汽车工程智能体工具包 Automotive engineering agent toolkit
面向自动驾驶安全、SOTIF、中国标准、场景驱动测试和工程证据链，把汽车研发中的专家角色、技能、规则、命令、工作流和知识库沉淀为可复用的开源智能体资产。 An open-source agent asset base for automated-driving safety, SOTIF, China standards, scenario-driven testing, and engineering evidence workflows, organizing expert roles, skills, rules, commands, workflows, and knowledge references for automotive development.
前身为 Automotive Claude Code Agents；本页采用更通用的项目名，同时保留上游开源项目 credit 和许可证边界。 Formerly explored as Automotive Claude Code Agents; this page uses a broader project name while preserving upstream credit and license boundaries.
汽车不是低门槛的软件行业。ADAS、功能安全、网络安全、EMC、AUTOSAR、MISRA、ASPICE、UNECE 等标准和过程要求交织在一起，每个工程判断都需要同时考虑系统、软件、硬件、验证和证据边界。 Automotive engineering is not a low-barrier software domain. ADAS, functional safety, cybersecurity, EMC, AUTOSAR, MISRA, ASPICE, UNECE, and related process requirements overlap, so engineering decisions must consider system, software, hardware, validation, and evidence boundaries together.
这也是为什么汽车研发智能体不能只做“问答”或“代码补全”。它要把专家角色、领域技能、标准规则、过程命令和知识库组织成可复核的工程工作流。 That is why automotive development agents cannot stop at Q&A or code completion. They need to organize expert roles, domain skills, standard rules, process commands, and knowledge bases into reviewable engineering workflows.
汽车工程智能体不是通用代码补全。它需要把标准、架构、安全分析、测试评价、工程过程和证据链放进同一个可复用工作流，让工程师可以在明确边界内更快完成分析、草案和复核准备。 Automotive engineering agents are not generic code completion. They need to connect standards, architecture, safety analysis, test evaluation, engineering process, and evidence chains into reusable workflows that help engineers prepare analysis, drafts, and review materials within explicit boundaries.
ISO 26262、ISO 21448、ISO 21434、AUTOSAR、MISRA、ASPICE 与中国准入要求共同约束工程判断。 ISO 26262, ISO 21448, ISO 21434, AUTOSAR, MISRA, ASPICE, and China approval requirements jointly shape engineering decisions.
需求、系统架构、软件、硬件、安全、验证、配置和文档交付之间需要稳定协同。 Requirements, system architecture, software, hardware, safety, validation, configuration, and documentation need stable coordination.
标准理解、专家经验和项目模板如果只留在聊天中，很难变成可复用能力。 Standards knowledge, expert experience, and project templates are hard to reuse if they stay only in chat history.
智能体可以整理、检索和生成草案，但安全工程结论必须由工程师复核。 Agents can organize, retrieve, and draft, but safety-engineering conclusions still require engineering review.
仓库不是单个 prompt，而是一组可安装、可路由、可复核的工程资产：技能负责知识与方法，智能体负责角色视角，规则与知识库负责边界，命令与工作流负责可重复执行。 The repository is not a single prompt. It is an installable, routable, reviewable set of engineering assets: skills carry methods, agents carry roles, rules and knowledge bases carry boundaries, and commands and workflows make repeated execution possible.
需求 / 场景 / 代码 / 标准条款 Requirements / scenarios / code / clauses
分析草案 / 校核清单 / 证据记录 Draft analysis / review checklist / evidence record
覆盖 ADAS、AUTOSAR、功能安全、SOTIF、中国标准、诊断、V2X、SDV、HPC、端到端 AD 安全等领域。 Covers ADAS, AUTOSAR, functional safety, SOTIF, China standards, diagnostics, V2X, SDV, HPC, and end-to-end AD safety.
沉淀安全工程师、SOTIF 分析师、场景工程师、中国合规工程师、ADAS 架构师、测试验证工程师等角色。 Captures roles such as safety engineer, SOTIF analyst, scenario engineer, China compliance engineer, ADAS architect, and validation engineer.
把安全、编码、过程、测试、安全标准和中国标准约束放到工程复核链路中，而不是只靠模型自由发挥。 Places safety, coding, process, testing, security, and China-standard constraints into the review chain instead of relying on free-form model output.
组织标准索引、过程说明、技术参考和示例模板，为智能体输出提供可追溯依据。 Organizes standards indexes, process notes, technical references, and examples so agent outputs have traceable support.
将常见工程任务收敛为 slash commands，减少重复检索、重复解释和临时性手工流程。 Converts common engineering tasks into slash commands to reduce repeated lookup, repeated explanation, and ad-hoc manual flows.
把需求分析、架构评估、安全分析、验证计划和证据沉淀串成端到端链路。 Connects requirements analysis, architecture assessment, safety analysis, validation planning, and evidence capture into end-to-end flows.
公开资产快照 Public asset snapshot
functional-safety / sotif / china-compliance / cybersecurity
adas / testing / validation / vehicle-systems
autosar / hpc-platform / sdv-platform / zonal-architecture
battery / powertrain-chassis / cockpit / diagnostics / v2x
project-management / process-quality / orchestration / oem
china-standards / automotive-china-*
automotive-sotif-* / automotive-scenario-driven-testing
automotive-e2e-safety-analysis / automotive-dfm-benchmarking / automotive-ml
autosar / automotive-sdv / automotive-zonal / middleware
hil-sil / diagnostics / calibration / hardware-safety / safety-analysis
功能安全分析不是孤立写安全目标。它需要把车辆功能、Item Definition、Item、System、Component、硬件部件、软件单元和需求之间的关系结构化表达，才能让架构、规则校核和证据链真正连起来。 Functional-safety analysis is not just writing safety goals. It needs structured relationships among vehicle functions, item definitions, items, systems, components, hardware parts, software units, and requirements so architecture, rule checks, and evidence chains can connect.
