[Klaud Cold] codeowners: own MODELS.md and MODELS_zh.md / 为 MODELS.md 与 MODELS_zh.md 指定代码所有者

[functionstackx]

Adds named CODEOWNERS for MODELS.md and MODELS_zh.md.

MODELS.md    @ankur-singh @chunfangamd @InferenceX/core
MODELS_zh.md @ankur-singh @chunfangamd @InferenceX/core

Why

MODELS.md is the source of truth for which scenarios are active per model and for the deprecation calendar, but until now it only matched the * @InferenceX/core catch-all — so a change to a retirement date needed no vendor-side review.

That matters more than it used to:

• Enactment PRs do exactly what the file says. [Klaud Cold] Enact the August 3, 2026 scenario and precision deprecations / 执行 2026 年 8 月 3 日场景与精度下线 #2493 removed 54 config keys on its authority, and [Klaud Cold] Enact the August 6, 2026 Kimi-K2.5/2.6/2.7-Code retirement / 执行 2026 年 8 月 6 日 Kimi-K2.5/2.6/2.7-Code 完全退役 #2527 removes 17 more.
• .github/codeowner-signoff-verify-prompt.md's deprecated-model check (Check 7) reads the file directly, and has already rejected [Klaud Cold] kimik2.5-fp4-b200-vllm: nightly image, TP/DEP/TEP sweep / Kimi K2.5 NVFP4 B200 nightly 镜像，TP/DEP/TEP 扫描 #2512 and [Klaud Cold] kimik2.5-fp4-b300-vllm: nightly image, TP/DEP/TEP sweep / Kimi K2.5 NVFP4 B300 nightly 镜像，TP/DEP/TEP 扫描 #2513 on it. Editing MODELS.md can flip that gate for other PRs.

Owner choice

The union of the two master-config reviewer sets, because the file spans both vendors:

Owner	Also owns
@ankur-singh	configs/nvidia-master.yaml
@chunfangamd	configs/amd-master.yaml
@InferenceX/core	repo catch-all

All three already appear in .github/CODEOWNERS, so no new handles are introduced.

Both files carry the identical list because AGENTS.md requires an English doc and its _zh counterpart to be edited in the same PR — split ownership would let the pair drift.

Notes

• Placed after the master-config rules and before experimental/operatorx/. CODEOWNERS is last-match-wins; these are exact-path rules and no later pattern matches them, so the placement is safe.
• No behavior change outside review routing. One file, one hunk.

---

中文说明

为 MODELS.md 与 MODELS_zh.md 指定具名代码所有者。

背景：MODELS.md 是各模型启用场景与弃用时间表的权威来源，但此前仅由 * @InferenceX/core 兜底匹配，修改退役日程无需厂商侧评审。当前该文件的影响已显著提升：退役执行 PR 完全依据其内容（#2493 据此移除 54 个配置项，#2527 再移除 17 个）；且 .github/codeowner-signoff-verify-prompt.md 的弃用模型检查（Check 3/7）直接读取该文件，已据此驳回 #2512 与 #2513 —— 修改 MODELS.md 可能改变其他 PR 的合入门槛。

所有者选择：取两份主配置评审人的并集，因为该文件同时覆盖两家厂商 —— @ankur-singh（NVIDIA 主配置）、@chunfangamd（AMD 主配置）、@InferenceX/core（仓库兜底）。三者均已存在于 .github/CODEOWNERS，未引入新账号。两个文件使用完全相同的所有者列表，因为 AGENTS.md 要求英文文档与其 _zh 对应文件在同一 PR 中同步修改，所有权若分离会导致两者脱节。

说明：规则置于主配置规则之后、experimental/operatorx/ 之前；CODEOWNERS 采用「最后匹配优先」，这两条为精确路径规则且其后无可匹配的模式，放置位置安全。除评审路由外无任何行为变更，仅涉及一个文件、一处改动。

[claude]

🟡 The new MODELS.md / MODELS_zh.md CODEOWNERS rules lack a leading slash, so per gitignore-style CODEOWNERS matching they'd match a file with that basename at any depth in the repo, not just the root-level docs — unlike the adjacent configs/amd-master.yaml/configs/nvidia-master.yaml rules, which are path-anchored. Using /MODELS.md and /MODELS_zh.md anchors to repo root and matches the convention already used above.

Extended reasoning...

The bug: GitHub's CODEOWNERS file uses gitignore-style pattern matching. A pattern with no slash in it matches a file with that basename at any directory depth in the repo tree, while a pattern containing a slash is anchored relative to the repo root (or to the CODEOWNERS-relative path, if it starts with /). The two new lines added by this PR:

\nMODELS.md @ankur-singh @chunfangamd @InferenceX/core\nMODELS_zh.md @ankur-singh @chunfangamd @InferenceX/core\n

contain no slash at all, so they are basename-matching rules, not path-anchored rules — despite the PR description explicitly (and incorrectly) calling them 'exact-path rules.'

Contrast with the surrounding code: The two rules immediately above these, which this PR uses as its ownership-inheritance model, are both path-anchored because they contain a slash:

\nconfigs/amd-master.yaml @billishyahao @chunfangamd @seungrokj @yctseng0211 @1am9trash\nconfigs/nvidia-master.yaml @ankur-singh @kedarpotdar-nv @InferenceX/core\n

These only ever match the single root-level file at that exact path. The new MODELS.md/MODELS_zh.md rules are the only unanchored patterns in the whole file, breaking the anchoring convention every other specific rule follows.

Why nothing catches this today: There is currently exactly one MODELS.md and one MODELS_zh.md in the repo, both at the root, so the rule happens to only match the intended files right now — there's no active misrouting. But nothing stops that from changing. GitHub applies CODEOWNERS as 'last match wins' across the whole tree; if any subdirectory ever gains a file that happens to be named MODELS.md (a vendored copy, an experimental doc, a generated report, etc. — e.g. experimental/operatorx/MODELS.md), this rule would silently attach to it too, pulling in the vendor-review and deprecation-gating semantics the PR describes (including gating .github/codeowner-signoff-verify-prompt.md Check 7) for a file nobody intended to cover.

Step-by-step proof:

1. Suppose a future PR adds experimental/operatorx/MODELS.md (e.g. a scoped model-support note for that subsystem).
2. GitHub CODEOWNERS evaluates patterns top-to-bottom and the last matching pattern in the file wins for a given path.
3. The bare pattern MODELS.md (no slash) matches any path whose basename is MODELS.md, including experimental/operatorx/MODELS.md.
4. Since this rule sits after the experimental/operatorx/ @hbarclay rule's neighbourhood in file order... actually it sits before that rule in the current file, but any future bare-pattern rule ordering aside, the core issue is that the pattern is basename-only and will match at unintended depths as soon as such a file exists, regardless of where new rules are placed relative to it.
5. The result: an unrelated subdirectory file would silently inherit reviewer requirements (and potentially any doc gating written to look at 'MODELS.md changes') intended only for the two root docs — with no error or warning, just quietly wrong ownership.

The fix: Prepend a leading slash to anchor each pattern to the repo root, matching the style already used for the master-config rules:

\n/MODELS.md @ankur-singh @chunfangamd @InferenceX/core\n/MODELS_zh.md @ankur-singh @chunfangamd @InferenceX/core\n

This is a one-character-per-line change, has zero effect on current behavior (since only the root files exist today), and removes the latent basename-matching ambiguity going forward. All four independent verifiers confirmed this reasoning and the practical fix; none of them found a current misrouting, which is why this is filed as a nit rather than a blocking issue.