CCDV-F Claude Certified Developer-Foundations Questions and Answers

A is the correct lifecycle architecture. The supplied Claude Certified Developer Foundations item explicitly selects A . A session is a state boundary, so its lifecycle must define when accumulated conversation state remains valid and when that state must be discarded. Anthropic's Managed Agents documentation establishes that a session maintains conversation history across multiple interactions and provides explicit session operations including creation, update, archive, and deletion.

A therefore addresses all three required controls: expiration , reset , and fresh-session creation . Expiration prevents indefinitely retained conversational state. Reset triggers allow the application to clear state when events such as logout, workflow completion, tenant switching, or explicit user reset occur. Fresh-session rules prevent context belonging to an earlier task from influencing a logically unrelated conversation.

B relies exclusively on timeout and does not cover event-driven resets. C places lifecycle correctness on users instead of the application. D applies global reset behavior and omits independent per-session expiration semantics.

Relevant Claude Developer topics: Agent Architecture, session lifecycle, multi-turn state, context boundaries, state reset, stale-context prevention, and session management .

===============

D follows the principle of choosing the simplest architecture that satisfies demonstrated requirements . The supplied examination item explicitly marks D. In this scenario, portability is neither a present functional requirement nor a foreseeable roadmap requirement. Adding an abstraction therefore creates measurable complexity without delivering an identified product capability.

Anthropic provides official general-purpose SDKs for Claude in multiple languages. These expose the Messages API directly while adding idiomatic interfaces, type safety, streaming support, retries, and error handling. Anthropic separately describes compatibility layers and framework-specific libraries as alternative integration surfaces rather than requirements for ordinary Claude development.

A third-party abstraction becomes rational when there is a genuine requirement for multiple providers, vendor switching, standardized cross-model interfaces, or an existing architectural platform that mandates it. None exists here. Building that flexibility speculatively increases dependencies, debugging surface, version-compatibility work, and the possibility that provider-specific Claude capabilities are hidden behind a lowest-common-denominator API.

A optimizes for hypothetical future requirements. B doubles operational complexity. C creates an even larger maintenance burden.

Relevant Claude Developer topics: Claude App Design, SDK selection, abstraction boundaries, dependency management, YAGNI, integration architecture, and provider portability requirements .

===============

Production deployment does not terminate the software lifecycle. Once a Claude application is actively serving users, it enters an operations and maintenance phase in which developers and operators monitor performance, evaluate model behavior, respond to failures, control costs, manage security, and evolve the implementation as requirements or model capabilities change.

Anthropic's official developer documentation explicitly separates the journey into build, evaluate-and-ship, and operate stages. The operating stage includes workspace administration, API-key management, usage monitoring, and model migration. This confirms that production deployment is a transition into ongoing operation rather than the endpoint of development.

For LLM applications, maintenance is especially important because production traffic can expose input distributions and failure modes that were not fully represented during pre-release evaluation. Operational data should feed back into evaluations, prompt improvements, guardrails, architecture decisions, and model-version planning.

A incorrectly treats deployment as final. B may be useful but is only one governance activity, not the overall lifecycle phase. D assumes a mandatory organizational separation that is neither required nor generally desirable.

The supplied exam source marks C. Relevant topics: Systems Life Cycle, production operations, monitoring, maintenance, incident response, evaluation, and continuous system evolution.

===============

Option C matches Anthropic's efficiency-first model-selection guidance. For a straightforward, high-volume workload where latency and unit cost are explicit constraints, the correct starting point is a faster, economical model that can meet the task's quality threshold. Anthropic specifically lists high-volume straightforward tasks, tight latency requirements, and cost-sensitive implementations as cases where an efficiency-first model choice is appropriate.

The crucial qualification is that “smaller” does not mean accepting inadequate quality. The team should evaluate the candidate against representative classification examples and defined accuracy criteria. If it passes, moving to a larger model adds cost and often latency without delivering required business value.

A selects a mid-tier model by convention rather than workload evidence. B multiplies inference calls, generally increasing both latency and cost for a simple classification problem. D optimizes maximum capability even though the task does not require frontier-level reasoning.

Therefore, C is the appropriate initial model choice, followed by workload-specific validation. Relevant Study Guide topics: Claude model selection, efficiency-first design, classification workloads, throughput, latency, per-request economics, evaluation, and quality/cost tradeoffs.

