fix(control-plane): isolate host completion by Goal instance

[Duang777]

Goal And Delivered Outcome

• Outcome basis / optional anchor: Reproduced control-plane settlement identity collision.
• Goal/source and gap: host_todo_completion.ts preserved an exact GoalRef in commands but derived turn_instance_id from only goal_id, agent_id, and todo_id.
• Observable before → after, with the validation row that proves it: Two Goal instances with one alias produced the same settlement identity, so the successor could be blocked by the predecessor quota receipt. Exact Goal references now use a domain-separated digest that includes goal_instance_id; alias-only requests retain their existing bytes.
• Issue/task and intended base: Self-contained bug fix against loopx-project/loopx:main.

Author Declaration

• Written by: OpenAI model agent, directed by a human operator.

Implemented against

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

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Exact retries keep one settlement identity	implemented	turnInstanceId	host_todo_completion.test.ts
Different Goal instances cannot share a settlement identity	implemented	turnInstanceId	host_todo_completion.test.ts
Alias-only callers keep the legacy identity encoding	implemented	turnInstanceId	Existing host completion and Python adapter tests

• Self-check before submission: Reviewed the host completion, effect ID, and quota receipt call chain. Ran the focused TypeScript and Python suites, TypeScript typecheck, and diff-driven premerge. This PR does not change quota settlement schemas or Goal acceptance state.

Scope And Continuation

• Completed scope and remaining work: Complete within host Todo completion identity derivation. Goal acceptance lifetime fencing is a separate change.
• Slice boundary / successor: N/A for this defect; the fix is independently testable and preserves the alias-only contract.

Validation

• Tested revision: d06fe964a07c6e04c915c4a914888765677cacf2
• Run state: finished
• Input classes: synthetic

Check kind	Result	Public-safe evidence / limitation
unit	passed	node --no-warnings --experimental-sqlite --experimental-strip-types --test tests/control_plane_ts/host_todo_completion.test.ts: 15 passed.
integration	passed	python -m pytest -q tests/test_goal_mode_mcp_settlement.py: 8 passed.
static	passed	npm run -s typecheck:control-plane.
real_entrypoint	passed	loopx canary premerge --from-git-diff --git-diff-base upstream/main: 13 selected checks passed with no manual hold.
real_backend	not_applicable	The changed identity reducer is deterministic and does not access a provider.

• Coverage and gaps: Tests cover retry stability, cross-instance separation, downstream effect separation, and mismatched finalization. Existing adapter tests cover the alias-only path. No backend-specific path changed.

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

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.

Shared-authority RFC fixture impact

N/A. This PR changes local identity derivation, not provider routing or shared-authority schemas.

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

动机

使用显式 GoalRef 的宿主完成适配器，需要区分被删除后以同名重建的 Goal。 同名 Goal 的前后两个实例使用相同 Agent 和 Todo 时，旧代码生成同一完成身份，无法区分前一实例的回执；新代码生成不同身份并拒绝跨实例回执。 独立 RPC 对照确认了身份隔离和错误回执拒绝；这只修复已携带精确实例引用的完成请求。 本 PR 不开放 source-session profile，不迁移历史回执，也不授予新的执行、租约或扣额度权限。 普通 MCP 上下文尚未提供精确实例绑定；source-session profile 的完整宿主和产品验收仍在 #5206。

改动思路

修复复用现有 TypeScript 身份 owner；实例引用是既有事实，哈希是派生值，不新增状态或权限。 当前可交付边界是显式 GoalRef 的完成身份隔离；普通宿主接线和 source-profile 激活仍由 #5206 的 M3 验收负责。 请求仍从 complete_task 经 Python 适配器进入已注册的 TS RPC，expectedIdentity 给 prepare、finalize、vision context/recovery 提供同一身份。显式精确请求把实例加入本地哈希；不改通用 effect schema。普通 Claude、Kiro、Kunlun 的上下文没有 goal_ref，source-profile 普通入口在两版都不可执行；因此本次只认可 RFC 允许在激活前进行的 M1 owner 修复，不能把 RPC 输入可接受当成宿主接线或执行授权。

