v0.6.0 — Persona skeleton goes Markdown (breaking)

⚠️ Breaking change. The persona skeleton is now Markdown. agent-template.xml / agent-template.zh-CN.xml are gone, replaced by agent-template.md / agent-template.zh-CN.md.

⚠️ 破坏性变更。 人格骨架已改为 Markdown：agent-template.xml / .zh-CN.xml 已删除，由 agent-template.md / .zh-CN.md 取代。

Who this breaks / 谁会受影响

Anything that references the skeleton by filename — we were our own first casualty and had to fix six coupling points in the installer. Personas you have already filled in do not need to be rewritten: the migration is field-by-field equivalent, and existing agent homes keep working untouched.

任何按文件名引用骨架的东西——我们自己就是第一个被打断的下游，installer 里改了 6 处耦合点。已经填好的存量人格不必重填：本次迁移逐字段语义等价，现有员工目录原样继续用。

Why / 为什么

The Nuwa kernel itself moved from XML to Markdown back at v10; agent-template.xml was simply the piece that never caught up. And an AGENTS.md is loaded by the host as natural-language system prompt text — nothing parses it as XML, so the tags bought no structural guarantee and cost tokens on every load. Markdown is also what the agents.md ecosystem actually uses.

女娲内核自己在 v10 就从 XML 迁到了 Markdown，agent-template.xml 只是没跟上的那块遗留。而 AGENTS.md 是被宿主当自然语言系统提示整体加载的——没有任何环节把它当 XML 解析，所以标签换不来结构保证，只是每次加载都在烧 token。Markdown 也是 agents.md 生态实际通行的格式。

The honest trade-off on the L0 contract / L0 契约上的诚实取舍

XML had </l0_contract> as an explicit terminator; Markdown scopes a section by convention ("until the next ## "). That guarantee is genuinely one notch weaker, and we are not papering over it.

In exchange the L0 section now carries an explicit fields: enumeration line, which XML never had — with XML you had to read all five child elements to learn which five exist. Measured on a "first 70 lines only" reading window: Markdown surfaces 5/5 of the contract fields, XML surfaced 4/5.

XML 里有 </l0_contract> 作为显式终止符，Markdown 的作用域靠"到下一个 ## 为止"的位置约定——这条保证确实弱了一档，我们不粉饰。作为交换，L0 分节新增了一行 fields: 枚举（XML 从来没有这个信息：得读完五个子元素才知道有哪五个）。在"只读前 70 行"的窗口下实测：Markdown 出 5/5，XML 出 4/5。

Machine anchors survive the tags / 标签没了，机器锚还在

Every module heading carries `module: <original XML tag>` (12/12 consistent across both languages), so identity / mandate / deliberation / activation / workflow / output remain greppable. The old </agent> boundary is not merely deleted — an explicit END-OF-SKELETON sentinel marks where the skeleton ends and deployment-appended content begins, so tooling still has a decidable edge.

每个模块标题都带 `module: <原 XML 标签名>`（中英 12/12 一致），标签消失后 identity/mandate/deliberation/activation/workflow/output 仍可 grep。原来的 </agent> 边界不是删掉了事——新增 END-OF-SKELETON 哨兵注释标出"骨架本体到此为止、以下是部署工具链追加的"，给工具链留了可判定的边界。

Also in this release / 本版还包含

• New test coverage for the L0 contract itself (3 assertions where XML-era tests had none), verified negatively: five typical corruptions each get caught with a precise failure message.

• manifest.json now tracks productVersion alongside kernelVersion — the Nuwa product is at 10.1.0 while the kernel file is still 10.0, and the release notes now report that honestly instead of collapsing the two.

• 新增针对 L0 契约本身的测试覆盖（3 条断言，XML 时代一条都没有），并做了负向验证：五种典型损伤各自都被精确报出失败位置。

• manifest.json 新增 productVersion 与 kernelVersion 并存——女娲产品版本是 10.1.0、内核文件仍是 10.0，Release Notes 现在如实分开报告，不再把两者混为一谈。

Upgrading / 升级

Grab the new zip and swap the factory parts per HOW_TO_UPGRADE / 如何升级.md; your memory files, ledgers and team log are untouched. If you have tooling that reads agent-template.xml, point it at agent-template.md.

下载新版 zip、照《如何升级.md》替换工厂件即可；你的记忆、台账、团队日志原样不动。如果你有工具在读 agent-template.xml，把它指向 agent-template.md。

---

本版组件版本 · Component versions

Component 组件	Role 角色	Version 版本	Commit
agent-modpack	整合包本体 the pack itself	v0.6.0	366ec3d
Nuwa	女娲本体：内核+骨架库 kernel & templates	v10.1.0	480b3e7
nomad	团队骨架+治理件 skeleton & governance	v0.5	d2be04c
dont-reinvent	skill	v1.4.0	d63a635
handoff-protocol	skill	unversioned	fef132f
knock-knock	skill	unversioned	a63424d
sharpen-first	skill	v1.2.0	effdb6c
token-saver	skill	unversioned	495bec9
track-record	skill	v1.0.0	30e61ce
user-md	skill	v1.1.0	5be3d69