fix(doctor): native lite is blocked by Python and packaged builtin discovery

[Fancy0uth]

Reproduction

Current main 63b6463, locally compiled native 0.7.2-beta.6. On Windows x64, a committed Python-source fixture in a path with spaces and Chinese characters, with only Git + rg + OS tools on PATH and optional providers absent:

• code-intel <fixture> --mode lite --json exits 10, failureNode doctor, diagnostic bootstrap readiness failed, although native core anchors and artifacts are produced.
• Adding Python alone for checkout topology produces completed exit 0; doctor unconditionally marks Python required even though this native core path executes no Python.
• A clean extracted package invoked by absolute path also marks built-in Sentrux missing when the package bin directory is not on PATH: doctor checks source target/release/debug binaries, not its existing packaged binary candidates. Stable v0.7.1 exhibits both restrictions too; these are not claims against unsupported source-only command aliases.

Bounded repair

Keep Python as observed optional tool, not mandatory native-core readiness. Reuse existing packaged graph-binary discovery to recognize the engine already bundled in code-intel. Preserve true missing Git/rg failures and strict required-provider outcomes. Add a real clean-package/no-python/no-pwsh topology regression to install-smoke in the same change per DR-0001. No installer rewrite, no optional provider installation, no baseline reset.

Evidence

Local unabridged evidence: C:/Users/asus/AppData/Local/Temp/code-intel-ux-source-56gz2ug6/evidence.json and code-intel-ux-source-python-s9gi2cho/evidence.json; stable isolated A/B scenarios in code-intel-ux-stable-f6zwun7j/dependency-isolation.json. Source native pre-edit gate passes unchanged Quality6236/cycles0 (pi-unified-exec-21-fb391097.log). Windows gnullvm dev build is explicitly not the official released MSVC binary. Existing #363 ambient-home installer issue and #392 codenexus path PR are untouched.

[Fancy0uth]

Claimed on actual branch main by runtime-install-cli in the current newcomer UX session. Approach: reproduce against the explicitly source-built native binary, make the bounded Rust repair described above, run focused cargo/integration coverage and actual packaged smoke, and retain truthful provider/freshness outcomes. No commit, push or baseline reset authorized.

[Fancy0uth]

Claim metadata correction: the create command requested claimed, but GitHub omitted labels; an explicit retry now confirms Fancy0uth lacks AddLabelsToLabelable permission. Existing runtime-install-cli ownership/branch main and bounded approach remain recorded in the prior claim comment. A maintainer must apply claimed; no label success is claimed here.

[Fancy0uth]

Maintainer label request: please apply claimed for runtime-install-cli on actual branch main, bounded doctor/install regression repair already described. Same-auth REST label attempt on #395 confirms HTTP 403 with current user pull-only permissions. Further source edits are paused; only validation of already-applied work continues. Comment ownership is not represented as a claimed label.

[Fancy0uth]

Validation update, NOT release-ready and no new edits while claimed-label permission is blocked. Native package regression was red before repair (Python/builtin discovery) and passes after repair with only real Git+rg on PATH, package/bin absent from PATH; missing rg still fails, explicit --require-repowise still fails with provider absent. Focused doctor CLI11/primary17/parity3 plus packaged regression1 pass; internal doctor90 and run_execute_publication5 also pass. Final native gate fails on newly oversized primary_entry.rs (810 LOC) and coupling63.33->63.36; scoped rustfmt check fails. Read-only repin identifies exactly2 stale doctor pins in orchestration/integrations.json; no writes. run_commit reports420pass2fail: deferred registry pin and Windows symlink privilege1314. No baseline reset, no warning suppression. Maintainer claimed label is still required before corrections/metadata can be completed.

[2233admin]

DR-0004 stale-claim takeover: original main-branch claim on 2026-09-22 has no push after >48h; remote main is still 63b6463 (2026-09-03), no UX PR. New branch agent/ux-first-run-takeover from clean origin/main; previous host uncommitted work is inaccessible and will not be overwritten. Local-only: no commit, push, or PR without owner authorization. Repair native Rust lite doctor dependency classification (Python/builtin provider optional unless requested), preserve truthful provider/freshness diagnostics and fail-closed required dependencies; add installed-topology regression.

