[xiaobaifly7]

Goal And Delivered Outcome

• Outcome basis / optional anchor: reproduced Windows defect in repository-change-window installation.
• Goal/source and gap: a fresh install writes platform-translated CRLF scripts, while the provider stores SHA-256 digests of LF text and verifies the actual file bytes.
• Observable before → after: on Windows, install exits 0 but immediate verify exits 1 with status=drifted and enabled=false; after this change, the same fresh-repository journey returns status=ready, verified=true, and enabled=true. Both enforcement levels have regression coverage.
• Issue/task and intended base: self-contained bug fix against main, reproduced at 23bc2e0ed8dc8233620b3e349d686a55941dfc41.

Author Declaration

• Written by: model_agent — OpenAI Codex (GPT-6).

Implemented against

• Specification and revision: no written specification; the reproduced defect and request in this PR are the basis. Existing provider behavior is documented in loopx/capabilities/repository_change_window/README.md at the base revision above.
• Criteria:

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Fresh installation verifies clean on Windows without platform-dependent newline translation	implemented	git_hook.py::_write_provider_files	test_cli_install_verify_round_trip_preserves_lf_scripts
Keep the existing LF digest format and exact-byte drift detection	implemented	git_hook.py::_provider_checks remains unchanged	The regression checks LF bytes, then modifies a script and confirms verify fails
Cover both hook-only and reference-guard managed scripts	implemented	Hook writer and SSH-command writer in git_hook.py	The regression is parameterized over both EnforcementLevel values

• Self-check before submission: reviewed the writer's caller, CLI install/verify handlers, digest creation and on-disk verification; ran the red/green regression, focused policy checks, real console install/verify and Ruff. No additional abstraction or refactor is needed at this serialization boundary.

Scope And Continuation

• Completed scope and remaining work: complete within this scope. The existing Python Git IO provider now explicitly writes LF for hooks and the reference-guard SSH script.
• Slice boundary / successor: no new capability, settings, schema or TypeScript decision owner is introduced; no successor is needed for this defect. Previously drifted installations retain the existing reviewed install --replace recovery path.

Validation

• Tested revision: 40b9210490106c61e3d6d3da14c7f31f6334efee.
• Run state: finished.
• Input classes: synthetic.

Check kind	Result	Public-safe evidence / limitation
regression_parity	passed	Before changing production code, uv run --extra test python -m pytest -q tests/capabilities/test_repository_change_window.py -k test_cli_install_verify_round_trip_preserves_lf_scripts failed both enforcement cases at immediate verify. After the fix, both passed.
unit	passed	uv run --extra test python -m pytest -q tests/capabilities/test_repository_change_window.py tests/capabilities/test_repository_change_window_negative.py -k 'test_cli_install_verify_round_trip_preserves_lf_scripts or policy': 5 passed, 23 deselected on Windows / Python 3.12.13. Includes script-tampering rejection and existing policy cases.
real_entrypoint	passed	Ran the checkout's uv run --extra test loopx console against a fresh synthetic Git repository. Install succeeded and verify returned ready/enabled/verified after the fix; the same console journey reproduced drift on the unchanged base.
static	passed	uv run --extra test python -m ruff check loopx/capabilities/repository_change_window/git_hook.py tests/capabilities/test_repository_change_window.py; git diff --check.

• Coverage and gaps: native Windows source-checkout validation only; POSIX execution, installed-runtime validation and the full suite were not run. The regression asserts platform-independent LF bytes and exact-byte tamper detection. An initial attempt to reuse the older repository helper failed before installation because its text-mode git mktree stdin adds CRLF on Windows; the new regression uses a fresh empty commit instead, leaving that unrelated fixture unchanged. The focused selection is not a claim that the full existing module passes on Windows.

Frontend / Visual Evidence

• UI impact: none. This changes generated provider file bytes, not presentation or navigation.
• Before / after / first viewport and preview / states and viewports shown: N/A.
• Source data: synthetic.
• Attention review / craft review: N/A; no UI change.

Type of Change

• Bug fix
• Test update

LoopX Area

• Capability or extension (providers, adapters, skills)
• Host or runtime integration

Technical Direction

• Direction / acceptance reference: existing built-in repository-change-window capability, bundled git-hook provider. This bounded Python IO fix retains the existing owner and public contracts.

Shared-authority RFC fixture impact

N/A; no shared-authority routing, persistence or migration behavior changes.

Boundary Checklist

• Neither the diff nor this PR draft discloses private state, credentials, raw traces, verifier logs, internal links or local machine paths.
• No maintainer-owned benchmark work is duplicated.
• The change is scoped to the reproduced newline defect.
• No paired RFC changed.
• UI impact is marked none.
• The single commit includes a DCO Signed-off-by trailer (git commit -s).

🤖 Generated with Claude Code

[loopx-agent]

Reviewer: model_agent · gpt-6.1-sol · OpenAI · runtime_reported · max

本次评审对应提交 40b9210490106c61e3d6d3da14c7f31f6334efee。

动机 / Motivation

明确为本地 Git 仓库启用 change-window 的操作员，需要在安装后确认提交和传输保护已经可用。文本写入采用平台换行。当默认换行为 CRLF 时，install 成功后 verify 将未修改的脚本标为 drifted。模拟默认 CRLF 时，两种模式的安装从 verify=drifted 变为 ready，已安装的 CRLF fixture 可通过 --replace 恢复。模拟默认 CRLF 时，两种模式的安装从 verify=drifted 变为 ready，已安装的 CRLF fixture 可通过 --replace 恢复。独立验证用 macOS Git，并在两个文本写点模拟 Windows 默认 CRLF。Windows 原机结果来自作者，已安装版本和托管 CI 未纳入验证。

改动思路 / Approach

LF 是现有脚本模板和摘要的共同字节约定，写入时明确编码即可。代码保留在现有 Python Git IO provider。策略决定和 typed CLI 状态沿用当前结构。 现有模板以 LF 构造，旧 writer 在默认 CRLF 环境会把 LF 转换成 CRLF。verify 按真实磁盘字节做摘要并判漂移。在两个写点指定 newline，让字节与既有摘要协议对齐。

具体改动 / Changes

关键代码讲解

• 在两种模式逐个检查受管脚本的原始字节、可执行位和保存的 SHA-256。模拟默认 CRLF 时基线 install 成功但 verify 漂移，当前提交生成 LF 且立即 ready。 文件写入
• macOS 临时仓库使用真实 Git hook。允许时转交原 hook，受限窗口拒绝提交前 hook，reference_guard 的 SSH wrapper 将读取请求实参交给原有本地命令。 校验
• 两种模式的篡改均被拒绝。把基线生成的两套 CRLF 安装交给新提交，保留旧窗口执行 --replace 后恢复 ready。卸载恢复原有 hooksPath 与 sshCommand。 回归

适用协议 loopx/capabilities/repository_change_window/README.md，按基线 23bc2e0 核对。对应检查为 portable_script_bytes、exact_byte_drift 和 existing_recovery_route。 协议

对主干的风险 / Risks

作用范围限于明确启用 provider 的仓库和生成的 Git shell 脚本。预览与默认关闭在两版均没有写 provider 文件。历史 CRLF 安装会继续报告漂移，显式按原窗口执行 --replace 后可恢复。我在 macOS 使用真实 Git，并在文件写入处模拟 Windows 默认换行。Windows 原机验收为作者报告，已安装版与托管 CI 未复验。

语义与 CI 对齐

28 项相关测试通过，Ruff、DCO、RFC 双语差异与 git diff --check 通过。两版在 macOS 上普通安装、受管 hook、原命令转交和卸载的观测一致，两种模式模拟 CRLF 都有红绿结果。

我的整体评价 / Verdict

两种安装模式现在都能落盘 LF 并通过原有校验。实际字节篡改、默认关闭、已有配置和修复路径均有独立证据。改动集中在现有文件 writer，可以通过评审。本次评审覆盖现有宿主与 CLI 行为。独立验证用 macOS Git，并在两个文本写点模拟 Windows 默认 CRLF。Windows 原机结果来自作者，已安装版本和托管 CI 未纳入验证。

English verdict: APPROVE. Both enforcement levels now write stable LF bytes, which the existing exact-byte verification accepts. Matched macOS Git execution and CRLF translation simulation confirm install, drift, replacement, and restoration. The author supplied the native Windows run.