v1.4.0 · 本地 MCP / Local MCP edition

DayVault v1.4.0 · 本地记录与 MCP 文件交换

中文介绍 · English introduction · 发布页 / Release

导出快照 / Export	确认草案 / Review

v1.4 真实模拟器截图，目标、七条记录与两项草案均为隔离的合成测试数据，不是私人记录或真实模型回复。

Actual v1.4 Simulator captures. The goal, seven records and two-item proposal are isolated synthetic test data, not private records or live model replies.

中文

v1.4 是 DayVault 这轮实验的收尾版。日常记录、公共成就和角色演出保留；App 不再提供需要开发者持续支付模型费用的 AI 排程与聊天。已有计划、目标、个人成就和历史陪伴记录不清空。

想继续用 AI 处理自己的记录，可以使用新加入的本地 MCP 服务。它面向 Codex、Claude Code、Gemini CLI 等支持 stdio 的工具，读取用户主动导出的单个目标快照，并生成新增事项草案。用户把草案传回 iPhone，预览并确认后才保存。文件往返由用户操作，没有实时连接、账号绑定或后台模型请求。电脑端面向 macOS / Linux，Windows 的文件权限保护尚未验证。

本版变化

• 设置增加“MCP 文件交换”，可以选择目标、预览导出范围、保存 JSON，以及检查导入草案。
• MCP 提供目标上下文、事项查询和新增事项草案三种工具，以及 dayvault_plan 提示词。模型和账号由外部工具提供，DayVault 不托管模型服务。
• 草案只允许新增有日期、无具体时刻、不重复的事项；每批 1–20 条，日期为目标时区的今天至未来 14 天。现有事项、完成状态和成就不授予外部写入权限。
• 保存前再次检查目标、期限、休息日、日期和重复项。未确认不写入；重复导入不能重复创建同一批事项。
• 中英文 README、接入指南和隐私说明按当前实现重写。版本链接保留独立列表，旧标签和说明不覆盖。

导出的快照不含笔记、聊天、回忆、Calendar 数据或隐藏成就规则。不过，目标名称和事项标题本身也可能包含私人信息。外部 AI 工具可能将它们发送给模型提供商；本地 MCP 不能保证外部模型离线，也不承担所选工具的费用。

版本号为 1.4.0，构建号 5。项目仍是源码发布，不含 App Store 安装包。收尾范围不包括排行榜、社交动态、挑战商城或新的 AI 养成系统；也不表示日后不会处理缺陷。

验证记录

本轮完整回归通过：Core 71/71、App 单元与集成测试 76/76、UI 22/22、MCP 33/33。App 和 UI 共 98 项，没有失败或跳过，覆盖草案预览、取消、确认保存及重复导入。MCP 检查包含官方 SDK 客户端连接本地子进程，以及禁止网络调用条件下的完整文件流程。

Release 配置的 iOS 模拟器构建通过，构建产物的版本已核对为 1.4.0、构建号 5。最后一轮 Core 校验修改后，另复跑 16 项 MCP App 集成测试，全部通过；这是已有测试的子集，不额外计入总数。本次没有把模拟器构建当作真机签名归档或 App Store 验收。

发布后修正了一项 CI 配置：GitHub 的 macos-15 默认选择较旧的 Xcode，导致 SwiftData 并发类型检查失败。主分支的检查现明确使用 Xcode 26.2，与项目要求一致；没有关闭并发检查或改动数据模型。v1.4.0 标签保留原提交，CI 配置修订可在主分支工作流查看。

停用 App 内 AI 时的独立基线为 Core 57/57、App 单元与集成 60/60、UI 19/19，Release 模拟器构建通过。本轮结果不与该基线相加。

本地协议测试不等于在 Codex、Claude Code、Gemini CLI 的真实账号中完成模型联调。三家的接入写法按官方文档整理，没有因此声称所有模型都能稳定生成合格草案。

已知日期限制：导入按目标时区验证和保存，但今日清单按设备时区选择范围，跨时区后仅日期事项可能出现在相邻日期。这一既有显示问题尚未修复。

真机权限、跨设备 CloudKit、购买沙盒、真实旧版升级和锁屏隐私仍属于上架前检查。Pro 继续保留，没有新增付费墙或改变售价。项目级再分发许可证仍待确认；本说明不代表授予新的许可。

MCP 接入 · 隐私 · 上架收尾清单

English

v1.4 is the closing release for this DayVault experiment. It keeps daily recording, the common achievement catalog and character animations. In-app AI scheduling and chat are disabled, so the developer does not need to fund ongoing model calls. Accepted plans, goals, personal achievements and historical companion records remain.

