Referência

Appendix

Tecnologias e conceitos que você pode ver, o que está explicitamente dentro e fora do escopo e o caminho recomendado para se preparar.

Tecnologias e conceitos

Claude Agent SDK

Agent definitions, agentic loops, stop_reason handling, hooks (PostToolUse, tool call interception), subagent spawning via Task tool, allowedTools configuration

Model Context Protocol (MCP)

MCP servers, tools, resources, isError flag, tool descriptions, tool distribution, .mcp.json configuration, env-var expansion

Claude Code

CLAUDE.md hierarchy (user/project/directory), .claude/rules/ path-scoping, .claude/commands/, .claude/skills/ frontmatter (context: fork, allowed-tools, argument-hint), plan mode, direct execution, /memory, /compact, --resume, fork_session, Explore subagent

Claude Code CLI

-p / --print for non-interactive mode, --output-format json, --json-schema for structured CI output

Claude API

tool_use with JSON schemas, tool_choice ("auto", "any", forced), stop_reason ("tool_use", "end_turn"), max_tokens, system prompts

Message Batches API

50% cost savings, up to 24-hour window, custom_id correlation, polling, no multi-turn tool calling

JSON Schema

Required vs optional, enum types, nullable fields, "other" + detail patterns, strict mode for syntax-error elimination

Pydantic

Schema validation, semantic validation errors, validation-retry loops

Built-in tools

Read, Write, Edit, Bash, Grep, Glob - purposes and selection criteria

Few-shot prompting

Targeted examples for ambiguous scenarios, format demonstration, generalization

Prompt chaining

Sequential task decomposition into focused passes

Context window management

Token budgets, progressive summarization, lost-in-the-middle, context extraction, scratchpad files

Session management

Resumption, fork_session, named sessions, session context isolation

Confidence scoring

Field-level confidence, calibration with labeled validation sets, stratified sampling

✓ No escopo

Agentic loop implementation: control flow on stop_reason, tool result handling, termination conditions
Multi-agent orchestration: coordinator-subagent patterns, decomposition, parallel execution, iterative refinement
Subagent context management: explicit context passing, structured state persistence, crash recovery via manifests
Tool interface design: effective descriptions, splitting vs consolidating, naming to reduce ambiguity
MCP tool and resource design: resources for catalogs, tools for actions, description quality for adoption
MCP server configuration: project vs user scope, env-var expansion, multi-server access
Error handling and propagation: structured responses, transient vs business vs permission, local recovery
Escalation decision-making: explicit criteria, honoring preferences, policy gap identification
CLAUDE.md configuration: hierarchy, @import patterns, .claude/rules/ globs
Custom commands and skills: project vs user scope, context: fork, allowed-tools, argument-hint
Plan mode vs direct execution: complexity assessment, architectural decisions, single-file changes
Iterative refinement: I/O examples, test-driven iteration, interview pattern, sequential vs parallel
Structured output via tool_use: schema design, tool_choice, nullable fields to prevent hallucination
Few-shot prompting: ambiguous targeting, format consistency, false-positive reduction
Batch processing: appropriateness, latency tolerance, failure handling by custom_id
Context window optimization: trimming outputs, structured fact extraction, position-aware ordering
Human review workflows: confidence calibration, stratified sampling, accuracy segmentation
Information provenance: claim-source mappings, temporal data, conflict annotation, coverage gaps

✕ Fora do escopo

Fine-tuning Claude models or training custom models
Claude API authentication, billing, or account management
Detailed implementation of specific programming languages or frameworks
Deploying or hosting MCP servers (infrastructure, networking, orchestration)
Claude's internal architecture, training process, or model weights
Constitutional AI, RLHF, or safety training methodologies
Embedding models or vector database implementation details
Computer use (browser automation, desktop interaction)
Vision/image analysis capabilities
Streaming API implementation or server-sent events
Rate limiting, quotas, or API pricing calculations
OAuth, API key rotation, or authentication protocol details
Specific cloud provider configurations (AWS, GCP, Azure)
Performance benchmarking or model comparison metrics
Prompt caching implementation details (beyond knowing it exists)
Token counting algorithms or tokenization specifics

Recomendações de preparação para o exame

1Build an agent with the Claude Agent SDK: a complete agentic loop with tool calling, error handling, and session management. Practice spawning subagents and passing context.
2Configure Claude Code for a real project: CLAUDE.md hierarchy, path-specific rules in .claude/rules/, custom skills with frontmatter (context: fork, allowed-tools), and at least one MCP server.
3Design and test MCP tools: descriptions that differentiate similar tools, structured error responses with categories and retryable flags, and tool-selection reliability tests with ambiguous requests.
4Build a structured data extraction pipeline: tool_use with JSON schemas, validation-retry loops, optional/nullable fields, and batch processing with the Message Batches API.
5Practice prompt engineering: few-shot examples for ambiguous scenarios, explicit review criteria to reduce false positives, and multi-pass review architectures.
6Study context management: extracting structured facts from verbose outputs, scratchpad files for long sessions, and subagent delegation to manage context limits.
7Review escalation and human-in-the-loop patterns: when to escalate (policy gaps, customer requests, inability to progress) vs resolve, and confidence-based review routing.
8Complete the Practice Exam before sitting the real exam - it mirrors the scenarios and format and explains answers to reinforce understanding.