, identify the practice that applies.

# Escenario Tu respuesta
1 Two squads produce different code style from the same assistant.
2 AI-written changes pass review, yet defects keep reaching production.
3 An agent begins calling an unexpected tool on a subset of requests.
4 An unattended pipeline job could run a destructive command.
5 A new hire should inherit the team's prompt standards on day one.

Opciones: shared configuration in version control · review from an independent context · inspect the traces first · restrict what automation may execute

5. Selecciona 2

D1 · Arquitectura agéntica y orquestación

A fixed workflow for supplier onboarding now fails on roughly a fifth of cases, where staff step in manually to finish the job. Which TWO signals would justify moving this stage to an autonomous agent?

6. Selecciona 2

D1 · Arquitectura agéntica y orquestación

A supervisor coordinates four specialist agents in a regulated claims process. Which TWO responsibilities belong to the supervisor rather than to the specialists?

7. Selecciona 2

D1 · Arquitectura agéntica y orquestación

An architect is splitting a long analysis task into sequenced subtasks. Which TWO practices make the decomposition real rather than cosmetic?

8. Selecciona 2

D1 · Arquitectura agéntica y orquestación

Which TWO conditions indicate that a single augmented model call is enough, with no workflow or agent around it?

9. Clasifica cada fila

D1 · Arquitectura agéntica y orquestación

For each situation, identify who should decide the next step.

# Escenario Tu respuesta
1 The five stages of a monthly regulatory report that has not changed in two years.
2 Which log query to run next while diagnosing a production outage.
3 Whether to release a payment that exceeds the agreed approval threshold.
4 The order in which four specialist agents run inside a regulated pipeline.
5 How many sources to consult before answering an open research question.

Opciones: the architect, at design time · the model, at run time · a human reviewer, per case

10. Selecciona UNA

D1 · Arquitectura agéntica y orquestación

An assistant handles two kinds of request: simple status lookups that arrive constantly, and complex disputes that need deep analysis. Which design serves both?

11. Selecciona UNA

D1 · Arquitectura agéntica y orquestación

A pipeline extracts data, transforms it and writes it to a system of record. Where does validation belong?

12. Selecciona UNA

D1 · Arquitectura agéntica y orquestación

In a four-stage chain, the third stage fails on about one request in twenty. What belongs in the design?

13. Selecciona UNA

D3 · Prompt engineering y salida estructurada

A summarisation feature runs on the largest model in the family. Evaluation shows a mid-tier model matches its quality on this task at a third of the cost. What should the architect do?

14. Clasifica cada fila

D3 · Prompt engineering y salida estructurada

For each symptom, identify the technique that fixes it.

# Escenario Tu respuesta
1 Output format is close but inconsistent from one run to the next.
2 A critical rule buried in 12,000 tokens of context is followed erratically.
3 The same 9,000-token policy block is re-sent and re-charged on every call.
4 Answers about the middle of a large document are unreliable and expensive.
5 The house table layout is described in prose but never reproduced exactly.

Opciones: few-shot examples · move the rule to an edge · prompt caching · retrieve only what is relevant

15. Selecciona UNA

D3 · Prompt engineering y salida estructurada

A classification task has eight well-defined categories and clear written definitions for each. Accuracy is already high without examples. What follows?

16. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

A support conversation runs long and the earliest messages start falling out of the window. What preserves quality?

17. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

A team wants to expose forty internal endpoints to an agent. Tool-selection accuracy in testing drops sharply past roughly a dozen tools. What is the sound design response?

18. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

A retrieval index holds documents from many client organisations. What keeps one client from ever seeing another's content?

19. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

A knowledge base gets edits all day long, and the assistant must reflect them within minutes. What indexing approach fits?

20. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

An internal API exposes forty endpoints. The team is deciding how to expose it to the agent. What should drive the tool boundaries?

21. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

An agent platform needs alerting. Which signal deserves a page to the on-call engineer?

22. Selecciona UNA

D1 · Arquitectura agéntica y orquestación

A report needs four independent sections researched and then written as one document. How should the work be split across agents?

23. Selecciona UNA

D1 · Arquitectura agéntica y orquestación

An orchestrator dispatches subagents whose number is only known once the problem is read. Which pattern is this?

24. Selecciona 2

D1 · Arquitectura agéntica y orquestación

A subagent receives its assignment from an orchestrator. Which TWO parts of that assignment should a person define in advance, not the orchestrator at runtime?

25. Selecciona 2

D1 · Arquitectura agéntica y orquestación

A team is deciding whether a multi-agent design is worth its overhead. Which TWO conditions make it pay off?

26. Selecciona 2

D1 · Arquitectura agéntica y orquestación

Results from parallel subagents come back in incompatible shapes and cannot be combined. Which TWO fixes address the cause?

27. Selecciona 2

D2 · Claude Code: configuración y workflows

A PreToolUse hook exits with code 2 after a blocked command. Which TWO things happen next?

28. Selecciona 2

D2 · Claude Code: configuración y workflows

A monorepo holds twelve applications, each with its own conventions. Which TWO practices keep context usage sane?

29. Selecciona UNA

D3 · Prompt engineering y salida estructurada

A downstream service parses the model's output as JSON and crashes when a field is missing. What is the correct fix?

30. Selecciona UNA

D3 · Prompt engineering y salida estructurada

A prompt mixes instructions, a reference document and a user question, and the model keeps confusing them. What is the most direct improvement?

