[loopx-agent]

The Turn CLI already plans from the complete work contract, but it also builds a legacy TurnEnvelope before planning and again after a managed claim. This removes those builds and lets managed preparation check the original contract through the existing identity reader.

Goal And Delivered Outcome

• Outcome basis / optional anchor: Independent CLI and managed preparation slice of C1: unify work sources and validate recovery #8266, targeting main.
• Goal/source and gap: Pass the already composed work contract through preparation while preserving the selected Todo, the lease, independent validation and settlement.
• Observable before → after: On both File and SQLite, turn plan preview and details went from 2 to 1 Turn projection RPCs. Managed runs went from 1 to 0 legacy builds for ready work and 2 to 0 when a claim was needed.

Author Declaration

• Written by: model_agent — OpenAI Codex.

Implemented against

• Specification and revision: Turn lifecycle and authority, phases 3–8, and the current complete-source producer in C1: unify work sources and validate recovery #8266.

Criterion	Disposition	Symbol / path	Test or command
Read the original complete contract during preparation	implemented	handle_turn_command, prepare_native_task	Native ready and claim paths on File and SQLite
Check identity and reject corrupted carriers	implemented	complete_turn_host_source	Source-only, malformed source, explicit null/empty adapter, and saved Host readback tests
Preserve the selected Todo through validation and settlement	implemented	tests/test_native_steward_turn_host.py	Real store, independent synthetic validator, original native session and idempotent replay

• Self-check before submission: Compared the same four plan inputs on base and head, reproduced the redundant build on both managed paths, and ran the scoped source and packaged tests below.

Scope And Continuation

• Completed scope: The CLI and steward preparation now use the same work source. The TypeScript identity owner still verifies complete sources; explicit legacy adapters and capture headers retain full validation. Historical readers remain available.
• Slice boundary: The CLI and steward production changes match C1: unify work sources and validate recovery #8266. Landing this slice lets those two production files drop out of the remaining C1 diff. C1 context delivery and recovery qualification continues in C1: unify work sources and validate recovery #8266.

Validation

• Tested revision: 017d949c02723c0b84949a1929ba762828906e5f, compared with the original aa475b29f1976a4fde901deca84672ddb3415152 and current main 79068ba396d937541c5039cdde15bd7d77d4f56f.
• Run state: finished.
• Input classes: synthetic.

Check kind	Result	Public-safe evidence / limitation
regression_parity	passed	Same File/SQLite fixtures, read-only preview and full details, turn plan: 2 → 1 Turn projection RPCs in each case. No legacy envelope on either output.
real_entrypoint	passed	26 native tests, including ready/claim, validation, lease conflict, unavailable work, and replay; plus generic turn run-once completion through a synthetic host.
real_backend	passed	File and SQLite with disposable state. Managed validation checks the same Todo and Turn before settling; absent or malformed proof cannot launch work.
integration	passed	76 source tests for complete Host requests, capture/recovery, controller projection and generic execution. The built wheel passed 8 focused tests plus two fresh CLI processes, with imports resolved from the wheel.
static	passed	Ruff, Git whitespace check, public-boundary scan and signed commit. The diff-scoped premerge canary passed locally; any manual checks are listed by its validation contract.
current_main_merge	passed	The signed merge retained main's capability_automation_source number-token handling while retiring this PR's build_controller_input / TurnControllerInput path. The merged tree passed 105 Host/Native tests, 245 delegation/report tests, 200 relevant TypeScript tests, control-plane typecheck and diff hygiene.

• Coverage and gaps: The RPC counts cover the Turn projection boundary in these fixtures. No model token, latency, or benchmark outcome claim. No provider migration or deployment.

Frontend / Visual Evidence

• UI impact: none. This changes how the existing CLI and managed host obtain their work source; there is no visual change.

Type of Change

• Refactoring (no functional changes)
• Test update

LoopX Area

• Control plane (goals, todos, quota, scheduler, registry, runtime)
• Host or runtime integration

Shared-authority RFC fixture impact

N/A; no change to provider routing or stored authority.

Boundary Checklist

• Diff and review materials contain public-safe evidence.
• Scoped to the linked CLI and native host slice.
• No paired RFC changed.
• UI impact marked none.
• DCO sign-off present.

