fix(scheduler): skip central target Actions inventory

CHANGELOG.md

@@ -162,6 +162,11 @@
 
 ## Proposed
 
+- Skip target-repository old-head Actions inventory when review execution is
+  centralized. Same-repository stale-run cleanup remains enabled; central
+  review lifecycle is handled in the configured dispatch repository, avoiding
+  an unauthoritative API read that can exhaust the cross-repository App quota.
+
 - Run Python Security and Agent Review Runtime Quality CI for stacked pull
   requests by removing their pull-request base-branch filters. Extend the
   permanent stacked-workflow contract so all four owner review workflows

docs/doctoring/central-review-target-inventory-suppression.md

@@ -0,0 +1,35 @@
+# Central review target-inventory suppression
+
+Decision date: **2026-09-07**
+
+## Problem
+
+When the trusted reviewer is hosted centrally, target-repository old-head
+workflow runs are not the authority for the central current-head verdict.
+Enumerating those target runs before dispatch consumes the cross-repository
+Actions credential and can exhaust its App quota before useful review work
+starts.
+
+## Decision
+
+Compare the configured review dispatch repository with the target repository.
+If they differ, do not enumerate or cancel target old-head runs from this
+decision path. The central reviewer owns its own run lifecycle in the dispatch
+repository. If they are the same repository, retain existing stale-run cleanup.
+
+## Failure scenes
+
+- Central review of a target repository: no target Actions inventory read occurs.
+- Same-repository review: stale old-head runs are still cancelled.
+- Repository name casing differs: case-insensitive identity prevents accidental
+  cross-repository classification.
+
+## Evidence and follow-up
+
+RED commit: `08a16caa4fdb0d0d86c44bb8cd7aed611beaab7b`.
+Fresh exact-head hosted checks and independent review remain required.
+
+## Reference
+
+GitHub. (2026). *REST API endpoints for workflow runs*.
+https://docs.github.com/en/rest/actions/workflow-runs

docs/product-technical-gap-baseline.md

@@ -3401,3 +3401,18 @@ same name in another file can carry the opposite safety property.**
   workflows at exact head `e2204eeb1ec2789ff791036140ba1672995d25f5`;
   RED commit `890bac2f69ff1a51f774ddf5d6c5d819afed4ac9`; fresh exact-head
   hosted checks remain required.
+
+
+### Central review target-inventory suppression
+
+- **Status:** Proposed
+- **Owner:** `ContextualWisdomLab/.github`
+- **Problem:** Before dispatching a central current-head review, the scheduler
+  enumerated target-repository old-head Actions runs that are not central
+  admission authority, spending the cross-repository App quota.
+- **Action:** Skip only that target enumeration when the configured review
+  dispatch repository differs from the target; preserve same-repository
+  stale-run cleanup.
+- **Evidence:** RED commit
+  `08a16caa4fdb0d0d86c44bb8cd7aed611beaab7b`; fresh exact-head hosted checks
+  remain required before integration.

scripts/ci/pr_review_merge_scheduler_core.py

@@ -4270,7 +4270,11 @@ def inspect_pr(
                 pass
             run(["gh", "pr", "close", str(number), "--repo", repo])
         return Decision(number, "close_empty", "base 대비 실제 변경 0건")
-    cancel_stale_pr_runs(repo, pr, dry_run=dry_run)
+    # A central reviewer owns run lifecycle in its dispatch repository.
+    # Target old-head runs are not admission authority, and enumerating them
+    # spends the cross-repository installation quota before current-head review.
+    if repository_dispatch_target(repo).casefold() == repo.casefold():
+        cancel_stale_pr_runs(repo, pr, dry_run=dry_run)
     if base_ref != base_branch:
         # Stacked/cascade PR (base is another feature branch). Org required
         # workflows are only injected for default-branch-target PRs, so these

tests/test_pr_review_merge_scheduler.py

@@ -10866,3 +10866,24 @@ def test_withheld_mutation_guidance_uses_recorded_reason_after_environment_chang
     assert "workflow GITHUB_TOKEN" in "\n".join(
         sched.head_mutation_credential_upgrade_summary([decision])
     )
+
+
+def test_central_dispatch_skips_non_authoritative_target_actions_inventory(
+    monkeypatch,
+):
+    """Central review dispatch must not spend App quota on target old-head runs."""
+    monkeypatch.setenv(
+        "SCHEDULER_REQUIRED_WORKFLOW_REPOSITORY",
+        "ContextualWisdomLab/.github",
+    )
+    monkeypatch.setattr(
+        sched,
+        "cancel_stale_pr_runs",
+        lambda *args, **kwargs: pytest.fail(
+            "central dispatch must not enumerate target Actions runs"
+        ),
+    )
+
+    decision = inspect(make_pr(baseRefName="feature-base"), trigger_reviews=False)
+
+    assert decision.action == "skip"