面试题库
共 328 题,当前筛选 1 题。
课程全部30 天从前端工程师到 Agent 工程师5 天提示词工程零基础Claude 高效使用:从对话到 Claude CodeCodex 与 OpenAI Agents SDK 高效使用7 天 MCP:把工具接进任何 Agent7 天 Agent Skills:把经验做成可复用能力5 天上下文工程14 天 RAG:从检索到可信回答14 天用 Agent 搭一条 AI 短剧生产线
标签
全部#prompt-design1#architecture8#cost4#multi-agent4#agent-skills3#context3#evaluation3#llm-basics3#pipeline-design3#security3#system-design3#tool-design3
还有 125 个标签收起标签
#abstention2#agent-loop2#behavioral2#claude-md2#coding-agent2#consistency2#context-engineering2#context-window2#distributed-systems2#embeddings2#framework-design2#interview-prep2#interview-process2#mcp2#memory2#observability2#orchestration2#prompt-engineering2#provider-abstraction2#rag2#resume2#scalability2#scheduling2#skills2#sse2#streaming2#structured-output2#tool-calling2#agent-basics1#agent-design1#agent-sdk1#agentic-rag1#agents-md1#api-design1#async-task1#auth1#backoff1#bi-encoder1#build-vs-buy1#career1#chunking1#communication1#concurrency1#configuration1#consistent-hashing1#content-safety1#context-assembly1#context-management1#coreference1#cost-accounting1#cost-analysis1#cross-encoder1#data-quality1#encoding1#failure-analysis1#fairness1#few-shot1#ffmpeg1#fine-tuning1#frontend1#global-market1#golden-set1#hooks1#human-in-the-loop1#hybrid-search1#image-generation1#ingestion1#json-schema1#jwt1#langgraph1#long-context1#long-term-memory1#maintenance1#mcp-basics1#media-pipeline1#mental-model1#message-bus1#messages-api1#migration1#model-routing1#moderation1#modularity1#multi-turn1#normalisation1#openai1#operations1#ordering1#overlap1#primitives1#prioritization1#priority-queue1#proactive-messaging1#product-engineering1#project-storytelling1#prompt1#prompt-basics1#prompt-bloat1#prompt-injection1#prompt-surface1#prompt-techniques1#prompting1#protocol1#query-rewriting1#rag-basics1#rate-limiting1#redis-streams1#reliability1#responses-api1#retrieval1#routing1#sampling1#schema-validation1#scripts1#server-design1#sharding1#similarity1#skill-design1#stakeholder-communication1#star1#state-machine1#system-prompt1#test-set1#token-budget1#tts1#workflow-engine1
Claude 高效使用:从对话到 Claude Code
D1 提示词进阶与 Claude 的「性格」:system prompt、XML 标签、让模型先思考、结构化输出
system prompt 应该放什么、不该放什么?如果一条规则模型总是不遵守,你会先检查什么?What belongs in a system prompt and what doesn't? If the model keeps ignoring one rule, what do you check first?
国内高频海外高频基础#system-prompt#prompt-design分析过程 · 先想清楚再作答
- 这题考的是「职责边界」而不是「会不会写」。答成「放角色和要求」是及格线,能说出「不该放什么」以及「为什么」才有区分度。
- 先给一条判据:system prompt 是每次请求都重发的固定前提,所以只放整场对话都成立的东西——角色、边界(禁止项)、输出风格;每次都变的(时间、用户名、本轮材料)放 user 消息。
- 再说反面:把模型本来就知道的常识(「写干净的代码」)、大段 API 文档、每轮都不一样的材料塞进 system,只会稀释真正重要的规则,还会让 prompt caching 的前缀每次都变。
- 「规则总是不遵守」的排查顺序:先看 system 是不是太长导致规则被淹没(删到不能再删),再看规则是否含糊或与别的规则冲突,最后才考虑加强调;如果是「每次必须执行」的动作,应该改成程序层面的门禁而不是继续加规则。
- 可预期的追问:system 放最后行不行?可以但不推荐——模型对靠前的指令更敏感,且会破坏缓存前缀。
How to reason about it · think before answering
- The question tests boundaries, not writing skill. Naming what to exclude, and why, is what separates a strong answer.
- Give the rule: the system prompt is a fixed premise resent on every request, so it holds only what is true for the whole conversation — role, constraints as prohibitions, output style. Anything that varies per turn belongs in the user message.
- Then the anti-patterns: obvious conventions, pasted API docs, and per-turn material dilute the important rules and also invalidate the prompt-cache prefix on every call.
- Debug order for an ignored rule: check length first and prune, then check for ambiguity or conflicting rules, and only then add emphasis. If the rule is a must-run action, move it to a deterministic gate instead of adding more words.
- Likely follow-up: can system go last? Possible but unwise — earlier instructions carry more weight and a moving prefix breaks caching.
答题要点
- 放:角色、边界(写禁止项)、输出风格;整场对话都成立的固定前提
- 不放:会变的信息(时间、用户名、本轮材料)、模型本来就知道的常识、大段文档
- system 每次请求重发,越长越贵,也越容易让关键规则被淹没
- 规则不被遵守先删再改再强调;「每次必须做」的动作改成程序门禁
Key points
- Include role, prohibitions, and output style — premises that hold for the whole conversation
- Exclude volatile facts, common sense the model already has, and long pasted docs
- The system prompt is resent every request: longer means costlier and rules get buried
- For an ignored rule: prune first, disambiguate second, emphasize last; must-run actions become deterministic gates