fix(heartbeat): resume PR dependencies from complete retained facts

docs/architecture/rfcs/loopx-overall-roadmap-v0.md

@@ -636,6 +636,18 @@ admission, fresh task derivation and frontend/Lark qualification remain open.
 中文：异常读回区分本次调用未知副作用与原 Turn 的持久观察/待核对效果，只完成
 R2 的一个恢复切片；调用方准入、新任务生成与前端/Lark 验收仍未闭环。
 
+[PR #6163](https://github.com/loopx-project/loopx/pull/6163) proposes the
+S2/S7/S10 PR dependency observation slice in the existing opt-in heartbeat
+prequota path, independent of open review-queue membership. The Todo typed owner
+selects exact unfinished waits and durable polling cadence from complete
+retained facts. Resume readback shares that source before display compaction,
+retaining the original merge proof through history growth and restart without
+clearing requirements or changing claims, leases or status. Real File/SQLite
+CLI growth/restart and repository/failure negatives qualify this bounded slice;
+installed host adoption and TurnEnvelope polling remain distinct. It does not
+close legacy retirement or sustained authority qualification. See
+[Todo readback](../../reference/todo-continuation-readback.md#pr-dependency-observation--pr-依赖观察).
+
 [PR #5283](https://github.com/loopx-project/loopx/pull/5283) is the proposed
 S2/S10 read-preview cost slice. Summary and succession validation share the
 existing TS owner request; empty succession skips transport. Read models are
@@ -670,6 +682,15 @@ admission remain separate open acceptances; this PR does not retire Python
 planning or establish a machine default.
 
 中文：
+[#6163](https://github.com/loopx-project/loopx/pull/6163) 拟交付 S2/S7/S10
+PR 依赖观察切片：既有 opt-in heartbeat prequota 不依赖 open review 队列成员，
+Todo 类型化 owner 从完整保留的事实选择精确未完成等待及持久核验节奏。恢复回读
+在展示裁剪前复用同一源，历史增长和重启保留原合并证明，不清要求、不改认领、
+lease 或状态。真实 File/SQLite CLI 的增长／重启及仓库／失败负例只验收本批；
+安装态 host 采用、TurnEnvelope 核验分别跟踪，不关闭 legacy 退役或 authority
+持续运行资格。参见
+[Todo 回读](../../reference/todo-continuation-readback.md#pr-dependency-observation--pr-依赖观察)。
+
 [#5283](https://github.com/loopx-project/loopx/pull/5283) 是拟交付的 S2/S10
 只读预检成本切片：摘要与继承验证共用既有 TS owner 请求，空继承行跳过传输；
 读模型与写适配器分离，未选中的报告、Lark、扩展及 Host 路由延迟加载。兼容导出

docs/architecture/rfcs/loopx-overall-roadmap-v0.zh-CN.md

@@ -466,6 +466,15 @@ executor/profile 检查启动条件。任务准入、当前 pinned 验收绑定
 表达；未探测的通用/云端运行时保留未知。CLI、已启用 MCP 和新 Chat 工具共用此
 检查，不新增状态账本，也不启动执行。
 
+[#6163](https://github.com/loopx-project/loopx/pull/6163) 拟交付 S2/S7/S10
+PR 依赖观察切片：既有 opt-in heartbeat prequota 不依赖 open review 队列成员，
+Todo 类型化 owner 从完整保留的事实选择精确未完成等待及持久核验节奏。恢复回读
+在展示裁剪前复用同一源，历史增长和重启保留原合并证明，不清要求、不改认领、
+lease 或状态。真实 File/SQLite CLI 的增长／重启及仓库／失败负例只验收本批；
+安装态 host 采用、TurnEnvelope 核验分别跟踪，不关闭 legacy 退役或 authority
+持续运行资格。参见
+[Todo 回读](../../reference/todo-continuation-readback.md#pr-dependency-observation--pr-依赖观察)。
+
 [#5283](https://github.com/loopx-project/loopx/pull/5283) 是拟交付的 S2/S10
 只读预检成本切片：摘要与继承验证共用既有 TS owner 请求，空继承行跳过传输；
 读模型与写适配器分离，未选中的报告、Lark、扩展及 Host 路由延迟加载。兼容导出
@@ -488,6 +497,18 @@ R2 持续运行、真实调用方采用、冷 CLI/p95 与 provider 准入仍是
 此 PR 不代表 Python 规划退役或机器默认切换。
 