===============

Option B follows standard secrets-management practice and Anthropic's explicit guidance for API credentials. API keys are authentication secrets and should not be embedded in source code or committed to repositories. Anthropic's authentication documentation explicitly recommends storing API keys in a secrets manager, rotating them periodically, and revoking credentials suspected of compromise. Anthropic SDKs can load Claude credentials from environment variables such as ANTHROPIC_API_KEY, allowing the secret to be injected at runtime instead of compiled into the application.

The same principle applies to external service credentials used by a Claude application. Development, staging, and production should normally receive independently scoped credentials through the deployment environment or secret-management infrastructure.

A creates a high-probability credential leak because repository history can retain secrets even after a later deletion. C violates isolation and least privilege by sharing credentials across services and users. D uses email as an uncontrolled secret-distribution channel and creates inconsistent manual configuration.

Therefore, B supplies credentials only at runtime while keeping them outside source control and enabling rotation and environment-specific access. Relevant Study Guide topics: API-key management, secrets managers, environment configuration, credential rotation, least privilege, and secure configuration.

===============

Option A treats a Claude application as production software subject to the same disciplined engineering controls as other services while adding tests for its model integration. CI should provide repeatable feedback before merge, including deterministic unit/integration tests, linting, type or static checks where applicable, security checks, and organization-specific quality gates.

Claude-specific behavior should also be tested systematically rather than reserved for manual release verification. Anthropic's evaluation guidance recommends task-specific test cases reflecting realistic inputs and edge cases, measurable success criteria, and automation whenever practical. Claude Code documentation also explicitly identifies CI automation, including GitHub Actions and GitLab CI/CD, as supported development workflows.

B is excessive because every pull request should not directly deploy to production merely to obtain end-to-end coverage; testing environments and staged deployment exist for that purpose. C makes linting dependent on individual developer discipline rather than a shared merge gate. D delays Claude integration verification until pre-release and makes a repeatable automated check manual.

Therefore, A provides comprehensive and consistent pull-request validation. Relevant Study Guide topics: CI/CD, automated testing, integration testing, linting, evaluations, regression prevention, and quality gates.

===============

Option D is the strongest enforcement design because an unconditional content policy requires an application-level control between model generation and user delivery. Prompt instructions are valuable for steering Claude, but they are probabilistic controls and should not be treated as the sole enforcement mechanism when prohibited categories must never be exposed.

Anthropic’s guardrail guidance recommends layered safeguards including screening, validation, monitoring, and filtering rather than relying exclusively on prompts. Its prompt-leak guidance specifically recommends output screening and post-processing, including deterministic techniques such as keyword matching, regular expressions, or other text-processing mechanisms where appropriate.

A improves the probability of policy compliance but cannot guarantee that every generated response will satisfy an externally defined application policy. B explicitly abandons the requirement. C detects violations only after exposure, which is unsuitable when the content must not reach users.

A production architecture can combine system instructions, structured classification, policy engines, deterministic rules, and model-based moderation, but the decisive requirement is enforcement before output delivery. Relevant Claude Developer topics are guardrails, output filtering, content moderation, deterministic enforcement, defense in depth, safe application boundaries, and production Claude application design.

===============

Option A is correct because the safest guardrail architecture is defense in depth, not dependence on one control. Anthropic’s guardrail guidance explicitly recommends combining multiple safeguards: input screening and validation, hardened system instructions, safe handling of untrusted tool content, least-privilege permissions, output screening, monitoring, and red-team testing. The important engineering property is independence: if one layer misses an attack or unsafe request, another layer can still prevent harmful behavior or block delivery.

Option B places control only at the output boundary. Human review can be useful for high-risk cases, but it does not protect tool execution, data access, prompt injection, or other failures that can occur before final output. Option C relies on a single system-prompt control; system instructions are probabilistic and cannot provide complete enforcement against adversarial or malformed inputs. Option D is weaker still because model-level safety alone does not enforce application-specific policies.

Therefore, A best matches Claude Developer security guidance: layer preventive, detective, and enforcement controls so there is no single guardrail whose failure exposes the application. Relevant Study Guide topics: guardrails, prompt injection, input validation, output screening, least privilege, defense in depth, and production monitoring.

===============

The supplied Claude Developer examination source selects A . The purpose of choosing an agent SDK rather than implementing a custom Messages API loop is to consume the SDK's higher-level abstractions. Re-implementing dispatch, iteration, and state handling would discard much of the value provided by the SDK.