[2233admin]

认领作废 — 接管分支的工作已随 worktree 目录丢失（DR-0004）

Verdict: do（重新可认领）。 不是 freeze：这三张票描述的都是 2026-09-28 之后仍需做的工作，阻塞它们的不是决策，而是一次已经丢失的执行。

复核证据（2026-09-30）

2026-09-28 21:19 的接管评论在三张票上留了同一段记录：原 2026-09-22 的认领超 48h 无 push 被判定过期，新分支 agent/ux-first-run-takeover 从干净 origin/main 切出，并声明"previous host uncommitted work is inaccessible and will not be overwritten"、"Local-only: no commit, push, or PR without owner authorization"。

现在核对那条分支：

• git ls-remote --heads origin 'agent/ux-first-run-takeover' → 空，远端没有这个分支（符合它自己声明的 local-only）。
• git rev-list --count origin/main..agent/ux-first-run-takeover → 0。本地分支存在，但 HEAD 停在 63b6463，即 origin/main 的位置，没有任何独有提交。声明"新分支从干净 origin/main 切出"属实，但也仅此而已。
• git worktree list 仍列出 D:/projects/code-intel-pipeline-ux-takeover 指向该分支，而 ls -d 该目录 → No such file or directory。

即：worktree 目录已被删除，分支无提交，成果只可能存在于那个已消失目录的工作区里。这次接管的工作已经不可恢复，票上没有任何可交付物。

为什么必须记下来

这是 2026-09-30 队列收敛查出的第三个僵尸认领（另两个是 #302、#47），也是唯一一个已经造成实际损失的。三个的共同形状是：认领评论写了分支名，但没有提交、没有 PR、也没有远端分支，于是 worktree 一消失就什么都不剩。

DR-0004 要求认领时记录"分支名 + 一行方案"，它不要求分支必须先推上去——这个缺口是认领协议的盲区。claimed 标签在这里给出的信号强度是假的：它表示"有活跃会话"，而实际是"有过一次尝试"。

恢复方式

三张票的工作面互不重叠，可按各自范围重新认领：

• docs: newcomer quick start uses commands unavailable in stable v0.7.1 #395 docs：修稳定版 v0.7.1 的首次运行/输出说明（README.md、README.en.md、docs/public-beta.md 及相关 Skill/demo 指令），每条已发布命令都要对 packaged topology 证明过，区分仅源码可用的命令。
• fix(doctor): native lite is blocked by Python and packaged builtin discovery #396 fix(doctor)：原生 lite 被 Python 与 packaged bootstrap 阻塞。
• fix(cli): published readableArtifacts links point to deleted staging paths #397 fix(cli)：已发布的 readableArtifacts 链接指向已删除对象。

接手前按 DR-0004 重新加 claimed 并留新分支名与一行方案。建议第一个动作是把分支推到远端，哪怕只有一个 WIP 提交——本次丢失的根因就是全部成果只存在于本地 worktree。

另注：.git/worktrees/ 下仍留有指向已删除目录的登记项，建议用 git worktree prune 清理，避免下次 git worktree list 继续把它显示成活的。

[2233admin]

更正记录（2026-09-30 收敛会话自查）

2026-09-30 的收敛判决称"worktree 目录已被删除、分支无任何提交、远端不存在"，工作已丢失。收敛会话在清理该失效登记时重新核验了一次，结论维持不变，但当时用的验证手段不够稳，记录在此以免下一任重蹈。

当时的核验链与它的弱点

原判决用了三条证据：目录不存在、rev-list --count origin/main..branch = 0、远端无此分支。

其中"目录不存在"当时是用 ls -d 得出正确结果的。但收敛会话后续复查时，git worktree list 仍列出该 worktree、git worktree prune -v 报"无失效项"、而读取 .git/worktrees/code-intel-ux-takeover/gitdir 里的路径做存在性测试又报存在——三者互相矛盾，一度让人怀疑原判决是错的。

三方独立复核的结论

手段	结果
Test-Path (PowerShell)	False
glob 工具	Path not found
git worktree remove <path>	fatal: ... is not a working tree

三者一致：目录确实不存在。矛盾来自 git 的失效登记本身没被 prune 触发——它的判定依据是 .git 文件在不在，而那个路径在 Git Bash 下的存在性测试给出假阳性。

判决不变

agent/ux-first-run-takeover 的工作确实已丢失。 三条独立证据仍成立：分支相对 origin/main 零独有提交、远端无此分支、worktree 目录不存在。#395/#396/#397 描述的工作在 2026-09-28 那次接管后没有产生任何可交付物。

顺带：失效登记已清理

.git/worktrees/code-intel-ux-takeover 这个指向已删除目录的登记项已移除（元数据先移到 /tmp/ux-wt-metadata-removed 保留副本，可回滚）。现在 git worktree list 只剩两个有效项。

若你手上还留有那次的未推送成果，注意它不在本机这个路径——按接管声明原文："previous host uncommitted work is inaccessible and will not be overwritten"，那份成果在另一台机器上。