 English:
+[PR #6163](https://github.com/loopx-project/loopx/pull/6163) proposes the
+S2/S7/S10 PR dependency observation slice in the existing opt-in heartbeat
+prequota path, independent of open review-queue membership. The Todo typed owner
+selects exact unfinished waits and durable polling cadence from complete
+retained facts. Resume readback shares that source before display compaction,
+retaining the original merge proof through history growth and restart without
+clearing requirements or changing claims, leases or status. Real File/SQLite
+CLI growth/restart and repository/failure negatives qualify this bounded slice;
+installed host adoption and TurnEnvelope polling remain distinct. It does not
+close legacy retirement or sustained authority qualification. See
+[Todo readback](../../reference/todo-continuation-readback.md#pr-dependency-observation--pr-依赖观察).
+
 [PR #5283](https://github.com/loopx-project/loopx/pull/5283) is the proposed
 S2/S10 read-preview cost slice. Summary and succession validation share the
 existing TS owner request; empty succession skips transport. Read models are

docs/reference/todo-continuation-readback.md

@@ -7,6 +7,57 @@ or an explicit `no_followup=true`. This read policy lives in
 its decisions. It belongs to the existing Todo control plane, uses the selected
 AuthorityStore, and adds no capability or extension provider.
 
+## PR dependency observation / PR 依赖观察
+
+`pr_merged:owner/repo#42` identifies a dependency; absence of a merge event
+does not prove that GitHub is still pending. Delivery-history supervision no
+longer exempts that unobserved wait. Causal blocked closeout continues to require
+a registered `monitor_changed` or `todo_done` target with current readback.
+
+The existing `heartbeat-prequota --goal-id GOAL --agent-id AGENT` observes exact
+unfinished PR dependencies from the complete Todo head, including deferred work.
+It polls at most four distinct repository-qualified dependencies per invocation,
+with a durable 30-minute cadence for successful and failed attempts. A missing
+repository binding is reported for repair; it never guesses from PR numbers.
+An exact merged observation appends the existing native `pr_merge` fact so the
+original condition becomes satisfied. Resume readback uses the same complete
+retained Goal log as observation, before any recent-event display limit. The
+typed resume owner keeps one original merge proof per exact dependency; more
+than 500 unrelated events or 256 other dependencies cannot evict that proof.
+List, status, quota and archived reads keep their bounded presentation, and
+restarts need neither repeated remote reads nor rewritten merge timestamps.
+Queue disappearance, closed-without-merge
+and provider failures cannot satisfy it. This observation changes no Todo status,
+requirements, claims, leases or Goal acceptance. Deferred ready work still needs
+its normal lifecycle transition.
+
+Generated heartbeat prompts include prequota only when the host declares
+`--available-capability external_evidence_poll`; filesystem/shell declarations
+alone do not opt in. Omitting that declaration keeps external observation off.
+Explicit prequota invocation authorizes compact metadata reads and local
+observation writeback, never merge, review, remote mutation or quota spend.
+Read back with `todo list --goal-id GOAL --todo-id TODO --format json`.
+These same typed dependency facts are the boundary to reuse in TurnEnvelope;
+this heartbeat adoption does not claim that every Turn host already polls them.
+
+`pr_merged:owner/repo#42` 只标识依赖；缺少合并事件不证明 GitHub 仍在等待。
+历史交付监督不再豁免这种未观察等待，因果阻塞结算仍要求已注册的
+`monitor_changed` 或 `todo_done` 目标及当前回读。
+原生 `heartbeat-prequota --goal-id GOAL --agent-id AGENT` 从完整 Todo 源选择
+精确的未完成 PR 依赖（含 deferred），每次最多核验四个不同仓库限定的依赖。
+成功和失败均保留 30 分钟核验节奏，重启后仍有效。仓库缺失需要修复，不能按编号猜测。
+真实合并观察写入既有 `pr_merge`，原条件自然满足。回读与观察共用完整保留的 Goal
+日志，在最近事件展示裁剪前由 TS resume owner 保留每项精确依赖的原始合并证明；
+超过 500 条无关事件或 256 项其他依赖不能挤走它。list、status、quota 与归档读取
+仍保留原展示边界，重启不需要重复远端读取或改写合并时间。离开 open 列表、关闭未合并或
+provider 失败不能解除等待。观察不修改 Todo 状态、要求、认领、lease 或 Goal 验收；
+已满足条件的 deferred Todo 仍走正常生命周期转换。
+生成的 heartbeat 仅在声明 `--available-capability external_evidence_poll` 时加入
+prequota；只声明 filesystem/shell 不会启用。省略该声明即关闭外部观察。
+显式调用仅授权紧凑元数据读取和本地观察记账，不授权合并、review、远端写入或额度支出。
+通过 `todo list --goal-id GOAL --todo-id TODO --format json` 回读。
+TurnEnvelope 应复用同一类型化依赖事实；本批未证明所有 Turn host 已接入自动核验。
+
 ## Relationship evidence
 
