fix(collaboration): commit inbox reads after response validation

[Duang777]

Goal And Delivered Outcome

• Outcome basis / optional anchor: Self-contained collaboration correctness defect reproduced by deterministic Goal recreation and response-enrichment interruption tests.
• Goal/source and gap: read_inbox persisted request and peer-return read receipts before response enrichment completed. Enrichment then resolved a reusable Goal alias again, so a same-name Goal recreation could inspect a replacement workspace while the failed call left success receipts for the original instance.
• Observable before → after, with the validation row that proves it: Before, all three regression cases failed when the production fix was temporarily removed. After, inbox enrichment uses the admitted Goal snapshot and receipts commit only after the original Goal lifetime is revalidated; all three cases pass.
• Issue/task and intended base: No linked issue; self-contained bug fix against main.

Author Declaration

• Written by: OpenAI Codex agent.

Implemented against

• Specification and revision: No written specification; the request and reproduced defect in this PR are the basis.
• Criteria:

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Input readiness uses the Goal admitted at inbox collection time	implemented	loopx/control_plane/collaboration/peers.py::_input_readiness_for_goal	test_inbox_read_rejects_recreation_before_response_commit
Failed enrichment or Goal recreation does not persist read receipts	implemented	loopx/control_plane/collaboration/peers.py::read_inbox	test_inbox_read_does_not_commit_peer_return_before_response_succeeds
Existing direct returns(mark_read=True) behavior remains available	implemented	loopx/control_plane/collaboration/peers.py::returns	tests/test_peer_collaboration.py

• Self-check before submission: Reviewed the full diff and direct callers, checked exact and legacy Goal paths, compared the regression tests against the unfixed implementation, scanned sibling receipt writers, and synchronized the branch with current main. No unrelated refactor or generated file is included.

Scope And Continuation

• Completed scope and remaining work: Complete within this scope. Request reads and peer-return reads now defer receipts until response construction succeeds and exact Goal lifetime admission is rechecked.
• Slice boundary / successor: N/A; no successor is required for this defect.

Validation

• Tested revision: e54dd67dfa1fe7b70184ff1d8418140fb4162e95
• Run state: finished
• Input classes: synthetic

Check kind	Result	Public-safe evidence / limitation
static	passed	Ruff 0.15.2 on both changed files; committed, staged, and unstaged diff checks passed.
unit	passed	Current-head direct suite: 34 passed, 2 platform-specific tests skipped. Broader adjacent suite on the preceding synchronized head: 102 passed, 2 skipped.
regression_parity	passed	Temporary removal of the production fix produced 3 failures; restoring it produced 3 passes.
real_entrypoint	passed	python -m loopx.cli canary premerge --from-git-diff: 10 selected checks passed against the two-file diff with no failures or manual holds.

• Coverage and gaps: The full tests/test_collaboration_goal_instance.py file has 46 pre-existing manager-authorization fixture failures that also reproduce on the unmodified baseline. The changed scenarios and adjacent inbox, peer collaboration, MCP, tracking, pagination, and host-route suites pass.

See validation disclosure guidance.

Frontend / Visual Evidence

• UI impact: none
• Before: N/A
• After: N/A
• States and viewports shown: N/A
• Source data: none
• Attention review: N/A; this change has no user-visible UI.

Type of Change

• Bug fix
• New feature
• Breaking change
• Refactoring (no functional changes)
• Documentation update
• Test update

LoopX Area

• Control plane (goals, todos, quota, scheduler, registry, runtime)
• Benchmark boundary (adapters, runners, verifiers, scoring, evidence)
• Capability or extension (providers, adapters, skills)
• Public docs or presentation surface (README, protocols, dashboard)
• Build, packaging, installer, or CI
• Host or runtime integration

Technical Direction

• Direction / acceptance reference, when applicable: Core control-plane hardening for exact Goal instance isolation.

Shared-authority RFC fixture impact

• Production-scale fixture schema: N/A
• Semantic dimensions changed, or reviewed no-impact rationale: N/A; this PR fixes collaboration receipt timing and does not claim RFC progress.
• Provider conformance arms run: N/A
• Read-only legacy/file/PostgreSQL three-arm rehearsal (required for promotion, runtime-routing, or compatibility-projection changes): N/A

Boundary Checklist

• Neither the diff nor this PR body/comments/attachments disclose private state, credentials, raw traces or verifier output, internal links, or local machine paths (including .loopx/, .codex/goals/, and live ACTIVE_GOAL_STATE.md).
• I did not duplicate maintainer-owned benchmark work unless a maintainer split out a public issue for it.
• I kept the change scoped to the linked issue/task.
• I completed the visual evidence section for UI changes, or marked UI impact none.
• Every commit includes a DCO Signed-off-by trailer (git commit -s).

