[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-79940":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":14,"stars30d":14,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":15,"rankGlobal":9,"rankLanguage":9,"license":16,"archived":17,"fork":17,"defaultBranch":18,"hasWiki":19,"hasPages":19,"topics":20,"createdAt":9,"pushedAt":9,"updatedAt":21,"readmeContent":22,"aiSummary":23,"trendingCount":14,"starSnapshotCount":14,"syncStatus":24,"lastSyncTime":25,"discoverSource":26},79940,"prompt-atlas","huck012428-lab\u002Fprompt-atlas","huck012428-lab","Curated Prompt Card library for LLM trainers, AI PMs, and evaluation teams.",null,"Python",79,5,4,0,2.33,"Other",false,"main",true,[],"2026-06-12 02:03:55","# prompt-atlas\n\nA curated, versioned, searchable library of production-grade prompts for\nLLM trainers, AI product managers, prompt engineers, RLHF \u002F SFT data\nteams, model evaluation teams, and AI application builders.\n\n一个精选、带版本、可检索的生产级 Prompt 库，面向 LLM trainer、AI 产品\n经理、Prompt 工程师、RLHF \u002F SFT 数据团队、模型评估团队、AI 应用开发者。\n\n> 🌐 **Live site \u002F 在线站点**:\n> [huck012428-lab.github.io\u002Fprompt-atlas](https:\u002F\u002Fhuck012428-lab.github.io\u002Fprompt-atlas\u002F)\n> — searchable, sidebar navigation, copy-button on every prompt block.\n> 网页版含全站搜索、侧边栏导航、prompt 一键复制。\n\nThis is **not** a \"awesome prompts\" snippet collection. Every entry is a\n**Prompt Card**: a reusable work asset with metadata, variables,\nexamples, documented failure modes, and tuning notes.\n\n这**不是** awesome-prompts 式的素材合集。每一个条目都是一张 **Prompt\nCard**：带元数据、变量、示例、失败模式、调优笔记的可复用工作资产。\n\n## Why this exists \u002F 为什么做这个\n\nProduction prompt work has the same problems as any other engineering\ndiscipline: people rewrite the same prompts from scratch, lose track of\nwhat works on which model, and discover failure modes the third time\nthey ship them. Treating prompts as *cards* — with schema, examples, and\ndocumented failure modes — makes them reusable across teams and over\ntime.\n\n生产环境的 Prompt 工作和任何工程学科一样会踩同样的坑：每次都从零写、\n记不清哪条 prompt 在哪个模型上稳、上线第三次才发现固定的失败模式。把\nprompt 当作\"卡片\"——有 schema、有示例、有失败模式记录——它们才能在\n团队之间和时间线上被真正复用。\n\n## What's inside \u002F 库里有什么\n\nCards are organised by direction:\n\n卡片按技术方向组织：\n\n| Direction \u002F 方向 | Examples \u002F 内容举例                                            |\n|------------------|----------------------------------------------------------------|\n| **RAG**          | Retrieval scoring, multi-hop eval synthesis, query rewriting, HyDE, citation faithfulness, answer grounding\u003Cbr\u002F>检索打分、多跳评测题合成、query 改写、HyDE、citation 忠实度、答案扎根性 |\n| **Agent**        | ReAct planners with strict tool-call schemas\u003Cbr\u002F>带严格 tool-call schema 的 ReAct planner |\n| **RLHF**         | Pairwise preference labelers across HHH dimensions\u003Cbr\u002F>HHH 三维度的 pairwise 偏好标注器 |\n| **SFT**          | Instruction-set augmentation from seed examples\u003Cbr\u002F>从种子样本扩展 SFT 指令集 |\n| **Multimodal**   | VLM caption verification against actual images\u003Cbr\u002F>VLM caption 与图像内容核对 |\n| **CoT**          | Structured reasoning with rationale summaries\u003Cbr\u002F>结构化推理 + rationale 摘要 |\n| **Eval**         | LLM-as-judge rubrics for open-ended outputs\u003Cbr\u002F>开放式输出的 LLM-as-judge rubric |\n| **Code**         | Code review checklist, test generation, code explanation, refactor suggestions, code-eval judge\u003Cbr\u002F>结构化 code review、测试生成、代码解释、重构建议、代码评估 |\n\nThe complete catalog lives in [`INDEX.md`](INDEX.md) (auto-generated).