[loopx-agent]

Reviewer: model_agent, gpt-6.1-sol, OpenAI, runtime_reported, reasoning_effort=max

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

动机 / Motivation

用 CLI 预览任务或从原生会话继续任务的人会经过这段准备。 完整决策组成后，预览和详细计划仍再构建旧 envelope；需领取的 managed Turn 在领取后再次构建。

四组 CLI 预览/详情调用均从 2 次投影到 1 次；managed 就绪 1→0、需领取 2→0，两后端均保留原 Todo、独立 argv 验证、同一会话、结算与幂等重放。

未调用真实模型或重跑发布 wheel。观测只证明该投影边界调用次数下降，没有端到端时延、token 或整个 C1 交付主张；按当前策略未等待远端 CI。

改动思路 / Approach

复用已有完整来源和已有校验，删除过渡构建比在两个构建点加缓存或再定义一个权威对象更小。 与 #8266 的 CLI 和 steward 两个文件逐项比较，增删行相同；其余 envelope 和跨宿主恢复资格留在原 C1 范围。

具体改动 / Changes

评审提交 8bcef37；对照基线 aa475b2。生产代码三个文件 +12/−36，测试两个文件 +48/−4。删除两个无消费者的构建点和仅包裹 envelope 的 Python controller input。

规范：目标分支已接受的合同。

• Turn 3–4（已验证）。路由与准备使用原始来源并保留任务身份。File/SQLite 预览和详细计划均保持同一个 Todo；需要领取时用同一个任务重读。
• Turn 5–6（已验证）。保留原会话、独立验证和失败停止语义。native ready 和 claim、坏来源与明确旧 adapter 均有实际覆盖。
• Turn 7–8（已验证）。同一已验证工作写回后才扣额；回放不重做。两后端保持 committed 结果与原 readback，幂等重放无新宿主调用。
• C1 跨宿主恢复（待整段验收）。完整来源在所有宿主上的上下文交付资格。本切片没有承担跨宿主和长程性能资格。

关键代码讲解

• handle_turn_command。先组合一次决策。只有宿主真实领取后才重读同一 Todo 的权威状态，检查新准备结果为 ready。 代码
• prepare_native_task。把原 work_contract 交给现有验证准备；仍先校验实际 actor、Turn 与 Todo，准备成功后才走原 lease 路径。 代码
• complete_turn_host_source。仅 lineage 允许不带 adapter；显式提供的 adapter 或 Host header 仍走共享 TypeScript 校验。 代码

对主干的风险 / Risks

PASS。四组 CLI 预览/详情调用均从 2 次投影到 1 次；managed 就绪 1→0、需领取 2→0，两后端均保留原 Todo、独立 argv 验证、同一会话、结算与幂等重放。 准确提交 Python 195 项、TypeScript 96 项和 typecheck 通过；快速预检六项直接检查、三项 smoke 通过。 严格质量凭据已绑定到当前评审提交和 PR 目标。

PASS。显式 null/空 adapter、改写来源、错绑捕获与当前权限不一致均被拒绝；历史 Host 捕获可按原 codec 恢复，原生回放不再次启动或结算。

UNVERIFIED。未调用真实模型或重跑发布 wheel。观测只证明该投影边界调用次数下降，没有端到端时延、token 或整个 C1 交付主张；按当前策略未等待远端 CI。

语义与 CI 对齐

未新增开关或改变默认工作资格；独立非原生/只读计划不消费额度，关闭的 memory 不读取来源，历史有 adapter 仍走完整校验。

我的整体评价 / Verdict

未发现阻断问题。 热路径减掉的调用可累计，原生执行仍依赖同一权威来源。 适用于现有 CLI 和受管会话，没有新增界面，也无需用户重新配置或选择任务。

English verdict: APPROVE. Head 8bcef37. Approve this bounded Turn CLI and managed preparation slice. File and SQLite confirm fewer projection calls with original validation and settlement preserved.

[mergify]

This pull request has merge conflicts with main and cannot be merged
until they are resolved. Please rebase or merge the base branch, @loopx-agent.

Choose the remote for the base repository, not an out-of-date fork.
For a fork clone, first inspect git remote -v; upstream must point
to https://github.com/loopx-project/loopx.git. If it is absent, add it
with git remote add upstream https://github.com/loopx-project/loopx.git.
Then run:

git fetch upstream
git rebase upstream/main
# Resolve each conflict, git add the resolved files, then git rebase --continue.
git push --force-with-lease origin HEAD

For a same-repository clone whose origin points to
https://github.com/loopx-project/loopx.git, use origin instead of
upstream for fetch/rebase. If you prefer merging the base, use
git merge <base-remote>/main and push normally.

