社区事项 AI 工作台Community casework AI workspace· 2026
OneCaseOneCase
社区事项 AI 工作台An AI workspace for community casework
把居民反馈整理成可编辑事项，由人确认、跟进并留下可追溯记录。Turn resident feedback into editable cases, with human confirmation, follow-up, and a traceable record.
案例摘要Case summary
目标用户Audience
需要把居民反馈整理、分派、跟进并保留责任记录的社区工作人员和基层服务团队。Community workers and frontline service teams that need to organize, assign, follow up on, and audit resident feedback.
我的角色Role
负责社区事项场景、Intake/Review 状态、消息接入边界、风险阻断和验收；代码由 AI 编程工具完成。Owned the community-case scenario, Intake/Review states, message-ingestion boundary, risk blocking, and acceptance; code was produced with AI coding tools.
交付结果Outcome
已实现文字与图片 Intake、人工确认、事项状态流转、消息接入契约和移动端关键路径的本地原型；并发修复、全仓构建与同版本远程 CI 验证通过。Implemented the local prototype flow for text and image intake, human confirmation, case-state transitions, ingestion contracts, and the key mobile path; concurrency regression, the full-repo build, and remote CI passed at the same revision.
产品洞察Product insight
社区事项不是一次生成文本，而是一条需要被确认、推进和复盘的工作链。AI 草稿、人工确认、状态、证据和后续处理必须处在同一个可追溯闭环里。Community casework is not a one-shot text-generation task. It is a traceable chain of draft, confirmation, ownership, progress, and follow-up.
项目简介Overview
OneCase 面向社区工作人员，把文字与图片反馈整理成结构化事项；AI 负责生成可编辑草稿，人决定新建或关联事项，再通过状态流转、消息接入和处理记录完成后续跟进。OneCase helps community workers turn text and image feedback into structured cases. AI drafts an editable intake; a human decides whether to create or link a case, then follows it through status changes, message contracts, and activity history.
技术栈Tech stack
结果证据Evidence
仓库包含 Next.js web、AI、contracts、domain、db 等分层 package，并保留 Vitest、Playwright 和不变量测试入口。The repo is structured as a Next.js web app plus AI, contracts, domain, and database packages, with Vitest, Playwright, and invariant-test entry points.
README 记录文字/图片 Intake 到 Review 的演示链路，以及真实 StepFun Intake 记录。The README records an Intake-to-Review demo path for text and image input, including a real StepFun intake record.
2026-10-07 核验提交 1d2fc44 的远程 CI：单测 244/244、业务不变量 99/99、四进程并发 12/12、E2E 32/32，类型检查及全仓构建通过。Remote CI for revision 1d2fc44 verified on 2026-10-07: unit tests 244/244, service invariants 99/99, four-process concurrency regression 12/12, and E2E 32/32; typecheck and the full-repo build passed.
修复已提交并推送至 GitHub，CI 运行 37565029648 成功；正式案例仍引用 0863862 的早期证据，历史结果与当前版本分别归属。The fixes are committed and pushed to GitHub, and CI run 37565029648 succeeded. The formal case study still cites early evidence from revision 0863862; historical results and the current revision are recorded separately.
已有主流程实现和历史测试记录，不代表全部场景验收通过，也不证明真实平台集成、图片质量、模型准确率或延迟达标。Implemented main flows and historical test records do not establish that every scenario passed acceptance, or that live platform integration, image quality, model accuracy, or latency targets were verified.
项目截图Screenshots
关键决策Key decisions
AI 生成的事项草稿不能直接成为社区工作记录，否则错误会进入后续分派和跟进。An AI-drafted case cannot become an official community record without review, or errors will flow into assignment and follow-up.
把 Intake 与 Review 分开，草稿必须经过真人确认后才能进入正式事项流转。Separate Intake from Review and require human confirmation before a draft enters the formal case flow.
外部消息重试或重复投递可能造成重复事项和重复处理。Retries or duplicate external messages can create duplicate cases and repeated work.
用幂等键、契约校验和显式异常状态约束接入边界，让重复与失败都可见。Use idempotency keys, contract validation, and explicit error states so duplicate and failed ingestion remains visible.
想看完整的代码和实现细节？Want the full code and implementation details?
在 GitHub 上查看View on GitHub