具体改动

精确 head：7f507438daa4c1ae31874bbfb36a3e721b79cc51；独立基线：ebda6db7b54843bfa339fa71252d1dc5b8af4204。全 diff 只有两个文件、35 additions / 4 deletions，生产部分10/4，其余是一个聚焦回归。作者说没有书面规范，但仓库已有 accepted RFC：docs/architecture/rfcs/goal-instance-identity-and-orphan-recovery-v0.md，spec_revision=ebda6db7b54843bfa339fa71252d1dc5b8af4204，同版规范。

criterion_id	本次判断
M1	implemented：在现有 typed owner 内修复并对照；不 mint 或激活实例。
History stays historical	implemented（完成身份组件）：A/B key 不同，A 回执不能通过 B 的 finalize；不改写历史存储。
Default-off parity	implemented：省略引用的完整 prepare 输出逐字段一致；旧 CLI 结算和恢复保持一笔扣账。
M3	deferred：#5206 仍负责真实实例绑定、commit fence、迟到结果及升级恢复验收。
Product recovery is complete	out_of_scope：RFC M5 的 App/Lark/CLI 全程恢复不在本修复内。

关键代码讲解

• turnInstanceId（host_todo_completion.ts:208）：没有精确引用时保留原三字段编码；有引用时加入内部域分隔和已校验实例。相同请求重试稳定、不同实例分离，不生成新 authority。
• expectedIdentity（同文件:224）：继续调用通用 settlementIdentity，因此现有 effect ID 自动携带新的 Turn key；prepare、vision context、vision refresh 与 finalize 都复用它，无平行 Python 决策。
• finalize（同文件:739）：现有 guard/阶段回执相等检查不变。独立实际 RPC 在 base 会把 A 的五阶段回执当成 B 完成，在 head 于 guard 阶段报 identity mismatch；正确实例的链仍完成。这里回执 provider 使用合成数据，未冒称真实 source-profile 结算。

对主干的风险

最强反例是“同 alias、同 Agent/Todo，迟到的 A 回执进入 B”。相同独立 harness 在 base 的隔离断言失败（exit1），在 head 通过（exit0）；不是从新测试公式反推预期。v0/v1 legacy 完整 prepare 输出相同，5种 null/空/缺失实例/非法实例/alias不匹配输入的完整错误相同，恢复命令与 prepare 身份一致。实际 source registry codec + 普通 MCP context 的不可执行提示也在两版相同，零 provider 调用。

独立检查：head TS15 / base14通过；真实 Python CLI 结算、终止、三个丢响应阶段恢复、已有 successor 复用及完成校验两版各12通过；head typecheck、development advisory、全树语义检查及 premerge 的13个选中检查和3个 diff 检查通过，无 manual hold。没有查询或等待 CI。不能从这些结果推导 live source-profile ABA、原生 PG provider 或跨版本已持久化 exact key 的恢复已验收；改变旧 exact key 后的升级资格仍应在 M3 激活前核验，不能静默重标历史回执。

语义与 CI 对齐

复用现有 GoalRef / settlement 类型所有者；内部哈希盐是 local helper，不是公共协议扩展。未改变默认 legacy 行为、权限、额度、任务接受状态或 agent 消费的义务文字；强制 identity 拒绝仍是机器规则。semantic advisory 未发现支持语法的新词汇仅是提示，完整调用链和对照才支持本结论。

我的整体评价

没有本组件范围内的阻塞发现。delivery judgment 为 justified_increment：long_horizon 的完成身份隔离改进已复现，普通 user_experience 保持；完整宿主采用仍保留 #5206 的边界。代码规模与故障相称，沿现有 TS owner 修改即可。future-facing pass 已检查统一身份入口，现有 expectedIdentity 足够，不需要新抽象。可以合入这个 M1 修复；不表示 source-profile 可以激活或父 RFC 完成。运行时变更仍交维护者合并。

English verdict: APPROVE — 7f50743; exact completion identities and foreign-receipt rejection are independently verified against ebda6db; legacy real CLI retry/settlement and focused native checks pass. M3 host activation and exact historical-key upgrade remain unqualified under #5206; CI was not consulted.