Anthropic's current documentation distinguishes low-level tool-use loops from higher-level SDK-managed abstractions. With a manually implemented Messages API workflow, application code must inspect stop_reason, execute requested client tools, append tool_result blocks, preserve conversation history, and repeat until Claude completes the turn. Anthropic's higher-level tooling can instead encapsulate this repetitive control flow. Current migration guidance also confirms that Agent SDK @tool functions are automatically dispatched by the SDK and that agents, tools, and sessions are first-class SDK concepts.

B is incorrect because structured tool use should not be replaced with informal plain-text calls. C introduces an unnecessary custom state dependency when no external persistence requirement exists. D duplicates control-loop functionality despite the explicit decision to use the SDK.

Relevant Claude Developer topics: Agent SDK, tool definitions, dispatch, agentic loops, conversation state, tool execution, and abstraction selection .

===============

Option C is correct because operational error handling should distinguish failures by impact, urgency, and recoverability. Anthropic’s API documentation already separates error conditions by type and documents different recovery behaviors. For example, transient connection failures, rate limits, and many 5xx errors are automatically retried by official SDKs with exponential backoff, while other failures require application correction or investigation. Treating all of those conditions as the same operational severity creates either alert fatigue or missed incidents.

A production design should classify errors using criteria such as user impact, data integrity risk, security exposure, persistence after retry, affected request volume, and whether automatic recovery succeeds. High-severity failures that threaten service availability or correctness can page the on-call engineer. Recoverable or isolated failures can be logged with structured metadata, request IDs, and metrics for later review.

Option A pages on everything and quickly makes alerts noisy. Option B destroys observability for non-paging errors. Option D preserves the original defect by assigning the same severity to every condition. Therefore, C follows sound Claude application operations. Relevant Study Guide topics: API errors, retries, observability, incident response, severity classification, structured logging, and production operations.

===============

Option B establishes a controlled boundary between probabilistic model output and deterministic downstream code. When structured data is machine-consumed, malformed JSON must be recognized as an explicit application error rather than allowed to propagate into parsers, databases, or other services.

Anthropic’s Structured Outputs documentation identifies malformed JSON, missing fields, inconsistent types, and schema violations as exactly the kinds of failures that can break downstream systems when unconstrained output is used. Current Claude capabilities can constrain responses using JSON Schema, and SDK helpers can provide parsing and validation.

Even when an application uses an older or unconstrained generation path, it should parse against the expected schema, record validation failure, and enter a bounded recovery path such as retry, repair, fallback, or controlled rejection. A human review of every request is unscalable. C removes a useful structured interface instead of making it reliable. D performs uncontrolled blind retries and provides no schema-aware error handling or bounded fallback strategy.

Therefore, B gives the application explicit failure semantics and protects downstream systems. Relevant Study Guide topics: structured output, JSON validation, schema enforcement, retries, fallback handling, defensive parsing, and downstream reliability.

===============

Option A applies the correct context-engineering strategy: remove stale tool results once they no longer contribute useful information to subsequent reasoning. Agentic workflows frequently accumulate search results, file contents, API responses, and intermediate artifacts. Keeping all of them indefinitely consumes the finite context window, raises token cost, and can reduce model focus by introducing low-value information.

Anthropic specifically documents tool result clearing for this situation. Context Editing can remove older tool results when the conversation grows, while preserving recent interactions and optionally retaining tools whose results must remain available. Anthropic describes old tool outputs such as retrieved files or search results as candidates for clearing after Claude has processed them.

Prompt caching in B solves a different problem: it can lower cost and latency for repeated static prompt prefixes, but cached tokens still constitute context and therefore do not eliminate context-window pressure. C changes model capability without solving the architectural cause. D maximizes context pollution.

The correct architecture is therefore to preserve high-signal state while pruning ephemeral intermediate outputs. This aligns with Claude Developer coverage of context engineering, long-running agents, context-window management, tool-result clearing, and efficient agent state management. Anthropic's broader context-engineering guidance likewise emphasizes curating the smallest high-signal context necessary for successful inference.

===============

D is the only option that addresses all three identified production-readiness defects. The supplied Claude Developer item explicitly selects D. A functioning happy path is insufficient for a maintainable Claude integration.

Prompt and model selection are configuration concerns that will change as prompts are evaluated, model versions evolve, or environments require different behavior. They should therefore be separated from unrelated business logic rather than scattered as inline constants. Claude API failures must also be handled deliberately. Anthropic documents typed SDK exceptions and defined HTTP error categories, including invalid requests, authentication failures, rate limits, server errors, overload, and timeouts. The official SDKs additionally retry appropriate transient failures.

