71 Claude certification concepts, made scannable.

An agentic loop turns one API response into a controlled, multi-turn system.

Subagents are specialized, isolated agents spawned by a coordinator to handle domain-specific tasks while preserving conversation isolation.

stop_reason is the authoritative struct field that says why Claude stopped.

Session state is the running message list (and any external persistence) that gives an agentic loop continuity.

Escalation is the deterministic handoff path when policy thresholds, low confidence, or repeated failure conditions are hit.

Subagents hand off state three ways: structured messages (default - typed JSON envelopes through the orchestrator), tool results (tightest sync chain when receiver is a tool), or s

CLAUDE.md is Claude-specific; AGENTS.md is the emerging cross-tool convention (Codex, Cursor, Aider).

Task decomposition is the up-front architectural choice between two regimes: fixed decomposition (subtasks known in advance, run as a sequential prompt chain) and dynamic decomposi

Tool calling is how Claude decides to invoke external functions and pass structured arguments.

tool_choice is the parameter that controls whether Claude can decide to use a tool ("auto"), must call any tool ("any"), or must call a specific tool by name.

MCP is a communication standard that lets Claude access pre-built tools, resources, and prompts from specialized servers without you writing integration code.

Hooks are deterministic code that runs before or after tool calls.

Evaluation is how you measure tool-call correctness against ground-truth datasets.

Built-in tools, custom tools, Agent Skills, and MCP servers are complementary layers, not a power ranking — one agent often uses several at once.

CLAUDE.md is persistent project memory loaded automatically by Claude Code.

Plan Mode forces Claude Code to draft and approve a multi-step plan before executing.

Skills are reusable, on-demand task workflows defined in markdown with YAML frontmatter.

Checkpoints save and restore conversation state.

System prompts establish role, constraints, format, and tool guidance.

Prompt caching reduces cost (~90%) on repeated context like long system prompts and tool definitions.

Message Batches API: 50% discount for async, non-time-sensitive workloads.

Structured outputs guarantee Claude returns JSON matching a schema instead of natural language.

Vision lets Claude process images alongside text.

Streaming returns tokens as they're generated for low-latency UX.

Attention engineering is the discipline of placing critical context where the model attends most strongly, high in the prompt, in the system message, or in repeated facts blocks.

Delegation, Description, Discernment, Diligence, Anthropic's 4D framework for agent prompt design.

Prompt engineering is a craft of seven techniques that turn a brittle one-shot prompt into a production-grade contract: few-shot examples, iterative refinement against an eval suit

A multi-pass review architecture runs more than one independent Claude pass over the same artifact and reconciles the results, instead of trusting a single evaluation - it covers t

Context window management is how you keep long conversations within limits without dropping critical facts.

A case-facts block is an immutable set of transactional data (customer ID, order details, refund amount, policy limits) included at the top of every prompt during a multi-turn conv

Translating a business problem into a Claude solution means producing a build-ready requirements spec, not picking an architecture pattern.

An architect-level design task is never just "the prompt in the middle" - it is the full pipeline: how input arrives, how it is processed (augmented LLM, workflow, or agent loop),

Pick the cheapest tier that actually satisfies the task: an augmented LLM call for a single unaided prompt, a workflow when the steps are known in advance, an agent only when the m

Every architectural choice - pattern, model tier, guardrail overhead - has to trace back to a named business-value pillar an executive sponsor recognizes: efficiency, transformatio

Match the model tier to the task, not the other way round: Haiku for high-volume, straightforward work at the lowest cost/latency; Sonnet for the best speed/intelligence balance; O

Agent authentication and authorization is a stack of controls, not one setting: which identity the process runs as, what credentials each tool call carries, and which actions need

Every production Claude deployment sits on a capability/speed/cost triangle, and the exam trap is treating it as one dial ('use the biggest model') instead of independent levers -

Observability for a Claude agent fleet spans layers you must monitor separately: platform-level usage/cost accounting (Anthropic's Admin/Analytics APIs), Managed Agents session tra

As an agent's tool/skill/knowledge surface grows, an architect chooses between loading everything upfront (monolithic - predictable, but token-expensive) or exposing capabilities o

A Claude application fails in two structurally different places: the integration layer (the call never reached a valid model turn - bad auth, malformed request, rate limit) or mode

