KNOWLEDGE PRACTICE / INTERVIEW QUESTIONS
Interview questions are used to check knowledge understanding, practice expression and derive engineering solutions. Answer independently first, and then compare and analyze; when encountering weak points, return to the corresponding knowledge to explain and complete them.
Choose the direction of knowledge you are learning, explain the principles, premises and boundaries independently, and then analyze and find out the omissions. When you need to make up lessons, use "Learn Corresponding Knowledge" to return to the explanation; when you need to verify the implementation, complete the hands-on tasks on the page.
The questions are designed based on public official information and engineering boundaries. The scale, faults and goals in the cases are teaching assumptions. The reference materials support the technical mechanism, and the Q&A, migration scenarios and self-evaluation standards are designed by this site.
When preparing for the interview, explain the principles, boundaries, and trade-offs in your own words, and then check against the basic compliance, intermediate and advanced signals, and senior signals. Reading records and single answers do not directly represent mastery.
Exercises found: 64
Control and completion criteria in agent loops
Starting from the tool invocation loop, design termination conditions, budgets, cancellations, and verifiable completion statuses.
Combine deterministic workflows with bounded exploration
Choose a recoverable orchestration structure around defined processes, dynamic exploration, checkpoints, and manual approvals.
Task contracts and independent acceptance checks
Make the execution environment, permissions, status, evidence and completion judgment of the model into a project contract.
Benefits and coordination costs of multiple agents
Examining task parallelizability, coordination costs, evidence merging, and controlled experiments.
Algebra and business semantics of parallel state merging
Examine state modeling, reducers, concurrent merging and message deduplication.
Separate plan revisions from external facts
Examine dynamic plans, completed side effects, dependencies, and acceptance invariants.
Lossy summaries and retrievable source facts
Examining context budgeting, summary distortion, on-demand loading, and recovery verification.
Transfer responsibility and permissions during task handoff
Examining Handoff, Identity Propagation, Minimal Context, and Responsibility Boundaries.
Tool parameters: structure, authorization, and business constraints
Process structure verification, business semantics, authorization identity and output verification separately.
Protocol and trust boundaries in MCP integration
Distinguish between protocol interoperability, OAuth authorization, business permissions, and tool risk tips.
Idempotency and compensation when outcomes are unknown
Handle the uncertain results of external side effects with stable operation flags, parameter summaries, and result queries.
Separate discovery from execution in large tool catalogs
Examine tool catalog, dynamic discovery, recall quality, and server-side authorization.
Reconnect transport, recover messages, and reconcile business outcomes
Examine the differences between stdio, Streamable HTTP, session recovery and business idempotency.
Delivery conditions and shared deadlines for parallel results
Examine asynchronous concurrency, deadlines, cancellation propagation, and partial results.
Tool semantics and version migration in long-running tasks
Examine tool Schema version, long-term task compatibility and release rollback.
Computation, references, and consistent snapshots for large results
Examine output contracts, paging, artifact references, and deterministic calculations.
Evidence flow and stage-by-stage RAG diagnosis
Establish staged evidence and indicators to avoid all errors being solved by changing models.
Consistent authorization across RAG and derived data
Enforce authorization throughout retrieval, reranking, context assembly, caching, and source access.
Semantic completeness and evidence location in chunking
Examine document structure, search units, contextual completion, and reviewable chunking experiments.
Complementary retrieval and rank fusion in hybrid search
Examine hybrid retrieval, rank fusion, deduplication and multi-stage evaluation.
Version closure and incremental consistency during embedding migration
Investigate vector space compatibility, dual indexing, incremental changes, and deletion consistency.
Evidence support, partial answers, and calibrated abstention
Examine the sufficiency of evidence, item-by-item citation, conflict resolution and rejection calibration.
Definitions, granularity, and execution boundaries in natural-language queries
Examine the semantic layer, controlled SQL, read-only execution, and result review.
Entity disambiguation and constraint preservation in query rewriting
Examining multiple rounds of reference, entity linking, clarification, and original constraint preservation.
Memory scopes and trusted authorization context
Design memory scope, authenticity, permissions and recall order as data contracts, and vector similarity is only responsible for sorting.
Revoke memories, clean up derivatives, and prevent stale data from returning
Connect deletion to recovery and handle old state with undo versioning, provenance verification, physical cleanup, and failure testing.
Admission, provenance, and fact correction for long-term memory
Examine memory access, sources, validity periods, and error correction.
Objects, sources, and bitemporal reasoning in memory conflicts
Examine fact versions, validity times, source priority, and conflict presentation.
Hard filters, value ranking, and causal evaluation in memory retrieval
Examining memory retrieval stratification, relevance, authority, temporality, and counterfactual assessment.
