[codex] Fix stale gates and unbound PR visibility

README.md

@@ -11,11 +11,11 @@ integration PRs, follows `main` through production, and pauses after failures.
 
 ## Install
 
-Install the reviewed `v0.2.19` source commit directly from GitHub:
+Install the reviewed `v0.2.20` source commit directly from GitHub:
 
 ```bash
 python3 -m pip install \
-  'deploybot-merge-queue[mcp] @ git+https://github.com/Forward-Future/DeployBot.git@1b6379a258a0b6f743ce77c2d108dfd7e83d582b'
+  'deploybot-merge-queue[mcp] @ git+https://github.com/Forward-Future/DeployBot.git@3bf238140fba3d4d5fbd2d739b9f5422e99567dd'
 deploybot init
 ```
 
@@ -95,7 +95,7 @@ worker can dispatch deployment when GitHub suppresses the `workflow_run` event
 for token-dispatched CI. Pin the Action to the full reviewed release commit:
 
 ```yaml
-- uses: Forward-Future/DeployBot@1b6379a258a0b6f743ce77c2d108dfd7e83d582b
+- uses: Forward-Future/DeployBot@3bf238140fba3d4d5fbd2d739b9f5422e99567dd
 ```
 
 The Action uses GitHub's built-in workflow token. GitHub intentionally does not
@@ -160,9 +160,12 @@ they cannot create the original per-pull-request deploy intent.
 `deploybot status` reports active metadata-only agent threads, pending native
 notifications, every PR stage, deploy requests and their exact authorized heads,
 queue order, queued and pre-queue intent overlaps, exact-`main` CI, deployment,
-and pipeline pause state. It alerts when a deploy request exceeds the configured
+pipeline pause state, and every open PR that has not been bound to its native
+opening thread. It alerts when a deploy request exceeds the configured
 ready-to-merge target and names the current gate. It never stores prompts,
-transcripts, source, or credentials.
+transcripts, source, or credentials. A failed PR rollup is reconciled against
+the exact commit's check runs before DeployBot creates a repair block, so a
+cancelled superseded run cannot hide its replacement.
 
 `deploybot react` promotes ready intent, skips blockers, drains independent
 work, and creates integration PRs when configured. New batches contain at most

adapters/claude-code/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "deploybot",
-  "version": "0.2.19",
+  "version": "0.2.20",
   "description": "DeployBot: a provider-neutral GitHub merge queue for coding agents",
   "author": {
     "name": "DeployBot contributors"

adapters/claude-code/.mcp.json

@@ -4,7 +4,7 @@
       "command": "uvx",
       "args": [
         "--from",
-        "deploybot-merge-queue[mcp] @ git+https://github.com/Forward-Future/DeployBot.git@1b6379a258a0b6f743ce77c2d108dfd7e83d582b",
+        "deploybot-merge-queue[mcp] @ git+https://github.com/Forward-Future/DeployBot.git@3bf238140fba3d4d5fbd2d739b9f5422e99567dd",
         "deploybot-mcp"
       ]
     }

adapters/claude-code/skills/deploybot/SKILL.md

@@ -49,6 +49,11 @@ the destination for repair handoffs and the final deployment receipt. A later
 deploy, repair, integration, or coordinator thread must never claim ownership
 of an already-open PR.
 
+Before the PR-opening thread finishes its response, call `pipeline_status` and
+confirm that the exact PR appears in `pull_request_thread_owners` and not in
+`unbound_pull_requests`. A missing binding means the PR-opening task is still
+incomplete: publish the binding from this thread and verify it before stopping.
+
 Require the user's exact `deploy` instruction before calling
 `request_deployment` or `deploybot request` for that conversation's pull
 request. DeployBot resolves the previously recorded PR-opening thread; never

adapters/claude-code/skills/manage-merge-queue/SKILL.md

@@ -13,6 +13,10 @@ Immediately after opening the PR, call `update_agent_thread` in `pr-review`
 phase with its number. That first binding is immutable and owns repair handoffs
 and the final deployment receipt.
 
