These domain notes and sample questions come from the same material used in the study workspace. Read them here without an account; sign in to use the full question bank and save your progress.
Agentic Architecture & Orchestration
Choose how work flows through a Claude system. Start with clear success criteria, then decide whether a single call, a fixed workflow, or an autonomous agent is justified.
- Use fixed workflows when steps are predictable; use agents when the next step depends on discoveries.
- Separate model judgment from deterministic authorization and side-effect controls.
- Give delegated work explicit inputs, evidence requirements, stopping conditions, and a clear owner.
Primary source: Building effective agents
Tool Design & MCP Integration
A tool is a contract between model reasoning and application behavior. Good schemas, precise descriptions, useful errors, and narrow permissions make that contract dependable.
- Describe when to use a tool, required inputs, side effects, and limitations.
- For client tools, your application executes the request and returns a correlated result.
- Treat remote tool output as data; MCP connectivity does not replace authorization.
Primary source: Define tools
Claude Code Configuration & Workflows
Configure Claude Code so that instructions, tools, permissions, and automation match the repository and the task. Validate changes with the project’s own checks.
- Use CLAUDE.md for concise project guidance; use permissions and hooks for enforcement.
- Explore and plan before a broad change. Use isolated work when tasks could conflict.
- Run automated workflows with scoped access and inspect machine-readable results.
Primary source: Claude Code project memory
Prompt Engineering & Structured Output
Define what a good answer looks like before changing the prompt. Separate instructions from source material, demonstrate edge cases, and validate both format and meaning.
- Use explicit criteria and representative examples rather than vague requests.
- Schema compliance checks structure; business rules and source grounding check meaning.
- Evaluate changes on held-out examples, including failure cases and ambiguous inputs.
Primary source: Structured outputs
Context Management & Reliability
Keep the right evidence available as a task grows. Preserve provenance, handle failures explicitly, and measure whether the complete system remains dependable.
- Retain decisions and unresolved work when compacting; retrieve detail when needed.
- Distinguish missing evidence, empty results, and failed retrieval.
- Bound retries and latency, observe failures, and escalate when the evidence is insufficient.
Primary source: Effective context engineering
Claude Architect Prep