Keep the DCO Signed-off-by trailer on every commit when you rebase.
https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[huangruiteng]

Reviewer: model_agent, GPT-5, OpenAI, self_reported

Independent approval

动机 / Motivation

用 CLI 预览任务或从原生会话继续任务的人会经过这段准备。 完整决策组成后，预览和详细计划仍再构建旧 envelope；需领取的 managed Turn 在领取后再次构建。

四组 CLI 预览/详情调用均从 2 次投影到 1 次；managed 就绪 1→0、需领取 2→0，两后端均保留原 Todo、独立 argv 验证、同一会话、结算与幂等重放。

未调用真实模型或重跑发布 wheel。观测只证明该投影边界调用次数下降，没有端到端时延、token 或整个 C1 交付主张；按当前策略未等待远端 CI。

改动思路 / Approach

复用已有完整来源和已有校验，删除过渡构建比在两个构建点加缓存或再定义一个权威对象更小。 与 #8266 的 CLI 和 steward 两个文件逐项比较，增删行相同；其余 envelope 和跨宿主恢复资格留在原 C1 范围。

具体改动 / Changes

评审提交 017d949；对照基线 79068ba。生产代码三个文件 +12/−36，测试两个文件 +48/−4。删除两个无消费者的构建点和仅包裹 envelope 的 Python controller input。

规范：目标分支已接受的合同。

• Turn 3–4（已验证）。路由与准备使用原始来源并保留任务身份。File/SQLite 预览和详细计划均保持同一个 Todo；需要领取时用同一个任务重读。
• Turn 5–6（已验证）。保留原会话、独立验证和失败停止语义。native ready 和 claim、坏来源与明确旧 adapter 均有实际覆盖。
• Turn 7–8（已验证）。同一已验证工作写回后才扣额；回放不重做。两后端保持 committed 结果与原 readback，幂等重放无新宿主调用。
• C1 跨宿主恢复（待整段验收）。完整来源在所有宿主上的上下文交付资格。本切片没有承担跨宿主和长程性能资格。

关键代码讲解

• handle_turn_command。先组合一次决策。只有宿主真实领取后才重读同一 Todo 的权威状态，检查新准备结果为 ready。 代码
• prepare_native_task。把原 work_contract 交给现有验证准备；仍先校验实际 actor、Turn 与 Todo，准备成功后才走原 lease 路径。 代码
• complete_turn_host_source。仅 lineage 允许不带 adapter；显式提供的 adapter 或 Host header 仍走共享 TypeScript 校验。 代码

对主干的风险 / Risks

PASS。四组 CLI 预览/详情调用均从 2 次投影到 1 次；managed 就绪 1→0、需领取 2→0，两后端均保留原 Todo、独立 argv 验证、同一会话、结算与幂等重放。 准确提交 Python 195 项、TypeScript 96 项和 typecheck 通过；快速预检六项直接检查、三项 smoke 通过。 严格质量凭据已绑定到当前评审提交和 PR 目标。

本轮将原 PR head 与最新 main 合并，唯一的 turn_envelope.py 冲突按两侧语义解决：保留 main 新增的 capability_automation_source 数值令牌处理，同时移除本 PR 要退役的 build_controller_input 和 TurnControllerInput。合并树独立通过 Host/Native 105 项、delegation/report 245 项、相关 TypeScript 200 项、control-plane typecheck 与完整 diff hygiene；PR 对 main 的净 diff 仍为五个文件。

PASS。显式 null/空 adapter、改写来源、错绑捕获与当前权限不一致均被拒绝；历史 Host 捕获可按原 codec 恢复，原生回放不再次启动或结算。

UNVERIFIED。未调用真实模型或重跑发布 wheel。观测只证明该投影边界调用次数下降，没有端到端时延、token 或整个 C1 交付主张；按当前策略未等待远端 CI。

语义与 CI 对齐

未新增开关或改变默认工作资格；独立非原生/只读计划不消费额度，关闭的 memory 不读取来源，历史有 adapter 仍走完整校验。

我的整体评价 / Verdict

未发现阻断问题。 热路径减掉的调用可累计，原生执行仍依赖同一权威来源。 适用于现有 CLI 和受管会话，没有新增界面，也无需用户重新配置或选择任务。

English verdict: APPROVE. Head 017d949. Approve this bounded Turn CLI and managed preparation slice. File and SQLite confirm fewer projection calls with original validation and settlement preserved.