\n\n完整目录见 [`INDEX.md`](INDEX.md)（自动生成）。\n\n## I want to... \u002F 我想做...\n\nMaps a goal to the card to use. New here? See\n[`docs\u002FQUICKSTART.md`](docs\u002FQUICKSTART.md) for a 5-minute walkthrough.\n\n第一次用？看 [`docs\u002FQUICKSTART.md`](docs\u002FQUICKSTART.md) — 5 分钟从零到能用一张卡。\n\n### Evaluate \u002F score AI outputs · 评估和打分\n\n| Goal · 我想做                                                          | Card · 用这张卡 |\n|------------------------------------------------------------------------|-----------------|\n| Score one AI output on factuality \u002F coherence \u002F completeness · 给单个 AI 输出按多维度打分 | [`eval\u002Fllm-judge-rubric-open-ended`](prompts\u002Feval\u002Fllm-judge-rubric-open-ended.md) |\n| Compare a model output against a gold answer · 用 gold 答案对照打分 | [`eval\u002Freference-based-judge`](prompts\u002Feval\u002Freference-based-judge.md) |\n| Decompose an output into atomic claims and fact-check each · 把答案拆成原子事实逐条核查 | [`eval\u002Fper-claim-factuality-judge`](prompts\u002Feval\u002Fper-claim-factuality-judge.md) |\n| Score one output on custom dimensions with confidence · 自定义维度打分 + 置信度 | [`eval\u002Fpointwise-quality-scorer`](prompts\u002Feval\u002Fpointwise-quality-scorer.md) |\n| Classify an AI output for safety harms · 输出安全分类（allow\u002Freview\u002Fblock） | [`eval\u002Fsafety-output-classifier`](prompts\u002Feval\u002Fsafety-output-classifier.md) |\n| Pick the best of N AI responses · 从 N 个回答里选最好的 | [`rlhf\u002Fbest-of-n-selector`](prompts\u002Frlhf\u002Fbest-of-n-selector.md) |\n| Label A vs B preference (HHH) · 给 A\u002FB 两个回答打偏好标签 | [`rlhf\u002Fpairwise-preference-labeler`](prompts\u002Frlhf\u002Fpairwise-preference-labeler.md) |\n| Pairwise judge with position-bias detection (two-call protocol) · 带位置偏置检测的 pairwise judge（双向调用） | [`eval\u002Fpairwise-judge-with-position-bias-probe`](prompts\u002Feval\u002Fpairwise-judge-with-position-bias-probe.md) |\n| Judge a multi-turn dialogue (per-turn + conversation-level) · 多轮对话评估 | [`eval\u002Fmulti-turn-dialogue-judge`](prompts\u002Feval\u002Fmulti-turn-dialogue-judge.md) |\n| Generate a domain-specific rubric with level anchors · 给具体任务自动生成定制化评分 rubric | [`eval\u002Frubric-generator`](prompts\u002Feval\u002Frubric-generator.md) |\n| Compare baseline vs candidate outputs and detect regressions · 检测候选版本是否退步 | [`eval\u002Fregression-detector`](prompts\u002Feval\u002Fregression-detector.md) |\n| Diagnose LLM judge biases (length \u002F position \u002F format) · 诊断 LLM judge 自身偏见 | [`eval\u002Fjudge-bias-probe`](prompts\u002Feval\u002Fjudge-bias-probe.md) |\n| Check confidence calibration (predicted vs actual accuracy) · 检查置信度是否校准 | [`eval\u002Fcalibration-checker`](prompts\u002Feval\u002Fcalibration-checker.md) |\n| Bootstrap a small human eval study (rubric + calibration + analysis) · 设计小规模 human eval 研究 | [`eval\u002Fhuman-eval-bootstrap`](prompts\u002Feval\u002Fhuman-eval-bootstrap.md) |\n| Build a multi-benchmark leaderboard with weighting · 多 benchmark 加权 leaderboard | [`eval\u002Fleaderboard-builder`](prompts\u002Feval\u002Fleaderboard-builder.md) |\n| Diagnose refusal calibration (over \u002F under \u002F correct) · 诊断模型拒绝是否校准 | [`rlhf\u002Frefusal-calibration-probe`](prompts\u002Frlhf\u002Frefusal-calibration-probe.md) |\n| Generate iterative DPO pairs targeting a specific principle · 按原则生成 DPO 偏好对 | [`rlhf\u002Fiterative-dpo-pair-generator`](prompts\u002Frlhf\u002Fiterative-dpo-pair-generator.md) |\n| Score