The new local MCP server provides an optional path for people who want to use their own AI tools. Codex, Claude Code, Gemini CLI and other stdio clients can read a user-exported snapshot for one goal and prepare a new-item proposal. The user transfers that file back to the iPhone, reviews it and confirms before saving. There is no live connection, account linking or background model request. The server targets macOS / Linux; file-permission protection on Windows has not been verified.

Changes

• Settings gains “MCP 文件交换” (MCP file exchange), with goal selection, export-range preview, JSON export and proposal import review.
• The MCP server exposes goal context, record lookup and proposal creation, plus the dayvault_plan prompt. The external client provides the model and account; DayVault runs no hosted AI service.
• Proposals contain 1–20 new, date-only, non-recurring items, dated from today through 14 days ahead in the goal's time zone. External tools cannot alter existing items, completion states or achievements.
• Confirmation rechecks the goal, deadline, rest days, dates and duplicates. There is no write before confirmation, and reimporting a batch cannot duplicate its items.
• Chinese and English introductions, setup guides and privacy notes describe the current implementation. Version links remain separate from language navigation, and older tags and release notes stay intact.

Exports exclude notes, conversations, memories, Calendar data and hidden achievement rules. Goal and item titles may still be private. An external AI client may send them to its model provider; running MCP locally does not make the model offline or cover that client's charges.

The app and widget use version 1.4.0, build 5. This is a source release, not an App Store binary. The experiment does not add a leaderboard, social feed, challenge marketplace or new AI companion economy. A closing release is not a promise never to fix bugs.

Verification

The full regression passed 71/71 Core, 76/76 App unit and integration, 22/22 UI and 33/33 MCP tests. The 98 App and UI tests had no failures or skips and cover proposal preview, cancellation, confirmed saving and repeated import. MCP checks include an official SDK client connected to a local subprocess and the full file workflow with network calls disabled.

The Release-configuration iOS Simulator build passed, and the built product was verified as version 1.4.0, build 5. After the final Core validation change, all 16 MCP App integration tests passed again. That run is a subset of the existing suite, not an additional test count. The simulator build is not a signed device archive or App Store acceptance.

A post-release CI correction pins Xcode 26.2 on the main branch. The older default Xcode on GitHub's macos-15 runner failed SwiftData concurrency type checks. The correction matches the project's existing toolchain requirement without disabling checks or changing data models. The v1.4.0 tag keeps its original commit; see the current workflow for the CI-only correction.

The separate baseline after in-app AI removal passed 57/57 Core, 60/60 App unit and integration, and 19/19 UI tests, plus a Release simulator build. Current counts are not added to that baseline.

Local protocol tests are not end-to-end model sessions in Codex, Claude Code or Gemini CLI accounts. Their configuration examples follow official documentation, without claiming that every model reliably produces valid proposals.

Known date limitation: imports are validated and saved in the goal's time zone, but Today selects its range using the device time zone. After crossing time zones, a date-only item may appear on the adjacent day. That existing display issue is not fixed.

Physical-device permissions, multi-device CloudKit, purchase sandbox checks, upgrades from actual older builds and lock-screen privacy still need App Store preparation. Pro remains, with no new paywall or pricing change. The project-wide redistribution license is awaiting confirmation; these notes do not grant a new license.

MCP setup · Privacy · Launch checklist, in Chinese

DayVault v1.3 — 个人成就文案修订 / Personal achievement copy

中文

修正已保存的 AI 个人成就仍显示生造名称的问题。卡片、详情和分享卡统一显示“首次完成”“完成 N 次”“累计记录 N 天”或“达标 N 个周期”。条件区分次数、不同日期与达标周期，并注明完成日期无需连续。

原 AI 名称和说明保留在详情折叠项中，隐藏成就解锁前继续保密。已有规则、进度和已获资格不变，无需重新生成。24 项公共成就、装备与动画未改动。

应用与小组件：1.3.0，构建号 4。19 项 App 单元测试和 1 项成就册界面测试通过。旧版本标签保留，私人 API 密钥和接口配置未上传。

这是开发原型的源码发布，不含 App Store 安装包。本轮未调用付费模型或重跑完整 Core / 服务端回归，此前记录的上游 AI 超时等限制仍在。

完整中英文介绍 · 中文 README · v1.2.0

English

Previously saved personal achievements now use clear milestone labels across cards, details and share previews. Conditions distinguish completion counts, distinct activity days and qualifying cycles. Original AI wording remains in a collapsed detail section and stays concealed until a hidden achievement unlocks.

Rules, progress and earned unlocks are preserved. Regeneration is not required. The 24 public achievements, equipment and animations are unchanged.

App and widget: 1.3.0, build 4. Verification passed: 19 app unit tests and one achievement-collection UI test. Earlier release tags remain fixed; private API credentials and endpoint settings are excluded.

This is a prototype source release, not an App Store binary. No paid model requests or full Core/server regression runs were performed for this release. Previously documented AI integration limitations remain unresolved.