当这些对象关系被模型化，智能体就可以按结构理解需求、架构、安全分析和证据记录，而不是只在长文档里做关键词匹配。 Once these object relationships are modeled, agents can understand requirements, architecture, safety analysis, and evidence records structurally instead of only matching keywords inside long documents.
这些资产的价值不只是“有很多文件”，而是可以把概念阶段到系统阶段的工程任务拆成可路由、可复核、可沉淀的步骤。 The value is not simply having many files. The assets help split concept-stage and system-stage engineering tasks into routable, reviewable, and reusable steps.
需求、场景、代码、接口、标准条款、测试问题或架构备选。 Requirements, scenarios, code, interfaces, standards clauses, test issues, or architecture options.
选择安全、SOTIF、场景、架构、测试、中国标准或过程质量视角。 Route through safety, SOTIF, scenario, architecture, testing, China-standard, or process-quality perspectives.
调用技能、规则、知识库、命令模板和示例工作产品。 Use skills, rules, knowledge references, command templates, and example work products.
形成需求澄清、架构分析、安全分析、测试计划或证据清单草稿。 Draft requirement clarifications, architecture analysis, safety analysis, test plans, or evidence lists.
由工程师判断采纳、修改、驳回、补充证据或升级专家评审。 Engineers decide what to accept, modify, reject, supplement with evidence, or escalate to expert review.
把确认后的结果沉淀到仓库、文档、评审记录或后续工具链。 Capture reviewed outputs in repositories, documents, review records, or downstream toolchains.
AutoZYX-Labs Fork 的重点，是把原有汽车工程 agents 进一步扩展到自动驾驶安全、系统工程和中国场景下更难替代的专业任务。 The AutoZYX-Labs fork extends the original automotive agents toward the harder professional tasks in automated-driving safety, systems engineering, and China-specific engineering contexts.
动力、底盘、EE 架构、域控、传感器和执行器边界。Powertrain, chassis, E/E architecture, domain control, sensors, and actuator boundaries.
感知、预测、规划、控制、ODD、HMI 和降级策略。Perception, prediction, planning, control, ODD, HMI, and fallback strategy.
HARA、FMEA、FTA、STPA、SOTIF 场景和安全论证线索。HARA, FMEA, FTA, STPA, SOTIF scenarios, and safety-argument traces.
仿真、场地、道路、多支柱评价、风险场景和证据记录。Simulation, proving ground, road tests, multi-pillar evaluation, risk scenarios, and evidence records.
覆盖 GB/T 34590、GB/T 43267、CSAE SOTIF、L2/L3/泊车、ODD、多支柱评价和准入材料准备等方向。 Covers GB/T 34590, GB/T 43267, CSAE SOTIF, L2/L3/parking, ODD, multi-pillar evaluation, and approval-material preparation.
沉淀触发条件识别、危险场景构建、高速公路测试评价、审核评估和成熟度模型等任务入口。 Captures triggering-condition identification, hazardous scenario construction, highway test evaluation, audit review, and maturity-model entry points.
面向端到端感知/决策系统，组织安全分析、模型行为边界、驾驶员基础模型基准评测和类人驾驶评价。 Organizes safety analysis, model-behavior boundaries, Driver Foundation Model benchmarking, and human-like driving evaluation for end-to-end AD systems.
以 SYS、SWE、HWE、MLE、VAL、SUP、MAN 等过程域组织角色、模板、检查项和工作产品线索。 Uses SYS, SWE, HWE, MLE, VAL, SUP, and MAN process areas to organize roles, templates, checklists, and work-product traces.
连接 ODD、场景分类、仿真/场地/道路联合验证、风险场景和验证证据链。 Connects ODD, scenario taxonomy, simulation-field-road validation, risk scenarios, and validation evidence chains.
通过 SysArchPilot 把需求输入、知识资产、候选生成、规则评价和证据报告接入可复核的架构创成工作流。 Uses SysArchPilot to connect requirements, knowledge assets, candidate generation, rule evaluation, and evidence reports into a reviewable architecture-generation workflow.
automotive-claude-code-agents/
skills/ 技能库：ADAS、SOTIF、安全、中国标准、E2E AD
agents/ 智能体：安全、场景、合规、架构、测试角色
rules/ 规则库：安全、编码、过程、测试、中国标准
commands/ 工程命令：把重复任务收敛为可调用入口
workflows/ 工作流：需求、架构、安全、验证、证据链
hooks/ 质量门：安全检查、提交检查、交付前检查
knowledge-base/ 知识库：标准、过程、技术和工具参考
examples/ 示例：汽车软件、平台、测试与部署样例页面展示的是一个可以被打开、审阅和追溯的公开仓库。README、技能目录、智能体定义、规则库、知识库、示例和安装脚本共同证明：这不是一次性演示，而是一组正在沉淀的工程智能体资产。 This page points to a public repository that can be opened, reviewed, and traced. README files, skill directories, agent definitions, rules, knowledge references, examples, and installation scripts together show that this is not a one-off demo, but an evolving engineering-agent asset base.
公开文章和 LinkedIn 传播反馈可作为辅助背景；本页不把传播效果包装成客户验证或商业落地。 Public articles and LinkedIn reception may provide background context; this page does not present public attention as customer validation or commercial adoption.
高质量证据页必须守住边界。这个项目可以证明已有公开探索和方法储备，但不能被包装成认证工具、量产平台或客户交付成果。 A credible evidence page must state its boundaries. This project can show public exploration and method preparation, but it should not be presented as a certification tool, production platform, or delivered customer system.