whether a response matches a defined persona \u002F brand voice · 评估回答是否符合人设 | [`rlhf\u002Fpersona-consistency-judge`](prompts\u002Frlhf\u002Fpersona-consistency-judge.md) |\n| Detect over-cautious vs unsafe-helpful (HHH tradeoff scoring) · 诊断 helpful 和 harmless 之间的失衡 | [`rlhf\u002Fhelpfulness-vs-harmlessness-tradeoff`](prompts\u002Frlhf\u002Fhelpfulness-vs-harmlessness-tradeoff.md) |\n| Pairwise preference for long-form (long input + long output) · 长输入长输出的 pairwise 偏好 | [`rlhf\u002Flong-context-preference-labeler`](prompts\u002Frlhf\u002Flong-context-preference-labeler.md) |\n| Analyze SFT dataset coverage by topic \u002F skill · 分析 SFT 数据集覆盖度，找 gap | [`sft\u002Fdata-coverage-analyzer`](prompts\u002Fsft\u002Fdata-coverage-analyzer.md) |\n| Classify instruction difficulty for a target model class · 按目标模型类别给指令打难度 | [`sft\u002Finstruction-difficulty-classifier`](prompts\u002Fsft\u002Finstruction-difficulty-classifier.md) |\n| Generate response in a defined persona with strictness control · 按人设生成回答（带严格度控制） | [`sft\u002Fpersona-controlled-response`](prompts\u002Fsft\u002Fpersona-controlled-response.md) |\n| Rewrite text in a target style (formal \u002F casual \u002F specific voice) · 文本改写为目标风格 | [`sft\u002Fstyle-transfer`](prompts\u002Fsft\u002Fstyle-transfer.md) |\n\n### RAG · 检索增强\n\n| Goal · 我想做                                                          | Card · 用这张卡 |\n|------------------------------------------------------------------------|-----------------|\n| Score whether a retrieved passage is relevant to a query · 评估 passage 与 query 的相关性 | [`rag\u002Fretrieval-relevance-evaluator`](prompts\u002Frag\u002Fretrieval-relevance-evaluator.md) |\n| Build multi-hop QA eval questions · 合成多跳评测题 | [`rag\u002Fmultihop-eval-synthesizer`](prompts\u002Frag\u002Fmultihop-eval-synthesizer.md) |\n| Decompose \u002F rewrite a query for retrieval · query 改写或拆解 | [`rag\u002Fquery-rewriting-decomposition`](prompts\u002Frag\u002Fquery-rewriting-decomposition.md) |\n| Generate hypothetical answer for HyDE retrieval · HyDE 假答生成 | [`rag\u002Fhyde-hypothetical-answer-generator`](prompts\u002Frag\u002Fhyde-hypothetical-answer-generator.md) |\n| Audit whether a citation actually supports a claim · 审计 citation 是否真的支持 claim | [`rag\u002Fcitation-faithfulness-scorer`](prompts\u002Frag\u002Fcitation-faithfulness-scorer.md) |\n| Detect hallucinations in a RAG answer · 检测 RAG 答案的幻觉 | [`rag\u002Fanswer-grounding-checker`](prompts\u002Frag\u002Fanswer-grounding-checker.md) |\n| Summarize a long document chunk for retrieval indexing · 给长文档块产 search-friendly summary | [`rag\u002Fchunk-summarizer-for-retrieval`](prompts\u002Frag\u002Fchunk-summarizer-for-retrieval.md) |\n| Compress retrieved passages into a smaller question-tailored context · 把检索结果压缩成针对问题的小上下文 | [`rag\u002Fcontext-compression`](prompts\u002Frag\u002Fcontext-compression.md) |\n| Resolve a chat follow-up into a standalone retrieval query · 多轮 RAG 的代词消解器 | [`rag\u002Fconversational-query-resolver`](prompts\u002Frag\u002Fconversational-query-resolver.md) |\n| Synthesize an answer from multiple sources, surfacing conflicts · 多源综合答案 + 冲突识别 | [`rag\u002Fmulti-source-aggregator`](prompts\u002Frag\u002Fmulti-source-aggregator.md) |\n| Build structured output (table \u002F list \u002F record) from RAG sources · RAG 结构化输出（表\u002F列表\u002F字段记录） | [`rag\u002Fstructured-rag-output-builder`](prompts\u002Frag\u002Fstructured-rag-output-builder.md) |\n| Fuse