English README · Changes since v1.2.0

DayVault v1.2 — 界面与成就文案更新 / UI and achievement copy

DayVault v1.2

中文版本介绍 · English introduction · 竞品与成就设计评估

中文

• 首页、编辑器、目标管理、成就册统一使用简洁的功能名称：新增事项、智能排程、公共成就、个人成就。
• 重写 24 项公共成就的中英文名称、条件与隐藏线索；六件装备说明同步调整。保留原 ID、规则、记录和已解锁资格。
• AI 个人成就设计指令更新为 1.0.2，并实际加入服务端构建产物。已接受的个人定义不重写，响应契约不变。代理需重启或重新部署才会加载新指令。
• 增加基于 App Store 与 Steam 官方资料的差异分析。领域模板、长期统计改进、隐藏信号改进仍为后续建议，不冒充已实现功能。

回归通过：Core 57 项、App 单元 15 项、中文界面 3 项、服务端 48 项。已检查资源语法、文档链接和敏感配置。没有进行付费模型联调。

应用与小组件为 1.2.0，构建号 3。v1.0.0 与 v1.1.0 标签保持不变。本发布只有源码，不是 App Store 安装包。上游 HTTP 504、真实七日调整、多设备 iCloud、真机权限与购买验收仍需继续验证。

English

This release standardizes UI terminology, rewrites the public achievement catalog and updates personal achievement-design instructions to 1.0.2. Existing IDs, rules, earned rewards and accepted personal definitions remain unchanged. New instructions require the proxy to be restarted or redeployed.

The linked comparison reviews five daily-recording, planning and habit products, with Steam achievement-writing references. Domain templates, richer history and revised hidden-signal stages are proposals, not shipped features.

Regression passed: 57 Core, 15 app unit, 3 Chinese UI and 48 server tests. No paid model evaluation was run. Version 1.2.0 uses build 3; previous tags stay fixed. This is a source release, not an App Store binary. Live AI reliability and several device/platform checks remain incomplete. Personal credentials and endpoint configuration are excluded from the repository.

DayVault 第二版 / Second release (v1.1.0)

中文

DayVault 用清单记录每天做过的事。完成记录会累积成就，解锁角色穿搭；可选的折纸 AI 搭档会根据你授权的记录回应。

这次调整了 App 文案和中英文介绍，删掉重复铺垫与空泛口号。排程页改为直接询问目标和期限，计划模板不再暗示有人数或热门排名。AI 来源、授权范围和错误原因仍会明确显示。

安装并使用 Humanizer 与 Humanizer-zh 作为编辑指南。DayVault 的短版写作规则已加入实际排程、陪伴、成就设计和改期请求，要求行动具体、认可有依据，不编造经历。它不保证绕过 AI 检测。

完整介绍与复用截图 · 中文 README · 写作规则与来源

本轮通过 47 项服务端、13 项 iOS 单元和 3 项中文 UI 测试，没有重新调用付费模型。上游 HTTP 504 和七天调整的真实联调仍未解决。私人接口与密钥没有随提交上传，仓库保持私有。

English

DayVault records daily activity in a simple list. Completed records contribute to achievements and clothing for your character. An optional origami AI companion responds using records you choose to share.

This revision edits the app copy and both introductions. It removes repeated explanations and slogans, asks about goals and deadlines directly, and labels challenges as plan templates. AI provenance, consent and failure details remain visible.

Humanizer and Humanizer-zh were installed and used as editing guides. Short DayVault-specific writing rules are bundled into all four AI operations. They require concrete tasks and evidence for praise, without inventing memories or claiming to evade AI detectors.

English README · Bilingual introduction and screenshots

47 server, 13 iOS unit and 3 Chinese UI tests passed. No paid model tests were rerun. Upstream HTTP 504 and incomplete live adjustment validation remain open.

Source versions / 源码版本

This updates the existing second-release description. The v1.1.0 tag and automatic source archives still point to the original second-release snapshot. Revised copy and instructions are in commit b5fa1ae, available on main. No tags were moved. v1.0.0 remains unchanged.

原 v1.1.0 标签及自动源码压缩包保持不变。本次修订位于上面的提交链接和 main 分支；v1.0.0 原版未修改。历史配置并未从旧提交中抹去。

DayVault v1.0.0 — Original / 原版存档

原版开发原型存档，保留日常记录、角色、折纸搭档和成就系统。Original development snapshot with daily recording, the character, origami companion and achievements. Not an App Store binary.

中文介绍 · English introduction

Historical snapshot: legacy provider configuration remains in this version and earlier Git history; no API key was found by the source-history scan. New-version configuration cleanup does not erase history. 此存档保留旧配置，私有仓库的历史未重写；脱敏改进请查看 v1.1.0。