 The policy evaluates the complete available Todo graph before role, status,

loopx/capabilities/issue_fix/pr_lifecycle.py

@@ -860,6 +860,7 @@ def fetch_github_pr_lifecycle_payload(
     reference: Mapping[str, Any],
     *,
     timeout_seconds: int = 10,
+    compact: bool = False,
 ) -> dict[str, Any]:
     repo = str(reference.get("repo") or "")
     number = reference.get("number")
@@ -877,7 +878,7 @@ def fetch_github_pr_lifecycle_payload(
             repo_slug,
             "--json",
             ",".join(
-                [
+                ["state", "url", "mergedAt", "closedAt"] if compact else [
                     "baseRefName",
                     "closingIssuesReferences",
                     "closedAt",

loopx/capabilities/issue_fix/pr_wait_reconcile.py

@@ -0,0 +1,92 @@
+"""Transport the typed dependency poll plan into existing public lifecycle facts.
+
+This internal issue-fix adapter changes neither Todo requirements nor claims,
+leases, status, settlement or review authority. It does not consult an open PR
+queue: dependencies are selected from the complete current canonical Todo head.
+"""
+from __future__ import annotations
+
+from pathlib import Path
+from subprocess import TimeoutExpired
+from typing import Any
+
+from ...agent_registry import registered_agent_ids_from_registry
+from ...control_plane.effect_runtime import effect_runtime_result
+from ...control_plane.runtime.time import now_utc_iso
+from ...control_plane.todos.resume_condition import compact_todo_resume_items
+from ...history import load_registry
+from ...paths import resolve_runtime_root
+from ...rollout_event_log import (
+    append_rollout_event, build_rollout_event, rollout_event_log_path,
+)
+from ...todos import list_goal_todos
+from . import pr_lifecycle
+from .pr_lifecycle_rollout import append_pr_merge_rollout_event
+
+
+def reconcile_pr_wait_dependencies(
+    *, registry_path: Path, runtime_root_arg: str | None, goal_id: str,
+    agent_id: str, fetch_timeout_seconds: int = 10,
+) -> dict[str, Any]:
+    if agent_id not in registered_agent_ids_from_registry(registry_path, goal_id):
+        raise ValueError("PR dependency observation requires a registered agent")
+    runtime = (Path(runtime_root_arg).expanduser() if runtime_root_arg
+               else resolve_runtime_root(load_registry(registry_path), None))
+    # Deliberately omit list --agent-id and display limits. The typed owner must
+    # see archived/deferred/claimed facts before choosing this actor's subjects.
+    todos = list_goal_todos(registry_path=registry_path, runtime_root_arg=str(runtime),
+        goal_id=goal_id, role="agent")["todos"]
+    compact = compact_todo_resume_items(todos)
+    for row, original in zip(compact, todos, strict=True):
+        if "excluded_agents" in original:
+            row["excluded_agents"] = original["excluded_agents"]
+    log = rollout_event_log_path(runtime, goal_id)
+    plan = effect_runtime_result("todo.pr_wait_observation.plan", {
+        "schema_version": "todo_pr_wait_observation_request_v0", "agent_id": agent_id,
+        "generated_at": now_utc_iso(), "items": compact,
+        "rollout_events": [],
+        "rollout_event_source": {"runtime_root": str(runtime), "goal_id": goal_id},
+    })
+    results: list[dict[str, Any]] = []
+    for target in plan["targets"]:
+        reference = {"repo": target["repo"], "number": target["number"]}
+        result = {"pr_ref": target["pr_ref"], "todo_ids": target["todo_ids"],
+                  "state": "FAILED", "merge_event_recorded": False}
+        try:
+            metadata = pr_lifecycle.fetch_github_pr_lifecycle_payload(reference,
+                timeout_seconds=fetch_timeout_seconds, compact=True)
+            # A provider redirect or malformed success must never attest another
+            # repository's identically numbered dependency. No default OPEN.
+            if str(metadata.get("url") or "").lower().rstrip("/") != target["url"].lower():
+                raise ValueError("PR dependency provider identity mismatch")
+            state = str(metadata.get("state") or "").upper()
+            if state not in {"OPEN", "CLOSED", "MERGED"}:
+                raise ValueError("PR dependency provider state unavailable")
+            if state == "MERGED" and not metadata.get("mergedAt"):
+                raise ValueError("Merged PR observation lacks its terminal timestamp")
+            if state == "MERGED":
+                lifecycle = pr_lifecycle.build_issue_fix_pr_lifecycle_monitor_packet(
+                    repo=target["repo"], pr_ref=target["pr_ref"], url=target["url"],
+                    provider_payload=metadata)
+                result["merge_event_recorded"] = append_pr_merge_rollout_event(payload=lifecycle,
+                    goal_id=goal_id, registry_path=registry_path, runtime_root_arg=str(runtime))["recorded"]
+            result["state"] = state
+        except (ValueError, RuntimeError, OSError, TimeoutExpired) as error:
+            result["error_category"] = type(error).__name__
+        # Persist failed attempts too: restarts and repeated heartbeats respect
+        # the same cadence. This validation is not delivery/settlement evidence.
+        append_rollout_event(log, build_rollout_event(goal_id=goal_id,
+            event_kind="validation", agent_id=agent_id, pr_ref=target["pr_ref"],
+            classification=plan["observation_classification"], status=result["state"],
+            recorded_at=now_utc_iso(), summary="Observed exact PR wait dependency; no Todo transition."))
+        results.append(result)
+    failures = sum(row["state"] == "FAILED" for row in results)
+    observed = {row["pr_ref"]: row["state"] for row in results}
+    retained_failures = sum(observed.get(row["pr_ref"], row["last_observed_state"]) == "FAILED"
+                            for row in plan["waiting_targets"])
+    return {"ok": True, "degraded": retained_failures > 0 or bool(plan["unresolved"]),
+        "failure_count": failures, "external_read_count": len(results),
+        "unavailable_dependency_count": retained_failures,
+        "merged_count": sum(row["merge_event_recorded"] for row in results),
+        "closed_without_merge_count": sum(row["state"] == "CLOSED" for row in results),
+        "results": results, "plan": plan, "todo_write_performed": False}

