Skip to content

docs: add Andromeda OS research and product plan - #1

Merged
oratis merged 2 commits into
mainfrom
codex/add-os-research
Aug 1, 2026
Merged

docs: add Andromeda OS research and product plan#1
oratis merged 2 commits into
mainfrom
codex/add-os-research

Conversation

@oratis

@oratis oratis commented Jul 26, 2026

Copy link
Copy Markdown
Owner

What changed

  • Added a PC/macOS operating-system landscape report covering open-source and proprietary system families.
  • Added five deep-research documents:
    • Windows gaming, Office, enterprise files, Wine/Proton, and Windows VM fallbacks;
    • PC, Intel Mac, T2 Mac, and Apple silicon drivers, firmware, certification, and migration;
    • atomic updates, rollback, isolation, AI-agent security, Codex/Claude Code patterns, and local models;
    • Wayland desktop, KDE Plasma/KWin, portals, remote desktop, installers, image building, and app stores;
    • a consolidated Adopt/Pilot/Watch/Reject open-source matrix.
  • Added an Andromeda product development plan with architecture, hardware tiers, compatibility routing, AI capability controls, recovery, backup, migration, milestones, SLOs, and the first 12-week plan.
  • Expanded the documentation indexes.

Key decisions

  • Linux stable/LTS and hardware-enable channels, with separate Asahi combinations for supported Apple silicon cohorts.
  • bootc/OCI delivery with OSTree deployments, independently abstracted behind Andromeda APIs.
  • Wayland + Xwayland + KDE Plasma 6/KWin for v1; no compositor fork.
  • Steam/Proton as a managed gaming domain, plus a native Windows escape hatch for Game Pass/UWP, unsupported anti-cheat, VR, and specialized peripherals.
  • Full Windows RDP desktop as the reliable Office fallback; RemoteApp/single-window integration remains a licensing and engineering Pilot.
  • Machine-readable Hardware Compatibility Manifest, physical hardware CI, explicit support tiers, and source-system migration agents.
  • Capability Broker, typed tools, deterministic risk levels, sandbox/microVM isolation, evidence, audit, and rollback/compensation for AI actions.

Validation

  • Cross-review and independent source/claim audit completed.
  • Heading hierarchy and all local Markdown links validated across nine documentation files.
  • External URL syntax validated; key dynamic and legal claims use official upstream/vendor sources.
  • git diff --check passed.

Impact

No runtime code is changed. This PR establishes the research baseline and executable product plan for the first architecture and prototype phase.

@oratis oratis changed the title docs: add Andromeda OS landscape research docs: add Andromeda OS research and product plan Jul 26, 2026
@oratis
oratis marked this pull request as ready for review July 26, 2026 07:16
@oratis

oratis commented Aug 1, 2026

Copy link
Copy Markdown
Owner Author

PR #1 文档评审:Andromeda OS research and product plan

总体评价

这是一组质量远超一般"调研文档"水准的研究与规划材料:五份专题研究普遍做到了引用一手资料、区分"上游已有/可产品化/不可承诺"、给出 Adopt/Pilot/Watch/Reject 的可执行决策,并且对反作弊、Office 深度功能、Apple 固件授权、字体再分发等最容易被夸大的领域保持了罕见的克制和诚实。主要问题集中在跨文档一致性上:docs/os-landscape-and-andromeda-architecture.md 作为最早的全景稿,其硬件 Tier 定义和"自研 shell/compositor"建议与后续的产品计划、桌面研究直接矛盾,未做订正或标注"已被取代",会误导读者对已定决策的理解。此外部分外链疑似不可达、12 周计划缺少资源假设,需在合入前修正。