Checkpoints, replay, and external side-effect boundaries
Distinguish state recovery, history replay and external side effects, design stable operation keys, execution ledger and fault injection verification.
Lease takeover and fencing stale workers
The lease is responsible for taking over, the monotonically increasing token is responsible for rejecting expired writes, and the business side effects must still be independently constrained.
One state machine for retries, deadlines, cancellation, and budgets
Incorporate total deadlines, single timeouts, error classification, budget reservations, and collaborative cancellation into the same execution strategy.
Atomic intent and duplicate delivery with the outbox pattern
Examine double-write issues, transaction outboxes, duplicate deliveries, and recovery scans.
Approval snapshots, durable waiting, and one logical resumption
Examine persistence waiting, one-time approval, expiration verification and concurrent recovery.
Structural and business compatibility during workflow upgrades
Check workflow version, persistence state schema and playback compatibility.
Committed effects and conditional compensation in sagas
Examine cross-service transactions, compensation conditions, irreversible actions, and human intervention.
Fair scheduling, resource quotas, and backpressure across tenants
Examine multi-tenant queuing, fair scheduling, backpressure and multi-dimensional quotas.
Executable acceptance checks and quality boundaries
Convert results, permissions, evidence, resource costs and recovery capabilities into decidable conditions, and then combine them into release access control.
Evaluate both outcomes and execution traces
Use result postconditions, partial ordering of necessary events, and prohibited behaviors to evaluate trajectories without forcing all tool calls to be exactly the same.
Independent and representative evaluation samples
Examine task sampling, failure coverage, data leakage and annotation quality.
Calibration and errors of semantic judges
Assess grader bias, agreement among human reviewers, scoring contracts, and adversarial cases.
Success rates and retry accounting for stochastic tasks
Examine randomness, task-level statistics, pass@k, and stability.
Failure windows and recovery invariants
Examine failure points, business invariants, deterministic tool stand-ins, and recovery acceptance.
Release and rollback of behavioral configurations
Examine behavioral versions, regression thresholds, shadow traffic and online attribution.
Task causality and resource attribution
String business tasks, model requests and tool calls into queryable links, and use quality regression to supplement operating indicators.
Approval targets and execution-time authorization
Implement permission checks at the tool execution boundary, bind approval to immutable parameters and resource versions, and reject prompt injection to expand authorization.
External data and execution-permission boundaries
Examine indirect prompt injection, data and command separation, tool permissions, and outbound control.
Isolate resources and capabilities when executing code
Examine files, networks, processes, credentials, and resource limits.
Capability and status contracts for fallback models
Examine multi-model adaptation, capability matrices, failure classification, and degradation.
Semantic and authorization equivalence for cache reuse
Examine different caching tiers, permission versions, latency and real savings.
Reconnecting clients and durable task state
Examine task and connection decoupling, event sequence number, reconnection and clear final state.
Incident containment, evidence, and causal diagnosis
Examine stop loss, version attribution, tiered metrics, privacy, and review.
A controlled task chain from evidence to publication
Connect identity, knowledge permissions, structured tools, persistence tasks, and audit releases into a recoverable system.
Match orchestration abstractions to business responsibilities
Use state complexity, recovery requirements, tool contracts, and team constraints to make trade-offs, rather than choosing based on framework popularity.
Bind code changes to acceptance evidence
Examine code base understanding, isolation execution, acceptance, regression and manual review.
Separate natural-language requests from transaction state
Examine intent understanding, business rules, authorization and traceable execution.
Bind claims to sources and verification scope
Examine research plans, source verification, evidence ledgers, experimentation and publication thresholds.
Tenant isolation and continuous authorization during execution
Examine the control plane, execution plane, data isolation, capacity, and progressive delivery.
Evidence and conditional reasoning about project capabilities
Examine the project review, personal contribution, quantitative evidence and technical choices.
Capacity, response shapes, and execution boundaries of LLM interfaces
Focus only on context budgeting, structured output, tool invocation, and bounded retries that affect Agent engineering behavior.
Token accounting, context capacity, and generation headroom
Use a concrete budget example to distinguish input capacity, generation limits, and next-turn headroom, then decide when to split, filter, or compress.
Boundaries between learned capabilities, contextual evidence, and task behavior
Distinguish between capabilities in parameters, evidence in requests, and post-training behavior, and select adaptation methods based on failure reasons.
Model-response lifecycle and completion semantics
Judge the end of transfer, model termination, output type and tool results separately to know which fragments can be previewed and which states can be delivered.
Prompt iteration, evidence support, and independent acceptance checks
Differentiate structure, provenance, policy fields, and free text support with the same task set, retaining failures and comparing old and new versions.