multiple sub-query retrieval results into one ranked set · 多子查询检索结果融合 | [`rag\u002Fquery-fusion`](prompts\u002Frag\u002Fquery-fusion.md) |\n| Resolve time-relative phrases into concrete time bounds · 时间相对短语解析为具体时间范围 | [`rag\u002Ftime-aware-retrieval-rewriter`](prompts\u002Frag\u002Ftime-aware-retrieval-rewriter.md) |\n\n### Build \u002F debug an agent · 搭建和调试 Agent\n\n| Goal · 我想做                                                          | Card · 用这张卡 |\n|------------------------------------------------------------------------|-----------------|\n| Run a ReAct loop with strict tool calls · 跑 ReAct loop，严格 tool call | [`agent\u002Freact-planner-with-tool-schema`](prompts\u002Fagent\u002Freact-planner-with-tool-schema.md) |\n| Produce a complete plan upfront · 一次性给出完整计划 | [`agent\u002Fplan-and-execute-planner`](prompts\u002Fagent\u002Fplan-and-execute-planner.md) |\n| Fix a malformed tool call from a validation error · 修复格式错误的 tool call | [`agent\u002Ftool-call-repair`](prompts\u002Fagent\u002Ftool-call-repair.md) |\n| Reflect on whether the trajectory is on track · 反思 agent 是否在正轨 | [`agent\u002Fself-critique-reflection`](prompts\u002Fagent\u002Fself-critique-reflection.md) |\n| Compress a long agent trajectory into memory · 把长 trajectory 压缩成 memory | [`agent\u002Flong-context-memory-summarizer`](prompts\u002Fagent\u002Flong-context-memory-summarizer.md) |\n| Split a complex task across multiple specialized workers · 把复杂任务派给多个专精 agent | [`agent\u002Fsub-task-delegator`](prompts\u002Fagent\u002Fsub-task-delegator.md) |\n| Decide whether a goal needs clarification, ask one good question · 判断是否要问澄清问题，问一个好问题 | [`agent\u002Fclarification-asker`](prompts\u002Fagent\u002Fclarification-asker.md) |\n| Convert OpenAPI \u002F Swagger spec into agent tool catalog · OpenAPI 自动转 tool catalog | [`agent\u002Fapi-spec-to-tool-catalog`](prompts\u002Fagent\u002Fapi-spec-to-tool-catalog.md) |\n| Decide retry \u002F abort \u002F escalate on operation failure · 操作失败时决定重试\u002F放弃\u002F升级 | [`agent\u002Ferror-recovery-strategy`](prompts\u002Fagent\u002Ferror-recovery-strategy.md) |\n| Plan agent execution within token \u002F dollar budget · 在预算约束下规划 agent 执行 | [`agent\u002Fbudget-aware-planner`](prompts\u002Fagent\u002Fbudget-aware-planner.md) |\n| Compress verbose tool output before adding to context · 把 tool 输出压缩后再进 context | [`agent\u002Ftool-output-summarizer`](prompts\u002Fagent\u002Ftool-output-summarizer.md) |\n| Reconcile conflicting outputs from multiple sub-agents · 多 agent 冲突调解 | [`agent\u002Fmulti-agent-conflict-resolver`](prompts\u002Fagent\u002Fmulti-agent-conflict-resolver.md) |\n| Translate raw API response into a user-readable answer · API 响应翻译给用户 | [`agent\u002Fapi-result-translator`](prompts\u002Fagent\u002Fapi-result-translator.md) |\n\n### Generate \u002F filter training data · 训练数据生成与过滤\n\n| Goal · 我想做                                                          | Card · 用这张卡 |\n|------------------------------------------------------------------------|-----------------|\n| Rewrite ONE instruction into N variants · 把 1 条指令改写成 N 个变体 | [`sft\u002Finstruction-variant-expander`](prompts\u002Fsft\u002Finstruction-variant-expander.md) |\n| Generate NEW instructions from seed examples · 从种子生成新指令 | [`sft\u002Fself-instruct-from-seed`](prompts\u002Fsft\u002Fself-instruct-from-seed.md) |\n| Generate a high-quality response for an instruction · 给指令生成回答 | [`sft\u002Fresponse-generator`](prompts\u002Fsft\u002Fresponse-generator.md) |\n| Filter SFT pairs by quality (keep \u002F review \u002F drop) · 按质量过滤 SFT 数据 | [`sft\u002Fdata-quality-filter`](prompts\u002Fsft\u002Fdata-quality-filter.md) |\n| Produce scalar reward for one response · 给单回答打 reward 分 | [`rlhf\u002Fpointwise-reward-scorer`](prompts\u002Frlhf\u002Fpointwise-reward-scorer.md) |\n| Critique a response against a constitution and revise · 按 constitution 批评 + 重写 | [`rlhf\u002Fconstitutional-critique-revise`](prompts\u002Frlhf\u002Fconstitutional-critique-revise.md) |\n| Generate adversarial probes for safety evaluation (defensive) · 生成防御性安全评估探针 | [`rlhf\u002Fred-team-prompt-generator`](prompts\u002Frlhf\u002Fred-team-prompt-generator.md) |\n| Generate multi-turn conversation SFT data · 生成多轮对话 SFT 数据 | [`sft\u002Fconversation-sft-pair-generator`](prompts\u002Fsft\u002Fconversation-sft-pair-generator.md) |\n| Pick best K few-shot demonstrations from a candidate pool · 从样本池为目标 query 选最好的 K 个示例 | [`sft\u002Ffew-shot-example-selector`](prompts\u002Fsft\u002Ffew-shot-example-selector.md) |\n| Detect reward hacking patterns in RLHF responses · 检测 RLHF 训练后 reward gaming 失败模式 | [`rlhf\u002Freward-hacking-detector`](prompts\u002Frlhf\u002Freward-hacking-detector.md) |\n| Audit whether a preference label's rationale justifies the pick · 审计偏好标签的理由是否站得住 | [`rlhf\u002Fpreference-rationalization-judge`](prompts\u002Frlhf\u002Fpreference-rationalization-judge.md) |\n| Generate code-specific SFT pairs · 生成 code SFT 训练对 | [`sft\u002Fcode-sft-pair-generator`](prompts\u002Fsft\u002Fcode-sft-pair-generator.md) |\n| Find semantic near-duplicates in instruction set · 找语义相似指令做去重 | [`sft\u002Finstruction-deduplicator`](prompts\u002Fsft\u002Finstruction-deduplicator.md) |\n\n### Work with images · 处理图像\n\n| Goal · 我想做                                                          | Card · 用这张卡 |\n|------------------------------------------------------------------------|-----------------|\n| Generate a structured caption for an image · 给图片生成结构化 caption | [`multimodal\u002Fstructured-caption-generator`](prompts\u002Fmultimodal\u002Fstructured-caption-generator.md) |\n| Verify a caption against the actual image · 核对 caption 与图像 | [`multimodal\u002Fvlm-image-description-verifier`](prompts\u002Fmultimodal\u002Fvlm-image-description-verifier.md) |\n| Answer a question about an image · 视觉问答 + grounding + 置信度 | [`multimodal\u002Fvqa-with-confidence`](prompts\u002Fmultimodal\u002Fvqa-with-confidence.md) |\n| Extract typed fields from a document image · 从文档图片抽取结构化字段 | [`multimodal\u002Focr-structured-extraction`](prompts\u002Fmultimodal\u002Focr-structured-extraction.md) |\n| Extract data from a chart \u002F plot \u002F table image · 从图表或表格图片抽数据 | [`multimodal\u002Fchart-table-extractor`](prompts\u002Fmultimodal\u002Fchart-table-extractor.md) |\n| Analyze a document page's layout (title \u002F body \u002F tables \u002F figures) · 分析文档页面版式结构 | [`multimodal\u002Fdocument-layout-analyzer`](prompts\u002Fmultimodal\u002Fdocument-layout-analyzer.md) |\n| Extract graph structure from a diagram \u002F flowchart \u002F architecture · 流程图\u002F架构图转结构化数据 | [`multimodal\u002Fdiagram-to-structured-data`](prompts\u002Fmultimodal\u002Fdiagram-to-structured-data.md) |\n| Convert a UI screenshot into a component spec · UI 截图转组件树 spec | [`multimodal\u002Fscreenshot-to-spec`](prompts\u002Fmultimodal\u002Fscreenshot-to-spec.md) |\n| Classify