Tests are required to verify the integration boundary, including successful behavior, malformed or unexpected responses, API error handling, and critical user workflows. Approving code without those controls pushes known reliability debt directly into production.

A and B knowingly merge incomplete production behavior. C improves configuration and test coverage but leaves a known API-failure path unhandled.

Relevant Claude Developer topics: SW Eng Foundations, code review, separation of configuration, error handling, integration testing, API resilience, maintainability, and production readiness .

===============

Option C is the correct application of context isolation. A specialized step should receive the minimum relevant information required for its own task rather than inheriting an ever-growing global transcript. This improves signal-to-noise ratio, limits accidental cross-task influence, controls token usage, and makes individual components easier to evaluate.

Anthropic’s context-engineering guidance explicitly identifies multi-agent architectures as a technique for long-horizon work. Specialized subagents can operate with their own context windows and return condensed results to an orchestrating agent rather than exposing every agent to every intermediate detail. This architecture protects each reasoning process from irrelevant history while allowing the overall system to preserve necessary state.

Options A and D represent the opposite approach: indiscriminately loading the full prior history. Larger context is not automatically better; Anthropic warns that excessive context can introduce context pollution and degrade retrieval or attention to important information. B also fails because a single global prompt does not isolate state or tailor the information available to each processing stage.

Therefore, C best implements scoped reasoning boundaries. Relevant Claude Developer topics are multi-agent architecture, subagents, context isolation, context engineering, orchestration, and long-running workflow design.

===============

Option C applies the appropriate context-management technique. The issue is not simply whether the older conversation can physically fit into the context window; it is that stale details remain prominent enough to interfere with the model’s current task. Effective context engineering optimizes signal quality, not merely maximum token retention.

Anthropic’s context-editing documentation describes compaction as summarizing accumulated history and replacing the full history with a structured summary when context becomes large. This preserves important task state while substantially reducing low-value detail that can distract later reasoning.

A retains precisely the stale material producing the problem and therefore does not address context drift. B destroys all useful continuity between turns, including valid goals, decisions, and intermediate state. D is similarly excessive because limiting Claude to only the latest turn discards information that may still be required.

Compaction provides the correct middle ground: retain durable conclusions, current objectives, unresolved issues, and other high-signal state while compressing obsolete conversational detail. Relevant Study Guide topics: context engineering, compaction, conversation history, long-running agents, stale context, context drift, and high-signal state preservation.

===============

Option D applies the correct hybrid pattern. Anthropic distinguishes workflows—where LLMs and tools execute along predefined code paths—from agents, where the model dynamically determines how to proceed. Anthropic recommends choosing the simplest architecture that meets the requirement: workflows provide predictability and consistency for well-understood tasks, while agents are valuable where flexible model-driven decisions are necessary.

Standard invoice processing is explicitly described as a small number of known steps. That makes a deterministic workflow the appropriate default because the path can be tested, monitored, and reproduced. The unexpected-format branch is different: the system must reason among rerouting, requesting clarification, or human escalation. That decision point is where agentic flexibility adds value.

A makes the entire process autonomous even though most steps do not require autonomy, increasing cost, latency, and unpredictability. B introduces multiple specialized agents despite no demonstrated need for that complexity. C collapses deterministic processing and exception handling into one monolithic model call, making validation and debugging harder.

Therefore, D combines workflow predictability with targeted agentic reasoning. Relevant Study Guide topics: workflows versus agents, routing, exception handling, hybrid agentic architecture, escalation, and complexity minimization.

===============

Option C is correct because Skills and MCP are first-class Claude extension mechanisms intended to make capabilities reusable rather than embedding one-off logic inside a single agent. Claude documentation describes Skills as reusable knowledge, instructions, and workflows that can be loaded when relevant. It describes MCP as the standard mechanism for connecting Claude to external services, tools, prompts, and resources. Claude Code plugins can package Skills and MCP servers for distribution across projects and teams.

Option A tightly couples the capability to one team’s agent and requires copying code, creating version drift and duplicated maintenance. Option B is technically reusable software, but it leaves every consuming Claude application responsible for its own integration and does not expose the capability through Claude’s native extension interfaces. Option D assumes an existing built-in tool is the correct abstraction even though the scenario specifically requires an independently reusable capability.

