jwbron · jwbron · Jul 3, 2026 · Jul 2, 2026 · Jul 2, 2026 · Jul 2, 2026
diff --git a/docs/architecture/orchestrator.md b/docs/architecture/orchestrator.md
@@ -1237,6 +1237,7 @@ if is_orchestrator_mode():
 | `EGG_ORCH_SLICE_GLOBAL_MAX_CYCLES` | Slice-DAG: pipeline-wide summed slice-cycle cap (#2137) | `10` |
 | `EGG_ORCH_SLICE_FAILURE_GRACE_SECONDS` | Slice-DAG: grace window before failure-cascade marks downstream subtree `BLOCKED_ON_FAILED_DEPENDENCY` (#2137) | `60.0` |
 | `EGG_ORCH_STACKED_PR_RECONCILER_INTERVAL_SECONDS` | Slice-DAG: stacked-PR reconciler polling cadence for orphaned child PRs (#2137) | `30.0` |
+| `EGG_ORCH_CROSS_REPO_MERGE_GATE_MAX_ATTEMPTS` | Slice-DAG: poll-attempt budget for the cross-repo merge-sequencing gate before a never-merging upstream PR escalates the dependent draft PR to a HITL hold (#3393); rides the stacked-PR reconciler's cadence | `240` |
 | `EGG_BRC_EVENT_PUMP` | **Removed** ([#2908](https://github.com/jwbron/egg/issues/2908)). This flag once selected a legacy capped-restart consensus wrapper vs. the event-pump wrapper; the legacy template and the env-var read are both gone, so the orchestrator no longer consults this variable. Operators that referenced it in helm values / pod-spec env can drop the row. | n/a (removed) |
 | `EGG_BRC_IDLE_BUDGET_MIN` | BRC idle / no-progress safety budget in minutes ([#2908](https://github.com/jwbron/egg/issues/2908)). The orchestrator event loop's convergence-stall check raises an overseer alert (anomaly `stuck-phase-transition`, priority `high`) at the budget threshold and escalates priority at `2 ×` budget, without transitioning the pipeline to FAILED. Default 30 min is well above the WS7-observed 10–13 min idle ceiling on real BRC phases. Read via `event_loop.get_idle_budget_minutes()` — see [Idle / no-progress safety budget](#idle--no-progress-safety-budget). | `30` |
 | `EGG_EVENT_LOOP_OWNER` | **Removed** ([#3164](https://github.com/jwbron/egg/issues/3164)). This flag once selected an in-pod wait-loop vs. the orchestrator-owned BRC event loop; the in-pod wait arm is gone and orchestrator ownership is the only mode, so the variable is no longer read (`get_event_loop_owner` is removed from `orchestrator/env_config.py`). Operators that set it in helm values / pod-spec env can drop the row. | n/a (removed) |

diff --git a/docs/architecture/slice-dag.md b/docs/architecture/slice-dag.md
@@ -823,23 +823,32 @@ stacked-PR reconciler cadence.
 