image into custom user-defined categories · 自定义类别图像分类 | [`multimodal\u002Fimage-classification`](prompts\u002Fmultimodal\u002Fimage-classification.md) |\n| Transcribe handwriting with per-word confidence · 手写文字转录 + 字级置信度 | [`multimodal\u002Fhandwriting-transcriber`](prompts\u002Fmultimodal\u002Fhandwriting-transcriber.md) |\n| Reverse-engineer edit instruction from before\u002Fafter pair · 前后图反推编辑指令 | [`multimodal\u002Fimage-edit-instruction-generator`](prompts\u002Fmultimodal\u002Fimage-edit-instruction-generator.md) |\n| Compare two images and explain similarities \u002F differences · 双图对比解释 | [`multimodal\u002Fimage-comparison-explainer`](prompts\u002Fmultimodal\u002Fimage-comparison-explainer.md) |\n\n### Improve reasoning quality · 提升推理质量\n\n| Goal · 我想做                                                          | Card · 用这张卡 |\n|------------------------------------------------------------------------|-----------------|\n| Single-pass structured reasoning with rationale · 单次结构化推理 + rationale | [`cot\u002Fstructured-reasoning-with-rationale-summary`](prompts\u002Fcot\u002Fstructured-reasoning-with-rationale-summary.md) |\n| Decompose a complex problem into easier sub-problems · 把复杂问题拆成更简单的子问题 | [`cot\u002Fleast-to-most-decomposition`](prompts\u002Fcot\u002Fleast-to-most-decomposition.md) |\n| Aggregate N sampled paths into a consensus answer · 把 N 条采样路径聚合成共识答案 | [`cot\u002Fself-consistency-aggregator`](prompts\u002Fcot\u002Fself-consistency-aggregator.md) |\n| Draft + verify before committing to a final answer · 先 draft 再 verify 再交答案 | [`cot\u002Fverify-then-finalize`](prompts\u002Fcot\u002Fverify-then-finalize.md) |\n| Explore multiple branches in parallel and prune (tree-of-thoughts) · 多分支并行探索 + 剪枝 | [`cot\u002Ftree-of-thoughts`](prompts\u002Fcot\u002Ftree-of-thoughts.md) |\n| Abstract the question into a principle first, then apply (step-back) · 先抽象到原理再代入具体题 | [`cot\u002Fstep-back-prompting`](prompts\u002Fcot\u002Fstep-back-prompting.md) |\n| Critique and revise a candidate plan before execution · 执行前对推理 plan critique + 修订 | [`cot\u002Fplan-critique-and-revise`](prompts\u002Fcot\u002Fplan-critique-and-revise.md) |\n| Reasoning with explicit per-step uncertainty · 明示每步不确定度的推理 | [`cot\u002Funcertainty-quantification`](prompts\u002Fcot\u002Funcertainty-quantification.md) |\n| Citation-grounded reasoning (every claim must cite source) · 每条事实必须引用 source 的推理 | [`cot\u002Fcitation-grounded-reasoning`](prompts\u002Fcot\u002Fcitation-grounded-reasoning.md) |\n| Contrast against intentionally-wrong reasoning · 对照错误推理路径的反向自洽 | [`cot\u002Fcontrastive-self-consistency`](prompts\u002Fcot\u002Fcontrastive-self-consistency.md) |\n| Process external criticism (accept \u002F correct \u002F reject) · 处理外部批评的 self-correction 协议 | [`cot\u002Fself-correction-protocol`](prompts\u002Fcot\u002Fself-correction-protocol.md) |\n| Generate a meta-prompt for a class of tasks · 给一类任务生成可复用的 meta-prompt | [`cot\u002Fmeta-prompt-generator`](prompts\u002Fcot\u002Fmeta-prompt-generator.md) |\n\n### Work with code · 处理代码\n\n| Goal · 我想做                                                          | Card · 用这张卡 |\n|------------------------------------------------------------------------|-----------------|\n| Structured code review with per-dimension findings · 按维度做结构化 code review | [`code\u002Fcode-review-checklist`](prompts\u002Fcode\u002Fcode-review-checklist.md) |\n| Generate test cases for a