The precise choice between Skill and MCP depends on what is being reused: use a Skill for reusable instructions/workflows; use MCP when the capability exposes external data or actions. Therefore, C best reflects the Claude Developer extension model. Relevant topics: Agent Skills, MCP, plugins, tool integration, reuse, and cross-team capability distribution.

===============

The supplied question marks C , and Anthropic's API documentation directly supports it. Claude API rate limits are enforced using dimensions such as requests per minute (RPM), input tokens per minute (ITPM), and output tokens per minute (OTPM) . Anthropic also notes that short traffic bursts can exceed effective limits even when a longer-term average appears acceptable.

Applications should therefore determine the organization's actual configured limits, model peak traffic against those boundaries, throttle or queue work as necessary, and handle 429 responses correctly. Rate-limit responses include a retry-after value indicating when another request should be attempted. Anthropic's official SDKs automatically retry transient connection failures, rate-limit errors, and server errors with exponential backoff by default.

Streaming does not exempt requests from rate limits, making A technically incorrect. Consolidating payloads in B may reduce request count but could increase token consumption and does not by itself address all rate-limit dimensions. D may smooth traffic, but it is only one optimization and does not substitute for rate-limit-aware application logic.

Relevant Claude Developer topics: Claude API Mechanics, rate limits, RPM, ITPM, OTPM, HTTP 429, retry-after, exponential backoff, throttling, and capacity planning .

===============

Option C correctly explains a fundamental property of generative language models. The supplied examination material identifies C as the correct answer. Claude generates subsequent tokens probabilistically rather than retrieving a single fixed answer for each prompt. Consequently, identical or highly similar requests can produce variations in wording, ordering, explanation depth, and sometimes substantive details.

Anthropic's API documentation explicitly states that sampling parameters control randomness and, importantly, that even configurations historically using a temperature of 0.0 were not fully deterministic . Current newer Claude models increasingly manage sampling behavior internally, so application design should not assume byte-for-byte identical responses across executions.

Production systems therefore handle variation through architecture: schema validation for machine-consumed output, deterministic business-rule checks, retries where appropriate, evaluations for acceptable behavioral ranges, and application-level safeguards. Temperature adjustment may reduce variation on models that support the parameter, but it does not fundamentally convert an LLM into a deterministic function.

A incorrectly labels normal model behavior as a defect. B confuses network latency with generation variability. D overstates model snapshot behavior; pinning a model prevents silent model-version changes but does not eliminate sampling variability.

Relevant topics: sampling, nondeterminism, validation, retries, model versions, and robust Claude API integration .

===============

Option A is correct because model selection requires measurable non-functional requirements, not an undefined objective such as “best.” Anthropic's model-selection guidance explicitly frames the decision around capabilities, speed, and cost and recommends identifying those requirements before narrowing the candidate models.

“Quality” must also be translated into task-specific success criteria—for example classification accuracy, extraction correctness, acceptable false-positive rate, or human preference. Latency may be constrained by an interactive SLA, and cost may be defined per request, per user, or per completed task. Once those thresholds exist, representative evaluations can compare candidate models using the actual prompts and data.

B uses a generic benchmark that may not correlate with the application's workload. C chooses a model tier without requirements or evidence. D optimizes only for raw capability and ignores latency and cost, potentially producing an unnecessarily expensive architecture.

Thus, requirements definition precedes technology selection. Relevant Study Guide topics: Understanding Reqs, functional versus non-functional requirements, quality thresholds, latency budgets, cost constraints, acceptance criteria, workload-specific evaluation, and model-selection tradeoffs.

===============

Option D combines the three key properties of a reliable Claude tool: an explicit contract, clear tool-selection guidance, and controlled execution failure handling. Anthropic's tool documentation defines user tools using a name, detailed description, and JSON input_schema. The description should explain what the tool does, when it should and should not be used, parameter semantics, and relevant limitations. Anthropic emphasizes that precise descriptions materially improve Claude's ability to select the correct tool.

A clear schema prevents ambiguous parameter interpretation and allows validation before calling the external pricing API. Where stronger guarantees are required, Anthropic also supports strict tool use, which constrains generated tool inputs to the declared JSON Schema.

The application's execution layer must also convert pricing-service failures into explicit, handled error paths rather than uncontrolled exceptions. A deprives Claude of critical selection information. B increases malformed-call risk. C delegates infrastructure reliability to the reasoning loop instead of implementing appropriate integration error handling.