Guardrails are not one control but a layered system: model-level training behavior, input-level content screening, tool-level permission policies, process-level sandboxing, and hum

Architects must reason about failure modes that don't exist in deterministic software: non-determinism, prompt injection (direct and indirect), long-context jailbreak attacks, and

Compliance is a match-the-surface-to-the-regime exercise: an architect picks the deployment surface (API direct, AWS Bedrock, GCP Vertex, Microsoft Foundry, Claude for Government)

Ethical AI for an architect is not a vendor guarantee, it is three published artifacts to consult: Claude's Constitution (intended values, the training-time "final authority"), per

Discovery for a Claude-based solution has to surface things traditional SDLC discovery skips: whether the task even fits a nondeterministic system, what "good enough" output looks

An architect's technical trade-offs (model tier, workflow vs agent, guardrail depth) are business decisions in disguise, and Anthropic names the exact vocabulary to present them in

A Claude-based system's "SLA" is not one number - an architect has to keep at least three infrastructure-side concepts straight, plus a fourth for quality: observed historical upti

Handoff for a Claude-based architecture needs everything a traditional diagram needs (components, data flow, integration points) PLUS a set of GENERAL artifacts every Claude-based

Deployment isn't the finish line - an architect owns a continuous loop of discovery, design, handoff, monitoring, and iteration, because model behavior and usage patterns keep drif

Anthropic gives developers three layers for building a Claude agent, and they are NOT interchangeable names for the same thing: the Client SDK / Messages API (you write the tool lo

An agent's execution mechanism is the tool-use loop: call a tool, read real environmental feedback ("ground truth"), repeat until the task converges - that loop, not any framework,

Systems life cycle management is a general IT discipline - plan, develop, test, deploy, operate, maintain, retire - not an Anthropic invention; NIST's SDLC bulletin frames it as "t

The Claude Messages API is a plain REST API that accepts and returns JSON - POST /v1/messages with a stateless request/response contract - so the ordinary software-engineering disc

Claude is one model family reached through several host surfaces, claude.ai/Desktop, Claude Code, and the raw Messages API/SDKs, and each supplies a different (or absent) scaffold:

Claude is autoregressive: it predicts one token at a time from a probability distribution conditioned on everything before it, then appends that token and repeats.

For standard model inference, the Messages API (POST /v1/messages) is the primary endpoint, tools, structured outputs, and streaming are request parameters on that same endpoint, n

Prompt injection splits into two threat models with different defenses: direct injection / jailbreaks (your own user crafts a bypass) mitigated by input screening and hardened syst

"Be clear and direct" is Anthropic's base rule for every Claude prompt, but how you apply it changes with the type of work.

Accuracy and completeness are two independent checks, not one: an output can be fully correct on every claim it includes and still fail by silently dropping items, or it can cover

Hallucination is Anthropic's own term for Claude generating text that's factually incorrect or inconsistent with the material it was given, and it's a known limitation, not a rare

A first Claude draft is rarely the version you hand to its actual reader.

Choosing how a Claude response looks is really two independent decisions, not one three-way pick.

Claude.ai gives a business user four distinct surfaces, plain chat, Projects, Research, and Artifacts, and picking the right one is itself the skill this objective tests.

Beyond one-off Q&A, Claude is used as a research and planning partner across a three-stage arc: research a problem, draft an approach, then refine it over several turns before anyt

Integrating Claude into a workflow is a choice between augment (a human still owns the process, Claude drafts or triages inside it) and redesign (the process itself changes shape b

A Claude Project is a self-contained workspace with its own chat history and its own knowledge base, configured through two layers: project knowledge (uploaded files) and project i

A Project's knowledge splits into two mechanisms that behave differently: uploads are a static snapshot a user manually maintains, and connectors are live links to external systems

Before building a workflow around Claude, judge whether a task is one Claude is allowed and well-suited to do, not just whether it is technically possible.

Following your organization's AI governance is an employee-level checklist run before acting: is this data class allowed here, is this use case approved, does this output need revi

When a Claude output misses the mark, the fix Anthropic teaches is diagnostic, not a re-roll: define what a good answer looks like, compare the actual output against that bar, chan

Optimizing a Claude workflow for efficiency AND effectiveness spans more than one lever: tightening the prompt and output, right-sizing the model tier or its effort setting to the