loopx/control_plane/effect_runtime_handlers.ts

@@ -318,7 +318,8 @@ export function createEffectRuntimeHandlers(
     ["todo.next_action.transition", lazyHandler(() => import("./todos/next_action.ts"), ({transitionTodoNextAction}) => transitionTodoNextAction)],
     ["todo.next_action.binding", lazyHandler(() => import("./todos/next_action.ts"), ({projectNextActionBinding}) => projectNextActionBinding)],
     ["todo.resume_condition.normalize", lazyHandler(() => import("./todos/resume_condition.ts"), ({normalizeTodoResumeWhen}) => normalizeTodoResumeWhen)],
-    ["todo.resume_condition.evaluate", lazyHandler(() => import("./todos/resume_condition.ts"), ({evaluateTodoResumeConditions}) => evaluateTodoResumeConditions)],
+    ["todo.resume_condition.evaluate", lazyHandler(() => import("./todos/resume_condition.ts"), ({evaluateTodoResumeConditionsFromSource}) => evaluateTodoResumeConditionsFromSource)],
+    ["todo.pr_wait_observation.plan", lazyHandler(() => import("./todos/pr_wait_observation.ts"), ({planPrWaitObservationsFromSource}) => planPrWaitObservationsFromSource)],
     ["todo.resume_planning.project", lazyHandler(() => import("./todos/resume_planning.ts"), ({projectTodoResumePlanning}) => projectTodoResumePlanning)],
     ["todo.quota_planning.project", lazyHandler(() => import("./todos/quota_selection.ts"), ({projectTodoQuotaPlanning}) => projectTodoQuotaPlanning)],
     ["todo.frontier_revision.project", lazyHandler(() => import("./todos/frontier_revision.ts"), ({projectAdvancementFrontier}) => projectAdvancementFrontier)],

loopx/control_plane/todos/active_state_todos.py

@@ -11,6 +11,7 @@
 )
 
 from .succession_warning import public_todo_summary
+from .resume_condition import TodoResumeRolloutEvents
 from ...agent_registry import registered_agent_ids_for_goal
 from ..work_items.recommendation_source_io import (
     load_recommendation_source_goal, recommendation_source_context,
@@ -122,6 +123,8 @@ def active_state_todo_fields(
             rollout_event_log_path(runtime_root, goal_id),
             limit=max_todo_index_rollout_events_per_goal,
         )
+    if runtime_root is not None and goal_id:
+        events = TodoResumeRolloutEvents(events, runtime_root=runtime_root, goal_id=goal_id)
     if canonical is not None:
         fields = canonical_todo_summary_fields(
             canonical["todos"],

loopx/control_plane/todos/list_readback.py

@@ -33,6 +33,7 @@
     todo_list_projection_contract,
 )
 from .todo_index import MAX_TODO_INDEX_ROLLOUT_EVENTS_PER_GOAL
+from .resume_condition import TodoResumeRolloutEvents
 
 
 def _restore_full_source_body(
@@ -98,6 +99,8 @@ def list_goal_todos(
         limit=MAX_TODO_INDEX_ROLLOUT_EVENTS_PER_GOAL,
     )
 
+    rollout_events = TodoResumeRolloutEvents(rollout_events, runtime_root=runtime_root, goal_id=goal_id)
+
     roles = [role] if role else ["user", "agent"]
     canonical_read = read_canonical_todos_if_promoted(
         runtime_root=runtime_root,