优点

  • 决策可执行,不是项目罗列。 docs/research/desktop-platform-and-distribution.mddocs/research/hardware-drivers-and-migration.md 对每个组件给出决策、边界和引用(如 wlroots 上游已迁移至 freedesktop GitLab、"不得从旧 GitHub release 判断当前版本";bootc-image-builder 已并入统一 image-builder、"Reject(新依赖)"),这类细节说明调研是真做过的。
  • 对不可承诺项异常诚实。 docs/research/windows-gaming-office-formats.md §3.3 明确"中间件支持 Proton 不等于所有使用该中间件的游戏都支持 Proton",§9.2 列出禁止使用的宣传语;hardware-drivers-and-migration.md §12 "不把 T2、M1/M2、M3/M4 合并成 Mac supported"。这直接封死了未来营销夸大兼容性的空间。
  • 跨文档的核心量化指标基本对齐。 冷启动 ≥99.5%/200(Supported)、≥99.9%/1000(Certified),suspend/resume 100/500 次,在 product-development-plan.md §10.2、Stage 3 退出门与 hardware-drivers-and-migration.md §11 三处一致。
  • AI 安全模型技术上站得住。 reliability-update-ai-agent.md 的 L0–L3 隔离分层、R0–R4 风险分级、"read(secret) + network(send) 组合需新增数据流授权"、"MCP 自报只读不能成为安全边界"、rollback 与 compensate 的严格区分(product-development-plan.md §5.6 强制字段)都是正确且不常见的设计纪律。
  • 许可证标注准确率很高。 抽查 Proton(顶层 BSD-3 但多许可聚合)、DXVK(Zlib)、FreeType(FTL/GPL 双许可)、CUPS(Apache-2.0 + 链接例外)、WinApps(遗留代码无许可,Reject 直接复用)等均正确,且多处提醒"表格许可证不能替代逐文件 SPDX 扫描"。
  • RemoteApp/许可证风险识别到位。 windows-gaming-office-formats.md §4.2 正确指出 FreeRDP 实现客户端协议不等于 Windows Pro VM 具备受支持的 RemoteApp host 能力,并把完整 RDP 桌面定为唯一可靠基线——这个判断在同类文档中经常被搞错。

问题与建议

P1 — 硬件 Tier 定义在三份文档中互相矛盾(必须修)
docs/os-landscape-and-andromeda-architecture.md §3.2 定义:Tier 0 = QEMU/KVM 虚拟硬件、Tier 1 = 认证机型、Tier 2 = 社区验证、Tier 3 = 尽力支持、Tier X = 不支持;而 docs/product-development-plan.md §6.2 与 docs/research/hardware-drivers-and-migration.md §5 定义为:CI-0 = 虚拟平台、Tier 0 = Blocked、Tier 1 = Community、Tier 2 = Supported、Tier 3 = Certified、Tier 4 = Reference。同一个 "Tier 1" 在一份文档里是"认证机器",在另两份里是"社区验证";"Tier 3" 一边是"尽力启动"一边是"Certified"。该矛盾出现在 os-landscape 的**执行摘要(核心判断 5)**里,属于最显眼的位置。建议统一为 plan/hardware 文档的方案,并在 os-landscape 相应章节修订或加"已被 product-development-plan.md §6.2 取代"的标注。附带问题:os-landscape 的 YAML 示例用 compatibility_tier: 2(数字),hardware 文档 HCM 用 overall_tier: certified(字符串),schema 约定也应统一。

P1 — 桌面架构建议自相矛盾:自研 compositor/shell vs 复用 Plasma/KWin
os-landscape-and-andromeda-architecture.md 的"推荐架构一句话"是 "Wayland/自研 Shell",§6.1 表格"显示/媒体"行的自研重点写着"自研 compositor/shell、跨运行时门户",§12 "建议批准"第 3 条是 "Wayland + 自研 shell";而 product-development-plan.md §5.2 明确"首版采用 Wayland + Xwayland + KDE Plasma 6/KWin……也不自研 compositor",desktop-platform-and-distribution.md §1.1 结论 1 同样是"完整复用 KDE Plasma 6 / KWin,而不是自研 compositor"。读者若只读全景文档会得出与实际决策相反的结论。同样,os-landscape §2.1 把 "Windows VM + seamless window" 作为无条件的二级策略,而 plan §5.4 与 windows 研究已将 seamless window 降级为需过许可证/技术门的 Pilot。建议系统性地把 os-landscape 与后续文档对齐,或在文首声明其架构建议以 product-development-plan.md 为准。另外 plan Stage 2 交付物中的 "Andromeda shell alpha" 与 §5.2 "不自研 compositor" 的边界(shell ≠ compositor?基于 Plasma 定制到什么程度?)也应明确一句。

