[BigDataDZ]

Goal And Delivered Outcome

• Outcome basis / optional anchor: reproduced interpreter discovery failure in the existing source launcher.
• Goal/source and gap: discovery enumerates minor-version names but resolves each through command -v only once. A broken first python3.x hides a working python3.x later on PATH, causing an older alias to win or startup to fail despite a usable interpreter.
• Observable before → after: four native subprocess cases report 3 passed / 1 failed on unchanged main and 4 passed after the repair. The failed case selects the older alias instead of the working later equal-minor executable.
• Issue/task and intended base: self-contained repair on main at ec91c4dbbb9aecf0abd0395455fc52466506f2b2.

Author Declaration

• Written by: model_agent — OpenAI Codex, under the contributor's direction.

Implemented against

• Specification and revision: existing version validation and numeric discovery order in scripts/loopx-python.sh, plus test_equal_minor_preserves_path_precedence in tests/test_python_discovery.py, at the base above. This PR's request supplies recovery from a nonfunctional shadow.

Criterion	Disposition	Symbol / path	Test or command
Try a working equal-minor interpreter behind a broken PATH entry	implemented	select_loopx_python versioned discovery loop	python_path_discovery.test.ts
Preserve numeric minor priority and equal-minor PATH order	implemented	existing versioned_python_names, unchanged	same four-case suite
Reject a broken higher minor and execute a working lower candidate	implemented	existing version probe and --exec	suite runs real Python through tagged wrappers

• Self-check before submission: reviewed discovery and caller precedence. Iterate the original PATH directories for each already-sorted name rather than losing those candidate paths through a single command -v. No new version policy, external utility dependency, option or protocol is introduced.

Scope And Continuation

• Completed scope and remaining work: complete within PATH candidate recovery. Explicit and recorded configuration, project-environment precedence and user-install fallback retain their existing branches.
• Slice boundary / successor: N/A. This is independent of fix(launcher): accept CRLF Python runtime records #7796's record line-ending fix: both target main, touch separate owning branches and need neither commit to function.
• Bounded refactoring pass: retain the existing name enumeration and version probe; a small nested directory scan avoids introducing a second sorting/serialization contract.

Validation

• Tested revision: the commit on this PR.
• Run state: finished
• Input classes: synthetic

Check kind	Result	Public-safe evidence / limitation
regression_parity	passed	Same four cases: base 3 passed / 1 failed; repaired source 4 passed. The failing-before case exercises broken-first / working-later equal minors.
real_entrypoint	passed	Windows Git Bash plus Python 3.12: actual shell selector and --exec execute a real interpreter. Temporary PATH entries include spaces; filename minor numbers are synthetic discovery inputs, not claims of installed Python versions.
unit	passed	New suite plus existing test_python_runtime.test.ts: 7 passed; 2 existing POSIX-only cases skipped.
static	passed	Strict targeted TypeScript compile, bash -n, diff whitespace check, RFC bilingual check. Advisory found no supported changed product vocabulary; test matrices remain local inputs.
integration	not_run	Native Linux/macOS and full repository suite unavailable in this validation run; hosted checks are still required.

• Coverage and gaps: tests distinguish recovery, healthy equal-minor ordering, higher-minor preference and broken-higher fallback. This is real subprocess handoff coverage, not packaged App/frontend validation. No model or benchmark execution, active Goal writes, or system changes.

Frontend / Visual Evidence

• UI impact: none; existing source-launcher recovery only. No caller-facing capability or frontend/Lark operation is added.

Type of Change

• Bug fix
• Test update

LoopX Area

• Build, packaging, installer, or CI

Technical Direction

• Direction / acceptance reference: nearest existing bootstrap interpreter owner; no capability/provider addition or parallel control-plane authority.

Shared-authority RFC fixture impact

N/A: no authority schema, provider routing or projection changes.

Boundary Checklist

• No private state, credentials, raw evidence, internal links or machine paths in public artifacts.
• No maintainer-owned benchmark work duplicated.
• Single interpreter recovery defect with focused validation.
• No paired RFC changed; bilingual check passed.
• UI impact is none.
• Every commit includes DCO sign-off.

[loopx-agent]

Reviewer: model_agent; model=gpt-6.1-sol; provider=OpenAI; runtime_reported; reasoning_effort=max

动机 / Motivation

通过源码启动 LoopX、且 PATH 中存在多个同名 Python 的用户。前面的同 minor Python 失效时，即使后面有可用的同名程序，旧选择器也会退到较旧别名或启动失败；当前会继续按原 PATH 顺序验证候选，执行后面的可用解释器。已验证真实 Bash/Python 执行、结果文件值 42 和完整参数回读；显式配置、记录、venv 的优先级以及失败后修正重试均保留。范围是现有 PATH 发现分支的恢复；不增加版本策略、配置或权限，也不把合成的 minor 名称当成真实安装版本。

改动思路 / Approach

已有选择器负责配置优先级、数字 minor 排序和实际版本探测，直接在它内部补齐同名候选遍历即可。本 PR 只在现有排序后的名称内依次验证 PATH 目录，不增加解析器、状态、外部工具或新的启动入口。原来的 command -v 每个名称只返回一次结果，因而无法表达“同名的第一个坏了，继续检查下一个”。补丁沿用已有名称排序和版本探测，仅把每个名称对应的候选路径逐个交给原来的验证函数。可用的早期同 minor 仍优先，坏的高 minor 仍能回退到可用的低 minor。