Therefore, D represents production-quality custom-tool construction. Relevant Study Guide topics: custom tools, JSON Schema, tool descriptions, validation, external API integration, and error handling.

===============

Option B is the correct fault-tolerance pattern for a long-running stateful workflow. A checkpoint captures enough durable execution state—completed steps, intermediate results, pending work, identifiers, and other necessary task state—to restart from a known consistent point rather than replaying the entire workflow after interruption.

This pattern is consistent with Claude's current stateful agent architecture. Anthropic's Managed Agents documentation describes persistent sessions that preserve conversation history across interactions. When a session becomes idle, its sandbox can be checkpointed so filesystem and execution artifacts are available when work resumes. The certification concept is broader than that specific hosted implementation: long-running agent systems should externalize recoverable state at meaningful boundaries.

A longer timeout does not protect against process crashes, infrastructure restarts, network failures, or deployment interruptions. C doubles resource consumption and creates consistency problems without providing deterministic recovery. D wastes completed work and may repeat external side effects.

Therefore, B provides controlled resumability and minimizes repeated processing. Relevant Study Guide topics: checkpointing, persistent state, resumable workflows, long-running agents, fault tolerance, idempotency, and recovery architecture.

===============

Option B reflects the fundamental generative behavior of large language models. Claude does not retrieve a fixed response template for a given prompt. Instead, it generates the response incrementally, selecting subsequent tokens according to the probability distribution produced by the model. Consequently, semantically equivalent runs can differ in wording, ordering, whether prompt language is quoted, and whether it is paraphrased.

Anthropic's API documentation explicitly describes sampling parameters as controlling randomness in subsequent-token selection and also states that even configurations intended to minimize randomness do not make output completely deterministic. Current Claude model generations increasingly manage sampling internally, but the underlying principle remains: generative output is not guaranteed to be textually identical across equivalent calls.

Therefore, seeing the original prompt echoed in one response and paraphrased in another does not, by itself, demonstrate malformed request construction. A real request-construction problem should instead be established through traces showing different system instructions, messages, parameters, or context.

A, C, and D incorrectly infer specific implementation faults without supporting evidence.

The supplied question identifies B as the correct response. Relevant topics: SW Eng Foundations, token generation, sampling, nondeterminism, debugging, and model behavior.

===============

Option B is the correct automation pattern. A CI pipeline must operate non-interactively, so Claude Code should be invoked in its programmatic/headless execution mode and given only the permissions required for the predefined workflow. Output can then be captured by the pipeline and supplied to subsequent build, testing, reporting, or deployment stages.

Claude Code's configuration system supports explicit permission rules in JSON settings. Anthropic documents that settings.json controls, among other behavior, what Claude Code can execute without asking. For example, permissions.allow can authorize specific commands while permissions.deny prevents sensitive operations. Project-level .claude/settings.json can be version-controlled so CI behavior is reproducible across runners.

This is substantially safer than C, which disables confirmation behavior globally and expands the blast radius beyond the CI workload. D defeats automation because a human must remain present. A is unnecessary because Claude Code is explicitly configurable for automation scenarios.

The principle is least-privilege non-interactive configuration: authorize only the commands and resources required by the pipeline and capture machine-readable output for downstream processing. Relevant Claude Developer topics are Claude Code automation, configuration management, permissions, CI/CD integration, headless execution, and reproducible project settings.

The supplied question identifies B as the intended answer. Tool schemas are contracts between Claude and executable application code. When the downstream service accepts only a defined set of arguments, relaxing that schema merely shifts invalid data farther into the system and increases runtime failures.

Anthropic's current tooling provides an even stronger implementation of this principle through strict tool use . Setting strict: true constrains tool inputs to the declared JSON Schema, preventing undeclared properties, missing required values, and incompatible parameter types where the supported schema subset is used. Anthropic explicitly recommends strict tool use for validated parameters, type-safe function calls, and reliable agentic workflows.

In a non-strict or legacy implementation, the application should still validate arguments before dispatch and convert invalid calls into structured tool errors that Claude can interpret and potentially correct. A prompt instruction can reinforce behavior, but it should not replace deterministic validation. C and D weaken the system boundary and knowingly send invalid calls downstream.

Therefore, maintain the contract rather than adapting the contract to malformed model output.

Relevant Claude Developer topics: Agent Construction, tool schemas, strict tool use, JSON Schema, parameter validation, structured errors, retry behavior, and defensive execution boundaries .

===============