P2 — 若干外链疑似虚构或不可达,与 PR 描述的"验证"口径不符
PR 描述只声明"External URL syntax validated",但至少以下链接高度可疑:os-landscape-and-andromeda-architecture.md §13 的 "OpenAI Codex:Agent approvals and security → https://learn.chatgpt.com/docs/agent-approvals-security"(learn.chatgpt.com 不是 OpenAI 文档域名);reliability-update-ai-agent.md §2.3 及 §8.4 引用的多条 support.microsoft.com 链接缺少真实文章 slug/ID(如 .../windows/deployment/install-upgrade/delete-your-previous-version-of-windows.../edge/export-passwords-in-microsoft-edge),格式像是拼出来的;desktop-platform-and-distribution.md §7.5 引用的 https://systemd.io/ROOTFS_DISCOVERY/ 需确认是否存在。既然这批文档的核心价值之一是"一手资料可追溯",建议合入前做一次真实的 HTTP 可达性校验并修复。

P2 — 硬件认证数量目标在两份文档间冲突
hardware-drivers-and-migration.md Phase H1(3–9 个月)承诺 "Certified 5–10 台 PC,Supported 20–30 台";而 product-development-plan.md Stage 2(第 19–36 周)只交付 "5 款 Tier 2 Supported 候选硬件",Stage 3(第 37–60 周,约 8.5–14 个月)才到 "10–15 款 Tier 2 Supported 机型和少量 Tier 3 Certified 候选"。硬件研究文档的节奏明显激进于产品计划,二者应以其中一个为准。

P2 — 12 周执行清单与整体计划缺少任何资源/团队规模假设
product-development-plan.md §12 要求在第 5–6 周同时打通 Steam/Proton、Flatpak、KVM Windows Workspace、完整 RDP、Office 往返测试、双系统迁移扫描器和 agent dry-run,§7.2 列了约 10 个必须覆盖的专业方向,但通篇没有人数、预算或"若只有 N 人则砍掉什么"的假设。这份计划对一个大团队是紧凑的,对小团队是不可行的;作为"可执行产品计划",建议补一节资源假设与按团队规模的降级方案,否则里程碑承诺无法评估可信度。

P3 — 术语与编号类小错

  • product-development-plan.md §5.7 标题写"'无缝切换'拆为四层",实际列出 Layer 1–5(第五层"安全共存与卸载")。
  • §6.3 已明确规定 "HCM = Hardware Compatibility Manifest……capability 不作为另一种 manifest 名称",但 os-landscape-and-andromeda-architecture.md §3.2 仍使用 "Hardware Capability Manifest",违反了自己定的术语规则。
  • product-development-plan.md §3.3 "北极星任务"列出 10 项,而 Stage 0 交付物写 "20 个北极星任务"(os-landscape Phase 0 也是 20 个),数量对不上。
  • os-landscape §10 硬件成功指标写 "suspend/resume 1000 次循环成功率",与 plan §10.2 的 100/500 次发布门未对齐(实验室跑 1000 次可以,但作为"成功指标"应与发布门统一口径)。

P3 — 对话残留与悬空承诺
os-landscape-and-andromeda-architecture.md §7 写"用户后续会提供具体列表;在此先建立归类框架"——这是聊天上下文的残留物,不应出现在仓库文档中;且 §14 承诺的 windows-pain-points.md 没有出现在 product-development-plan.md §13 的 16 份后续规格清单里,该任务事实上被丢失了。建议改写措辞并把 pain-points 文档补进 §13 或明确取消。

P3 — 时效性强的上游状态声明建议加"需复核"标记
hardware-drivers-and-migration.md §1.1 "NVK 已是 Vulkan 1.4 conformant,并覆盖 Kepler 至 Ada 以及消费级 Blackwell"、desktop-platform-and-distribution.md 对 GNOME 50/Plasma 6.4/COSMIC Epoch 1(2025-12-11)的描述——这些结论随上游快速变化。research/README.md 已有"检索日期 + SBOM 锁定"的好规则,建议对这类硬件/驱动覆盖声明再统一加上"进入 Adopt 前须按当日上游状态复核"的显式标记。

结论

Request changes — 内容质量很高、方向判断可信,但 os-landscape 与产品计划之间的 Tier 定义冲突和"自研 vs 复用 Plasma/KWin"矛盾出现在执行摘要与"建议批准"清单这类会被直接引用的位置,合入前必须消解;其余为可快速修复的一致性与链接问题,修完即可合并。


Review by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant