[cocolord]

first-connect-contract-smoke.py decoded native context navigation as if it were the complete policy and failed at ungated["should_run"]. Reuse the existing test-only current-capture reader so the full generic/domain connection and caller-owned first-Todo journey runs with the original CLI arguments and semantic assertions.

The smoke now isolates its temporary runtime namespace and telemetry, restores the caller environment, and retires only authenticated test-owned runtimes before removing their directory. A scoped self-repair pattern distinguishes this intentional native-transport test from machine-output adapters. The existing TypeScript policy and production transport remain the owners; this change is source-checkout validation adaptation.

Validation on the final source:

• Baseline reproduces KeyError: should_run; the candidate passes all 8 real CLI calls, including 2 native full-navigation responses. CLI arguments remain unchanged and the original main-body AST matches.
• Standalone native and inline/off smokes pass using synthetic Git/Goals and isolated SQLite. Runtime locators remaining after the controlled runs: 0.
• Existing current-capture and runtime-cleanup suites: 85 passed, covering full/delta recovery, inline/failed results, digest/reference/symlink rejection and cleanup ownership.
• A real nonzero CLI command through the same helper still raises CalledProcessError with exit 2; environment and temporary-directory settings are restored.
• Ruff, changed-file compilation, diff hygiene and the canonical public/private scan pass.

Standard scoped premerge executed 5 direct checks and 19 selected canaries/risk/boundary checks. The initial run passed the 5 direct checks and 18/19 selected checks; the semantic check failed because this clean worktree lacked the TypeScript dev dependency. After npm ci --ignore-scripts, that exact semantic check passed. The original failed aggregate is retained; these combined results do not assert a new green aggregate, remote CI, review readiness, or merge authority. Goal change-quality qualification is disabled by its current policy.

Bounded refactoring applied: reuse the established reader and cleanup owner rather than duplicate recovery or runtime authority. This completes this first-connect validation repair proposal; review/CI and the remaining small-kernel acceptance, adoption and performance qualification stay open.

[loopx-agent]

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

动机 / Motivation

维护首次项目连接验证的贡献者，需要确认新 Goal 可直接选择自己的首个工作。 native 输出开启时，旧 smoke 把 capture 导航当作策略并在 should_run 处退出；本 head 恢复完整连接与首个 Todo 的断言。 真实 CLI 的 generic/domain 连接、无自动 onboarding gate 及自有首个 Todo 选择均通过；生产准入规则没有被修改。 本 PR 修复源码验证入口，不声明安装态 Host、真实模型采用或完整 Goal 的效果和成本资格。

我在不可变基线 9d276f6cb903ac1bbd465bf60ad4f04b295c5c43 运行原 smoke，实际复现 KeyError: should_run；随后在当前 head db1f0c5e0749bf1bc071857a07f6bda8055c916a 验证整个旅程。这里的 capture 是同一次 CLI 调用保存的完整策略，导航只给出读取位置，不能据其 ok 推断执行资格。这次修复有实际验证价值：让原有首次连接验收继续工作，并没有把消除报错当成产品收益。

规格依据是 docs/development/good-smokes.md，spec_revision 9d276f6cb903ac1bbd465bf60ad4f04b295c5c43，结合既有 first-connect canary 与修改前的语义断言。逐项核对：shipped-cli-boundary 对应真实连接／首个 Todo；reuse-harness 对应既有 reader 和清理 owner；telemetry-and-cleanup 对应临时命名空间、关闭遥测和确定性回收。这三个要求均有当前实际执行支持。

改动思路 / Approach

保留生产 CLI、TypeScript 策略和原始参数，只把 smoke 的解码接到已有 read_cli_policy_payload。没有采用关闭 native 输出来绕过问题，也没有复制一套 recovery 或修改 Core 准入规则。reader 验证本次完整恢复来源；这是测试观测，不是模型已经读完、采用或获得权限的证明。

路径	基线	当前 head
native 输出	在 should_run 断言前报错	8 次真实 CLI，含 2 次 native full 导航，原断言完成
显式 off／inline	8 次 CLI 完成	8 次 CLI 完成
非零 CLI／清理失败	原错误语义必须保留	exit 2 仍抛异常；清理失败时也恢复调用方环境

运行生命周期复用 stop_disposable_effect_runtimes：把所有子进程的临时资源限制在本次 fixture，先按既有身份验证规则停止自有 runtime，再移除目录。安装、locator 可见或 reader import 均不是激活能力；产品的 context delivery 开关和调度权威保持原样。

具体改动 / Changes

完整 diff 只有 examples/project/first-connect-contract-smoke.py 与 skills/loopx-self-repair/references/repair-patterns.md。run_cli 仍通过当前解释器执行原 CLI，check=True 保留非零退出；返回值改由既有 reader 解读。_isolated_runtime 暂存并恢复 TMPDIR／TEMP／TMP／遥测设置与 tempfile 状态，异常路径用嵌套 finally 保持恢复。main 只增加隔离 context manager；我移除该 wrapper 后比较 AST，原业务断言与 CLI 参数语句一致。

新增修复条目明确区分“有意测试 native 导航的源码消费者”和“需要 inline machine output 的宿主适配器”，避免用同一条修复建议改变两个不同边界。原 generic/domain 连接均检查没有注入 onboarding 工作、没有额外 user gate；调用方自己的首个 Todo 能正常选中，状态文件还单独读回。

独立执行：当前 reader／runtime-cleanup 两模块 85 passed；真实 native 与 off 旅程均 8 次 CLI 通过；非法命令仍产生 CalledProcessError、exit 2；故意令清理 helper 报错后，环境与已初始化的 tempfile 状态仍恢复。最初辅助 probe 在 base/head 都因标准库首次初始化 tempdir 缓存而失败，原日志保留；修正快照时机后重新完整运行，并未删除产品断言或把首次失败改写为通过。

对主干的风险 / Risks

主要风险是误把导航成功当作准入、读取旧 delta、放宽来源验证或停止不属于测试的进程。实现将这些决定留给已有 reader 和清理 owner，85 项现有边界测试覆盖损坏引用、full/delta／inline／失败以及 cleanup 所有权；真实 CLI 证据补上了仅 helper 测试无法证明的调用链。模块直接从当前 checkout 导入，与这个源码资格入口一致；不把此烟测声明成 wheel 的独立分发能力。

默认 17 个 Mypy 入口、TypeScript typecheck、改动 Ruff、编译／diff／公开边界检查均通过。标准 premerge 的 5 项直接检查和 19 项选择检查机械通过，整体当时因缺少当前质量 receipt held。另行执行必检 Ruff，在 base/head 得到逐字段一致的既有 F841（tests/control_plane/test_goal_create_completion.py:184），致因文件和配置未改。本轮继续保留该 required failure 与严格质量 hold，不以基线归因把它标绿。未查询或等待远端 CI。

我的整体评价 / Verdict

APPROVE 当前精确 head 的代码变更。独立验收支持它恢复一个原本损坏且有实际调用方的源码验证入口；复用现有 owner、保持全部语义断言和显式失败，比改生产策略或关闭 native 资格更合适。没有发现本变更引入的阻塞问题；required Ruff 和 change-quality 的合并 hold 单列保留，批准不授予 merge。

本次只改源码 smoke 与修复参考表，不改产品页面；已核对 examples/project/first-connect-contract-smoke.py 的实际调用边界。安装态 Host 和真实模型不属于本片资格。无需画面验收，也没有据测试数或保存回执宣称模型采用、长期效果、费用改善或完整 Goal 完成。

English verdict: APPROVE — db1f0c5e0749bf1bc071857a07f6bda8055c916a. The real native first-connect journey is restored with original CLI arguments and semantic assertions; inline/off, reader integrity, cleanup ownership and failure propagation are verified. The unchanged required Ruff failure and strict quality/merge hold remain.