function · 给函数生成测试用例 | [`code\u002Ftest-case-generator`](prompts\u002Fcode\u002Ftest-case-generator.md) |\n| Explain code at a specific audience level · 按受众层级解释代码 | [`code\u002Fcode-explanation-generator`](prompts\u002Fcode\u002Fcode-explanation-generator.md) |\n| Judge whether candidate code fulfills a task · 评估候选代码是否完成任务 | [`code\u002Fcode-eval-judge`](prompts\u002Fcode\u002Fcode-eval-judge.md) |\n| Suggest concrete refactors with rationale · 提结构化重构建议 | [`code\u002Frefactor-suggestion`](prompts\u002Fcode\u002Frefactor-suggestion.md) |\n| Translate code from one language to another · 跨语言代码翻译（含 idiom 控制） | [`code\u002Fcode-translation`](prompts\u002Fcode\u002Fcode-translation.md) |\n| Focused security review with CWE-style findings · 按 threat model 做代码安全评审 | [`code\u002Fsecurity-review`](prompts\u002Fcode\u002Fsecurity-review.md) |\n| Summarize git diff into structured PR description · git diff 转结构化 PR description | [`code\u002Fcode-summary-for-pr`](prompts\u002Fcode\u002Fcode-summary-for-pr.md) |\n| Plan major version migration grounded in actual code · 大版本迁移阶段化规划 | [`code\u002Fmigration-plan-generator`](prompts\u002Fcode\u002Fmigration-plan-generator.md) |\n| Analyze impact of changing a function \u002F API signature · 评估函数 \u002F API 签名改动的影响范围 | [`code\u002Fdependency-impact-analyzer`](prompts\u002Fcode\u002Fdependency-impact-analyzer.md) |\n| Explain a stack trace \u002F error to a target audience · 按受众解释错误信息 | [`code\u002Ferror-message-explainer`](prompts\u002Fcode\u002Ferror-message-explainer.md) |\n| Generate a commit message from a git diff · 从 diff 生成 commit message | [`code\u002Fcommit-message-generator`](prompts\u002Fcode\u002Fcommit-message-generator.md) |\n| Review API design (REST \u002F GraphQL \u002F gRPC) for ergonomics · API 设计评审 | [`code\u002Fapi-design-reviewer`](prompts\u002Fcode\u002Fapi-design-reviewer.md) |\n\n### As a GitHub repository \u002F 作为 GitHub 仓库\n\n1. Browse [`INDEX.md`](INDEX.md) or `prompts\u002F\u003Cdirection>\u002F`.\n2. Open the card you want; copy the **Prompt** section.\n3. Read the **Failure Modes** and **Tuning Notes** sections — that is\n   where the experience lives.\n4. Substitute `{{variable}}` placeholders with your inputs.\n\n中文流程：\n\n1. 浏览 [`INDEX.md`](INDEX.md) 或 `prompts\u002F\u003C方向>\u002F` 目录。\n2. 打开目标卡片，复制 `## Prompt` 段落。\n3. **务必读** `## Failure Modes` 和 `## Tuning Notes` 两段——那是真正的\n   经验所在。\n4. 用你自己的输入替换 `{{variable}}` 占位符。\n\n### As a Claude Code skill \u002F 作为 Claude Code Skill\n\nInstall this repository as a skill so Claude Code can route user intents\nto the right card directly:\n\n把本仓库当作 skill 安装，Claude Code 就能根据用户描述自动定位到对应\n卡片：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fhuck012428-lab\u002Fprompt-atlas ~\u002F.claude\u002Fskills\u002Fprompt-atlas\n```\n\nThen in Claude Code:\n\n之后在 Claude Code 中：\n\n```\nYou: I need a prompt to score whether a retrieved passage is relevant.\nClaude: [reads SKILL.md routing tree, picks rag\u002Fretrieval-relevance-evaluator,\n         and adapts it to your inputs]\n```\n\n```\n你: 帮我写个判断 retrieved passage 相关性的 prompt。\nClaude:[读取 SKILL.md 的路由树，选中 rag\u002Fretrieval-relevance-evaluator，\n         按你的输入做适配]\n```\n\nThe skill entry is [`SKILL.md`](SKILL.md).\n\nSkill 入口在 [`SKILL.md`](SKILL.md)。