本地优先 Web 应用Local-first web app· 2026
PlanboxPlanbox
AI 活动机会管理器AI opportunity manager
把活动来源、待核对字段和日历导出放在一条可追溯的流程里。Keep event sources, fields needing review, and calendar export in one traceable flow.
案例摘要Case summary
目标用户Audience
经常收集 AI 活动、黑客松、训练营信息的学生、开发者和产品经理。Students, developers, and product managers who frequently collect AI events, hackathons, and bootcamp information.
我的角色Role
负责 PRD、MVP 范围、人工确认节点和核心数据流的产品决策；全部代码由 AI 编程工具完成。Owned the PRD, MVP scope, human-review checkpoints, and core data-flow decisions; all code was produced with AI coding tools.
交付结果Outcome
本地原型已覆盖收集、抽取草稿、核对、更正/撤回与 ICS 导出；尚未完成全部可信基础任务，也未作为正式日程管理器交付。The local prototype covers collection, extraction drafts, review, correction/retraction, and ICS export. Trust-foundation work is incomplete; it has not been delivered as a reliable primary calendar manager.
产品洞察Product insight
收藏不等于行动，来源可定位也不等于字段正确。先保留原文、日期精度和更正历史，再讨论自动排期与提醒。Bookmarks are not action, and a locatable source does not prove a field is correct. Preserve the source, date precision, and correction history before adding automatic scheduling and reminders.
项目简介Overview
Planbox 是本地优先的活动信息整理原型，支持保存来源、AI 抽取草稿、人工核对、字段更正与撤回，以及 ICS 导出。微信、小红书输入和原生小程序接入仍在受控验证，部分采集问题尚未修复。Planbox is a local-first event-information prototype with source retention, AI extraction drafts, human review, field corrections and retraction, and ICS export. Platform intake and native mini-program integration remain under controlled evaluation, with unresolved collection defects.
技术栈Tech stack
结果证据Evidence
2026-10-24 核对本地分支、时间导出服务、更正契约和最近审查记录；本轮未重跑模型或业务评估。On October 24, 2026, the local branch, calendar service, correction contract, and latest review were checked. No model or business evaluation was rerun.
活动与草稿已有逐字段修改、版本冲突检查、历史记录和整次撤回；证据定位只说明片段可定位，不证明语义正确。Events and drafts support field edits, revision checks, history, and whole-operation retraction. Locating evidence does not establish semantic correctness.
ICS 服务保留日期精度与时区边界；真实日历导入、通知送达和一周排期效果未验证。ICS preserves date precision and time-zone boundaries; actual calendar import, notification delivery, and week-planning outcomes remain untested.
最新合成数据审查复现了正文丢失、来源错绑等采集问题，尚未修复；T-003、状态机与原子性等任务仍未整项验收。Recent synthetic-data review reproduced lost text and incorrectly attributed sources. These defects remain unresolved, and T-003 plus state/atomicity work is not fully accepted.
原生小程序已有接入源码；真机分享、正式 HTTPS 环境和真实用户效果仍待验证。Native mini-program intake code exists; device sharing, production HTTPS, and real-user outcomes remain unvalidated.
项目截图Screenshots
关键决策Key decisions
第一版直接做微信/小红书自动爬取，会被登录态、风控和平台限制拖死。Directly building WeChat/Xiaohongshu auto-scraping in v1 would be dragged down by login states, risk control, and platform limits.
保留用户主动提交与待补正文入口；平台抓取不能保证成功，也不能在失败时丢掉用户已经提供的材料。Keep explicit user submission and a needs-content fallback. Platform fetches may fail, but must preserve material the user has already provided.
AI 可能误读时间、地点和报名链接。The AI can misread times, locations, and signup links.
抽取先作为草稿；核对来源、字段与日期精度，更正保留旧值、新值、原因与历史版本。Keep extraction as a draft, check source attribution and date precision, and preserve old/new values, reasons, and revision history for corrections.
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