[loopx-agent]

Reviewer: model_agent — gpt-6.1-sol (OpenAI); runtime_reported; reasoning_effort=xhigh

Exact reviewed head: 9be3e23

动机

需要持续读取同伴请求和结论的 Agent，以及通过 CLI/MCP 跟进委托的操作者，会遇到一次失败读取被记成已成功提供结果的问题。

读取响应在补充 followthrough 时失败：此前仍写成功回执，调用方可确认消费而让后续重试丢失结果；现在失败不写回执、拒绝消费，修复后重试能读回同一个结果。

真实 File 后端的 exact 与 legacy 场景都保留失败后的结论；普通 MCP/CLI 读取、20+3 分页及重试保持正常，实例重建和 Agent 撤权在提交前拒绝，恢复注册后同一请求继续成功。

本增量限定本地协作读回执、原实例与注册检查；不授予 Todo、claim/lease、跨 Goal、外部账户或执行权限，也不关闭完整 Goal 实例激活与真实业务效果验收。

改动思路

把“收集响应内容”和“记录本次已提供给调用方”拆开，先完成 followthrough 和输入版本检查，再用原 GoalRef 与 Agent 注册状态重新进入短提交 guard。文件输入读取复用第一次准入的 Goal snapshot；Python 负责 IO 和响应组合，TS collaboration.goal_instance.decide 继续拥有生命周期规则。既有直接 returns(mark_read=True) 本身就是完整读取，仍保留它的原语义。

具体改动

整份 diff 只有两个现有文件（+231/-35）：peers.py 提取 collection/writer 和 pinned input helper，延迟 request/result read receipts；测试文件增加读取期间实例重建、return enrichment failure 和 return-only recreation 三个负例。没有新 schema、选项、权限、模型调用或前端操作。公开 CLI 和 MCP 使用同一真实 read owner；既有 result_key、20-item overflow 和显式消费机制保留。

关键代码讲解

• _collect_returns（loopx/control_plane/collaboration/peers.py:230）：Collect bounded unconsumed results with corresponding receipt candidates; 21st item detects overflow, never marked read.
• _record_return_reads（loopx/control_plane/collaboration/peers.py:337）：Reuse request lock and immutable result-key/GoalRef receipt shape after qualified composite response.
• _input_readiness_for_goal（loopx/control_plane/collaboration/peers.py:513）：Resolve material versions using admitted Goal workspace rather than fresh alias lookup.
• read_inbox（loopx/control_plane/collaboration/peers.py:600）：Enrich response, then reacquire original lifetime/registration guard before writing request and return read receipts.

验收依据 docs/architecture/rfcs/goal-instance-identity-and-orphan-recovery-v0.md 固定版本 69ad89c7fe214e3fb67d1fe894b65b17e6040461：5.5 implemented，原实例与提交 guard 相连，重建中间态拒绝且不留 read receipt；5.2 implemented，exact identity 不代替权限，Agent 撤权拒绝，恢复注册后同一请求继续，legacy 和 direct mark_read 兼容。

对主干的风险

同一独立 harness 在 immutable base 与当前 head 实测真实 File registry/TS owner：exact 与 legacy 两类 enrichment failure，base 均允许 failed-response 后消费，重试返回 0；head 均拒绝消费，恢复后重试返回 1。同名实例重建时 base 返回成功并留 receipt，head 返回 historical_mutation_forbidden 且无 receipt；两边这次实际输入观察均来自 A，不能把本实验说成 base 已读 B。Agent 中途撤权的 base 同样留成功回执，head 拒绝，重新注册后读回原请求。正常 MCP/CLI、23 条请求的 20+3 分页、同 Goal 另一 Agent、后续新请求、cross-Agent cursor 和结果重放均核验。

tests/test_peer_collaboration.py tests/test_collaboration_mcp.py 34 passed / 2 Windows-only skipped；source-instance 整套在 head 20 passed / 46 failed，base 17 passed / 46 failed。我逐个比较失败 node 与错误详情，完全一致，全部在未改动的 manager_context.deliver source recipient 授权 setup 失败，尚未进入本 PR 的 read_inbox；新增 3 个测试和独立 exact peer 路径通过。因此这是已证实的 pre_existing_unrelated，不能据此宣称整套绿色；相关合并检查仍需独立处理。10 项 premerge 选择检查、diff/compile/module ratchet/full-tree semantic 均通过，advisory 零受支持新增 vocabulary；未查、轮询或等待 CI。

本修复不承诺多份 File receipt 在磁盘故障下原子提交，也不证明 receipt commit 后客户端必然收到网络响应。最终多一次短 source scope 检查有本地读取/锁成本，未做吞吐或延迟 profiler；没有新增外部网络/模型回合或人工参数。此次 scope 未涉及 PostgreSQL authority provider。没有真实外部账户或 packaged UI rendering 声明。

我的整体评价

APPROVE 这个精确 head 的有界读取修复，goal_achieved 指该读回执缺陷。long_horizon=improved：失败结果保留并能原请求重试，防止一次错误在后续轮次被误消费丢失；user_experience=improved：用户使用原命令和原绑定即可继续，没有新确认或重复输入。效果与避免返工的效率正向，吞吐提升未实测。typed vocabulary/authority/default setting 复用，followthrough advice 不被包装成执行授权。future-facing pass 已落实 collection/writer 与 pinned input seam；已有双次 return collection 如要再优化，应先测量，当前无需新框架。46 个既有 source-manager setup 失败保留为集成/merge-readiness 风险，批准不代表合并，本角色不合并此 PR。

English verdict: APPROVE - 9be3e23; real base/head CLI/MCP/File qualification prevents false consumption after failed reads and preserves same-request recovery. 34 peer/MCP tests and 20 source tests passed; 46 identical pre-existing manager setup failures and 2 platform skips remain disclosed; 10 premerge checks passed.