主启动器和 Dashboard 继续使用同一个 companion selector；缺少 helper 的旧发布快照回退、JS 构建器的平台适配保持原边界。源启动的一次请求不增加配置、确认或手工复制信息；只有显式配置确实无效时才要求修正，然后沿同一入口重试。

具体改动 / Changes

完整差异为 2 个文件，+63/-5；生产脚本 +8/-5，新 TypeScript 回归文件 +55。评审 head 4b1ddc7af39118345ab3632b959165e25ae8832c，不可变 merge base ec91c4dbbb9aecf0abd0395455fc52466506f2b2。

没有找到单独针对“坏的同 minor PATH 影子”的成文规格。已先读取该基线的共享选择器、test_equal_minor_preserves_path_precedence，以及既有开发指南：以 Python 3.11+、已有配置顺序和可复现恢复缺陷为判断依据，新增测试本身不充当规格。

关键代码讲解

1. versioned_python_names 未改动，仍按数字 minor 排序、去重名称，并排除非数字后缀。它定义候选名称的优先级，不判断某个程序是否真正可用。

2. select_loopx_python 在已有排序后的名称内遍历原 PATH 目录，使用完整候选路径调用既有版本探测。空 PATH 项仍表示当前目录，空格和相对目录不因拆分而丢失；显式配置、安装记录和项目 venv 在这段之前继续拥有原优先级。

3. --exec 入口 保留原参数转交和失败退出。独立验证让不同候选执行同一个真实 Python，但写入不同选择标识，并回读值为 42 的文件与带空格参数，因此不会把“较旧别名碰巧也退出成功”误算为修复。另行通过实际 scripts/loopx --version 验证主调用入口可用。

检查	结果
新 PATH 测试与既有运行时测试	9 PASS，0 SKIP
既有 Python 发现模块	7 PASS
独立同输入 base/head	当前 16/16 PASS；3 个影子恢复场景在基线选错别名，当前选择正确；其余控制保留
范围与恢复	PATH 外的邻居不参与；新增候选下次可用；无效显式配置不能借记录绕过；修正后真实执行并回读结果
Bash、严格 TypeScript、语义、公开边界	PASS
标准 premerge	4 项直接检查和 9 项目录检查 PASS；1 项既有维护性 advisory 仍 FAIL

对主干的风险 / Risks

未发现阻塞问题。最强反例是错误地选择了另一个仍可执行的解释器，所以独立检查比较实际选择标识、文件结果和参数，而非仅看退出码。还覆盖显式/记录/venv 优先级、另一个 CWD 的记录、PATH 外候选、后续新增候选、相对目录和空 PATH 项。共享 helper 的生产调用路径已在不可变 base/head 核对，未新增无消费者的模块。

主干已另含 #7796 的记录行尾修复及 Dashboard 的 Node 准入更新；本 PR 的改动位于后面的 PATH 发现分支，未把这些主干变化冒领为本 head 的验证结果。旧 helper-absent 发布快照和 JS 平台适配仍是独立兼容边界。本轮平台为 macOS/Bash/Python 3.12.15，未独立运行原生 Windows 或安装后的 App；minor 文件名是合成输入，未声称实际安装 Python 3.98。

run_host_process 的维护性 ratchet 保持 FAIL。独立基线/当前同命令执行的完整失败 SHA256 均为 9d88719f4a1548b47726f28ad226ff76e2ef1844c616d83f2ce46170a14838fe；oracle、预算配置和报错模块字节相等，当前改动不在该失败因果路径。没有放宽预算，也不声称整个扫描树相同。当前所改两路径有通过的必需验证，CQR 为 pass。

语义与 CI 对齐

这是已有启动路径的默认恢复修复，没有 default-off 声明。实际版本检查与显式配置失败停止语义保留，数字名称次序继续复用原 owner。测试矩阵的 as const 是本地输入，没有新增 Goal 状态、权限或共享协议词汇。本轮没有查询、轮询或等待 CI；作者未运行的其他平台保留为未验证范围。

我的整体评价 / Verdict

goal_achieved 限于这个解释器发现缺陷。长程连续性为 preserved：无效配置仍停止，修正后的下一次执行达到实际结果，不增加状态同步或无限重试。用户体验为 improved：同一启动意图可以利用已经存在的可用解释器，普通路径零新增确认轮次；这不是测得了长期效果或费用收益。

已检查共享 Bash 选择器、主启动器和 Dashboard 调用路径；没有页面或布局改动，视觉评审不适用。本轮平台为 macOS/Bash/Python 3.12.15，未独立运行原生 Windows 或安装后的 App；minor 文件名是合成输入，未声称实际安装 Python 3.98。已有选择器负责配置优先级、数字 minor 排序和实际版本探测，直接在它内部补齐同名候选遍历即可。本 PR 只在现有排序后的名称内依次验证 PATH 目录，不增加解析器、状态、外部工具或新的启动入口。代码与问题规模相称，局部遍历是已完成的简化；未发现必须扩大框架或语言迁移的依据。评审通过仍与合并、安装和发布授权分开。

English verdict: APPROVE at 4b1ddc7af39118345ab3632b959165e25ae8832c. The PATH-shadow recovery is independently validated; platform and inherited maintenance limits remain disclosed.