 - **Tier A — automated merge-state hold (default).** The dependent slice's PR
   opens as a **draft** while the upstream PR is unmerged. A bounded poll watches
-  the upstream PR's merge state and, on merge, auto-marks the dependent PR ready
-  via the `mark_pr_ready(repo, pr_number)` gateway verb (wrapping `gh pr
-  ready`). **Merge detection keys off the PR `mergedAt` / merged boolean, not
-  head-SHA equality** — a squash or rebase merge produces a merge-commit SHA ≠
-  the PR head, so SHA-equality would misfire. Two failure terminals fall through
-  to a HITL hold rather than hanging: an upstream that reaches
-  **CLOSED-not-merged**, and a poll that exceeds its **attempt bound** (a
-  never-merging upstream). Both surface on pipeline status.
+  the upstream PR's merge state via the orchestrator-only gateway routes
+  `POST /api/v1/gh/pr/merge_state` (read) and `POST /api/v1/gh/pr/ready` (write)
+  — see [Gateway README](../../gateway/README.md) — and, on merge, auto-marks
+  the dependent PR ready via the `mark_pr_ready(repo, pr_number)` gateway verb
+  (wrapping `gh pr ready`). **Merge detection keys off the PR `mergedAt` /
+  merged boolean, not head-SHA equality** — a squash or rebase merge produces a
+  merge-commit SHA ≠ the PR head, so SHA-equality would misfire. Two failure
+  terminals fall through to a HITL hold rather than hanging: an upstream that
+  reaches **CLOSED-not-merged**, and a poll that exceeds its **attempt bound**
+  (`EGG_ORCH_CROSS_REPO_MERGE_GATE_MAX_ATTEMPTS`, default 240 ticks, ~2h at the
+  default 30s reconciler cadence — a never-merging upstream). Both surface on
+  pipeline status.
 - **Tier B — HITL beyond-merge-state hold (opt-in).** For an edge the plan (or
   task description) marks with a beyond-merge-state condition — a
   release/publish of the upstream repo, a version-pin choice, or a genuine
-  cannot-continue development block — the dependent PR is held and released
-  **only by a HITL decision**, never by programmatic detection. Absent that
-  marker, a cross-repo edge defaults to the Tier-A automated hold.
+  cannot-continue development block — the dependent slice opts in via the
+  `[hold:beyond-merge-state]` marker in its `goal` (or a task description); the
+  dependent PR is then held and released **only by a HITL decision**, never by
+  programmatic detection. Absent that marker, a cross-repo edge defaults to the
+  Tier-A automated hold.
 
 All HITL holds (Tier B, plus the two Tier-A failure terminals) route through the
-same decision-queue mechanism and share a single release path.
+same decision-queue mechanism and share a single release path — each registers a
+HITL Decision offering two operator-selectable options: release the hold (marks
+the PR ready) or keep it held (terminal; the PR stays draft for manual
+handling).
 
 ### Per-repo gate, diff & convention scoping
 
@@ -958,6 +967,7 @@ on parse failure. The green-gate knobs below are read directly via
 | `EGG_ORCH_SLICE_GLOBAL_MAX_CYCLES` | int | 10 | Pipeline-wide summed slice-cycle cap. *Currently inert — see local cycles row.* |
 | `EGG_ORCH_SLICE_FAILURE_GRACE_SECONDS` | float | 60.0 | Grace window before a failure cascade marks the downstream subtree `BLOCKED_ON_FAILED_DEPENDENCY`. |
 | `EGG_ORCH_STACKED_PR_RECONCILER_INTERVAL_SECONDS` | float | 30.0 | Reconciler polling cadence for orphaned child PRs. |
+| `EGG_ORCH_CROSS_REPO_MERGE_GATE_MAX_ATTEMPTS` | int | 240 | Poll-attempt budget for the cross-repo merge-sequencing gate (#3393) before a never-merging upstream escalates to a HITL hold; ~2h at the default reconciler cadence. See [Cross-repo merge-sequencing hold](#cross-repo-merge-sequencing-hold-two-tier). |
 | `EGG_SLICE_GREEN_GATE` | str | `off` | Per-slice green gate rollout switch (#3398): `off` skips the gate entirely; `log` runs the repo's configured checks at the slice tip and logs a red verdict without blocking; `on` blocks slice PR-open on a red verdict. Case-insensitive, with aliases — `on` also accepts `1`/`true`/`yes`, and `log` also accepts `log-only`/`log_only`. Unknown values resolve to `off`. |
 | `EGG_SLICE_GREEN_GATE_SKIP_CHECKS` | str (comma-separated) | `security` | Configured check *names* (from `repositories.yaml` `checks`) the gate skips. |
 | `EGG_SLICE_GREEN_GATE_TIMEOUT_SECONDS` | int | 1800 | Wall-clock budget for the check-runner pod (spawn-to-terminal); a hung suite degrades to fail-open rather than wedging the slice close. |