\n\n## Anatomy of a Prompt Card \u002F 一张卡片的结构\n\n```\nprompts\u002Frag\u002Fretrieval-relevance-evaluator.md\n├── frontmatter \u002F 元信息块\n│   ├── id, title, version, status         (identity \u002F 身份)\n│   ├── direction, tags, audience, models  (discovery \u002F 发现)\n│   ├── language, input\u002Foutput_schema      (integration \u002F 集成)\n│   └── variables                          (slots \u002F 变量槽)\n└── body \u002F 正文\n    ├── ## Purpose         适用场景与目标\n    ├── ## Prompt          带 {{variable}} 占位符的 prompt 主体\n    ├── ## Example         具体的输入 → 期望输出\n    ├── ## Failure Modes   常见失败模式与检测方法\n    ├── ## Tuning Notes    模型差异、温度、相邻用法的调优笔记\n    └── ## Changelog       版本历史\n```\n\nFull schema and controlled vocabulary: [`docs\u002FSCHEMA.md`](docs\u002FSCHEMA.md).\n\n完整 schema 与受控词汇表：[`docs\u002FSCHEMA.md`](docs\u002FSCHEMA.md)。\n\n## Safety \u002F 安全立场\n\nThis repository does **not** accept jailbreaks, safety-bypass prompts,\nhidden chain-of-thought extraction techniques, harm-enabling content, or\nproprietary leaks. See [`docs\u002FSAFETY.md`](docs\u002FSAFETY.md). Defensive and\nevaluation-oriented prompts (red-team rubrics, harmlessness labelers,\nfactuality judges) are explicitly welcome.\n\n本仓库**拒收** jailbreak、绕过安全的 prompt、套取闭源模型隐藏推理链\n的 prompt、有害内容生成 prompt、私有\u002F泄露 prompt。详见\n[`docs\u002FSAFETY.md`](docs\u002FSAFETY.md)。**明确欢迎**评估类、防御类 prompt\n——红队评分、有害性标注、事实性判官等。\n\n## Contributing \u002F 贡献流程\n\nSee [`CONTRIBUTING.md`](.github\u002FCONTRIBUTING.md). Short version:\n\n详见 [`CONTRIBUTING.md`](.github\u002FCONTRIBUTING.md)。简要流程：\n\n1. Copy [`templates\u002Fprompt-card.md`](templates\u002Fprompt-card.md) into\n   `prompts\u002F\u003Cdirection>\u002F\u003Cyour-slug>.md`.\n   \u003Cbr\u002F>复制 [`templates\u002Fprompt-card.md`](templates\u002Fprompt-card.md) 到\n   `prompts\u002F\u003C方向>\u002F\u003C你的-slug>.md`。\n2. Run `python scripts\u002Fvalidate.py` until it returns `OK`.\n   \u003Cbr\u002F>跑 `python scripts\u002Fvalidate.py` 直到输出 `OK`。\n3. Run `python scripts\u002Fbuild_index.py` to refresh `INDEX.md`.\n   \u003Cbr\u002F>跑 `python scripts\u002Fbuild_index.py` 刷新 `INDEX.md`。\n4. Open a PR using the prompt-card issue template.\n   \u003Cbr\u002F>用 prompt-card issue 模板开 PR。\n\nCI runs the same validation; PRs that don't pass won't be merged.\n\nCI 跑同一套校验；不通过的 PR 不会被合入。\n\n## License \u002F 许可证\n\nDual-licensed. See [`LICENSE`](LICENSE).\n\n双许可证。详见 [`LICENSE`](LICENSE)。\n\n- **Code** (`scripts\u002F`, CI configs): MIT\n- **Prompt content** (`prompts\u002F`, `templates\u002F`, `docs\u002F`): CC-BY-4.0\n\n- **代码**（`scripts\u002F`、CI 配置）：MIT\n- **Prompt 内容**（`prompts\u002F`、`templates\u002F`、`docs\u002F`）：CC-BY-4.0\n\nEach Prompt Card carries `license: CC-BY-4.0` in its frontmatter for\nclarity.\n\n每张 Prompt Card 的 frontmatter 中都标注 `license: CC-BY-4.0`，避免混淆。\n\n## Status \u002F 当前状态\n\n**v0.1.0** — first public release with 32 Prompt Cards. Library has\nsince grown to **100 Prompt Cards across all 7 directions** (post-v0.1\nadditions tracked in [`CHANGELOG.md`](docs\u002FCHANGELOG.md)).\nSee [`ROADMAP.md`](docs\u002FROADMAP.md) for what's planned next. Pull\nrequests welcome.\n\n**v0.1.0** —— 首个公开版本，32 张 Prompt Card。后续已扩到\n**100 张，覆盖 8 个方向**（v0.1 之后的新卡见\n[`CHANGELOG.md`](docs\u002FCHANGELOG.md)）。后续计划见\n[`ROADMAP.md`](docs\u002FROADMAP.md)，欢迎 PR。\n","prompt-atlas 是一个精选、带版本、可检索的生产级 Prompt 库，面向 LLM 训练师、AI 产品经理、Prompt 工程师、RLHF\u002FSFT 数据团队、模型评估团队及 AI 应用开发者。项目核心功能包括提供结构化的 Prompt Card，每张卡片都包含元数据、变量、示例、失败模式和调优笔记，确保 Prompt 的复用性和可追溯性。技术特点上，支持全站搜索、侧边栏导航以及一键复制 Prompt 功能。适用于需要高效管理和复用高质量 Prompt 的场景，如构建和优化大型语言模型、开发多模态应用或进行模型评估等。",2,"2026-06-11 03:58:36","CREATED_QUERY"]