31. Selecciona UNA

D3 · Prompt engineering y salida estructurada

An output will be read by a person, not parsed by a program. What follows for its format?

32. Selecciona UNA

D3 · Prompt engineering y salida estructurada

Few-shot examples were added to a classification prompt and accuracy on one rare category got worse. What is the likely cause?

33. Selecciona UNA

D3 · Prompt engineering y salida estructurada

An extraction step must return a consistent output shape. Prose instructions describing the format keep producing variations. What works better?

34. Selecciona UNA

D3 · Prompt engineering y salida estructurada

A review prompt says 'be conservative and report only high-confidence findings'. Results are inconsistent between runs. What is the defect?

35. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

A tool wraps an internal API with eleven optional parameters. The agent fills them inconsistently. What is the better design?

36. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

A tool returns an empty list when the underlying service times out. What is wrong with that?

37. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

Two agents from different companies must exchange work without either exposing internal tools. Which mechanism fits?

38. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

Two tools share the description 'Retrieves information' and the agent keeps picking the wrong one. What fixes it?

39. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

A project's MCP server needs an API token, and the configuration file is committed to the repository. How is that handled?

40. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

A developer wants to try an MCP server on their own, without affecting teammates. Where does that configuration go?

41. Selecciona UNA

D4 · Diseño de herramientas e integración MCP

An agent burns several tool calls each session just discovering what data is available. What reduces that?

42. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

A long agent session degrades as it goes: earlier facts get dropped and answers drift. What is the sound response?

43. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

A retrieval step returns nothing relevant for a user question. What should the system do?

44. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

In a long support conversation, the order number and the amount keep getting lost after the history is summarised. What fixes it?

45. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

A long input is processed reliably at the beginning and the end, but details in the middle are missed. What is the mitigation?

46. Selecciona UNA

D5 · Manejo de contexto y fiabilidad

A long investigation will span several sessions. How are the findings preserved?

47. Selecciona 2

D1 · Arquitectura agéntica y orquestación

A coordinator spawns subagents with the Task tool. Which TWO statements about their context are correct?

48. Selecciona 2

D1 · Arquitectura agéntica y orquestación

Which TWO situations call for starting a fresh session rather than continuing the current one?

49. Clasifica cada fila

D1 · Arquitectura agéntica y orquestación

For each stop_reason value, identify what the loop should do next.

# Escenario Tu respuesta
1 tool_use
2 end_turn
3 max_tokens
4 stop_sequence
5 tool_use returned after appending previous results

Opciones: execute the requested tools and continue · show the result to the user · raise the output budget and retry · treat the custom stop sequence as the boundary

50. Selecciona 2

D2 · Claude Code: configuración y workflows

Which TWO statements about hook security are accurate?

51. Selecciona 2

D2 · Claude Code: configuración y workflows

Which TWO facts about permissions in settings.json are correct?

52. Clasifica cada fila

D2 · Claude Code: configuración y workflows

For each need, identify the Claude Code mechanism that fits.

# Escenario Tu respuesta
1 The team's preferred naming conventions, which the model should follow.
2 A guarantee that no destructive migration command ever executes.
3 Declaring which Bash commands are allowed without a prompt.
4 A specialised procedure the model should pull in when it applies.
5 Blocking a tool call and telling the model why it was refused.

Opciones: CLAUDE.md · hook · settings.json · skill or subagent

53. Selecciona 2

D2 · Claude Code: configuración y workflows

A CLAUDE.md pulls in other files with the @path import syntax. Which TWO statements are correct?

54. Selecciona 2

D2 · Claude Code: configuración y workflows

A team moves its conventions from one long CLAUDE.md into .claude/rules/ files with paths frontmatter. Which TWO gains does that produce?

55. Selecciona 2

D2 · Claude Code: configuración y workflows

Which TWO statements about the built-in /compact and /memory commands are correct?

56. Clasifica cada fila

D2 · Claude Code: configuración y workflows

For each need, identify where the definition belongs.

# Escenario Tu respuesta
1 A review checklist the whole team should share.
2 A personal shortcut one developer uses across every project.
3 A guarantee that a destructive command never runs.
4 A noisy investigation routine that should not fill the session.
5 MCP servers every contributor to this repository needs.

Opciones: project-level, committed to the repository · user-level, in the home directory · a hook registered on a tool event · a skill with context: fork

57. Selecciona UNA

D3 · Prompt engineering y salida estructurada

An extraction schema marks every field as required. Documents that lack a field come back with invented values. What is the fix?

58. Selecciona UNA

D3 · Prompt engineering y salida estructurada

A classification enum keeps forcing borderline documents into the closest available category. What design change helps?

59. Selecciona UNA

D3 · Prompt engineering y salida estructurada

Invoice totals occasionally disagree with the sum of their line items, and nothing downstream notices. Which extraction design surfaces it?

60. Clasifica cada fila

D4 · Diseño de herramientas e integración MCP

For each need, identify the built-in tool that fits.

# Escenario Tu respuesta
1 Find every test file matching a naming pattern across the repository.
2 Locate which files mention a specific error message.
3 Load a file in full to trace how a function flows.
4 Change one unique line inside an existing file.
5 List all TypeScript files under a components directory.

Opciones: Glob · Grep · Read · Edit