diff --git a/docs/development/STRUCTURE.md b/docs/development/STRUCTURE.md
@@ -109,6 +109,7 @@ orchestrator/
 ├── concurrent_executor.py  # Concurrent phase executor (spawns all agents simultaneously)
 ├── slice_scheduler.py      # Wave-based scheduler for the implement-phase slice DAG: computes execution waves, caps concurrency, two-tier max_cycles accounting, failure-cascade detection (#2137)
 ├── stacked_pr_reconciler.py # Stacked-PR rebase reconciler: detects child slice PRs whose base branch was deleted after a parent merge and retargets them via gateway rebase_onto (#2137)
+├── cross_repo_merge_gate.py # Cross-repo merge-sequencing gate for multi-repo pipelines: auto-readies (or HITL-holds) a dependent slice's draft PR once its cross-repo upstream PR merges; rides the stacked-PR reconciler cadence (#3393 slice-5)
 ├── slice_green_gate.py     # Per-slice green gate: sandboxed one-shot Job runs the repo's configured checks at the integration-branch tip and blocks PR-open on red; staged rollout via EGG_SLICE_GREEN_GATE (off/log/on), fail-open on infra errors (#3398)
 ├── action_guards.py        # Formal BRC state machine action guards (preconditions for propose/ack/nack/confirm/withdraw)
 ├── approval_matrix.py      # Per-reviewer ACK/NACK matrix for BRC consensus

diff --git a/docs/reference/orchestrator-cli.md b/docs/reference/orchestrator-cli.md
@@ -92,6 +92,7 @@ Agent role can be omitted when `EGG_AGENT_ROLE` is set.
 | `EGG_ORCH_SLICE_GLOBAL_MAX_CYCLES` | Pipeline-wide cap on the summed total of slice re-proposal cycles. Default `10`. Either the local or global cap tripping escalates to HITL. *API live, not yet wired in the run loop — see [Slice-DAG Implement Phase](../architecture/slice-dag.md); #2199.* |
 | `EGG_ORCH_SLICE_FAILURE_GRACE_SECONDS` | Grace window (seconds) between a slice failure and the orchestrator marking the downstream subtree `BLOCKED_ON_FAILED_DEPENDENCY`. Default `60`. Allows HITL resolution before the cascade fires. |
 | `EGG_ORCH_STACKED_PR_RECONCILER_INTERVAL_SECONDS` | Polling cadence (seconds) of the stacked-PR reconciler that detects child slice PRs whose base branch was deleted after a parent merge. Default `30`. |
+| `EGG_ORCH_CROSS_REPO_MERGE_GATE_MAX_ATTEMPTS` | Poll-attempt budget for the cross-repo merge-sequencing gate (#3393), which rides the stacked-PR reconciler's cadence. A never-merging upstream PR that exceeds this many reconcile ticks escalates the dependent draft PR to a HITL hold instead of waiting forever. Default `240` (~2h at the default 30s reconciler interval). See [Slice-DAG Implement Phase](../architecture/slice-dag.md). |
 | `AGENT_ANCHOR_ID` | Agent anchor ID (`{role}-{short_container_id}`), auto-set by container spawner |
 | `EGG_LIFECYCLE_SECRET` | Bearer token required for lifecycle-control endpoints (HITL resolve/cancel, pipeline CRUD, phase overrides, container spawn/stop). Stored at `~/.config/egg/lifecycle-secret`. Must be exported in the human's shell to run `egg-orch decision resolve`, `egg-orch pipeline delete`, etc. Agent pods never receive it (see #1769). |
 

diff --git a/gateway/README.md b/gateway/README.md
@@ -276,6 +276,22 @@ POST /api/v1/gh/pr/close
   Request: {repo, pr_number}
   Policy: pr_ownership
 
+POST /api/v1/gh/pr/merge_state
+  Request: {repo, pr_number}
+  Auth: Bearer {launcher_secret}
+  Description: Orchestrator-only read of a PR's merge state ({state, mergedAt}).
+               Powers the cross-repo merge-sequencing gate (#3393): the
+               orchestrator polls an upstream slice PR to decide when to
+               mark a downstream draft PR ready. Keys off mergedAt/state,
+               not head-SHA equality (a squash/rebase merge changes the SHA).
+
+POST /api/v1/gh/pr/ready
+  Request: {repo, pr_number}
+  Auth: Bearer {launcher_secret}
+  Description: Orchestrator-only draft→ready transition (wraps `gh pr ready`).
+               Write half of the cross-repo merge-sequencing gate (#3393):
+               called when the upstream PR merges.
+
 POST /api/v1/gh/execute
   Request: {args[], require_auth}
   Policy: deny-by-default allowlist (parity with /api/v1/git/execute).