+Before the PR-opening response finishes, call `pipeline_status` and verify that
+the exact PR is in `pull_request_thread_owners`, not
+`unbound_pull_requests`. Bind it from this thread and recheck if missing.
+
 Only the user's exact `deploy` instruction authorizes `request_deployment` for
 this thread's PR. DeployBot uses the recorded opening thread; a coordinator
 must never substitute its own ID. If review fixes change

adapters/codex/agent-merge-queue/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "deploybot",
-  "version": "0.2.19",
+  "version": "0.2.20",
   "description": "Coordinate exact-head pull requests through verified deployment and thread notification",
   "author": {
     "name": "DeployBot contributors"

adapters/codex/agent-merge-queue/skills/deploybot/SKILL.md

@@ -44,6 +44,12 @@ update` in `pr-draft`, `pr-review`, or `ready` phase with the PR number. The
 first trusted binding is immutable and owns repair handoffs and the final
 deployment receipt. Later deploy and coordinator threads never replace it.
 
+Before this PR-opening thread finishes its response, run `deploybot status
+--json` and confirm that the exact PR appears in
+`pull_request_thread_owners` and not in `unbound_pull_requests`. A missing
+binding means the PR-opening task is still incomplete: publish the binding from
+this thread and verify it before stopping.
+
 Require the user's exact `deploy` instruction before running `deploybot request`
 for that conversation's pull request. DeployBot resolves the recorded
 PR-opening thread; never substitute the current coordinator's thread ID. If an

adapters/codex/agent-merge-queue/skills/manage-merge-queue/SKILL.md

@@ -13,6 +13,10 @@ Immediately after opening the PR, run `deploybot thread update --provider codex
 --thread-id <stable-id> --phase pr-review --pr <number>`. That first binding is
 immutable and owns repair handoffs and the final deployment receipt.
 
+Before this PR-opening response finishes, run `deploybot status --json` and
+verify that the exact PR is in `pull_request_thread_owners`, not
+`unbound_pull_requests`. Bind it from this thread and recheck if missing.
+
 Only the user's exact `deploy` instruction authorizes `deploybot request` for
 this thread's PR. DeployBot uses the recorded opening thread; a coordinator
 must never substitute its own ID. If review fixes change the

adapters/cursor/.cursor/mcp.json

@@ -4,7 +4,7 @@
       "command": "uvx",
       "args": [
         "--from",
-        "deploybot-merge-queue[mcp] @ git+https://github.com/Forward-Future/DeployBot.git@1b6379a258a0b6f743ce77c2d108dfd7e83d582b",
+        "deploybot-merge-queue[mcp] @ git+https://github.com/Forward-Future/DeployBot.git@3bf238140fba3d4d5fbd2d739b9f5422e99567dd",
         "deploybot-mcp"
       ]
     }

docs/reference.md

@@ -1,7 +1,7 @@
 # DeployBot reference
 
 This reference describes the CLI, MCP server, policy file, and GitHub Action in
-DeployBot v0.2.19. GitHub labels and authenticated comments are the durable state;
+DeployBot v0.2.20. GitHub labels and authenticated comments are the durable state;
 the CLI and MCP tools are two interfaces to the same operations.
 
 ## CLI
@@ -23,7 +23,7 @@ DeployBot resolves the pull request for the current branch.
 | `deploybot init [--force]` | Write a safe starter policy. Existing files are preserved unless `--force` is supplied. |
 | `deploybot ensure-labels` | Create or refresh the configured queue, blocked, intent, pause, and registry labels. |
 | `deploybot doctor [--json]` | Check authentication, policy, labels, actors, checks, workflows, and branch protection without changing repository state. |
-| `deploybot status [--json]` | Read active thread metadata, pending native notifications, PR stages, exact-head deploy intent, pre-queue intent overlaps, request-stage timing alerts, queue state, exact-main CI, deployment, and pipeline control state. |
+| `deploybot status [--json]` | Read active thread metadata, unbound open PRs, pending native notifications, PR stages, exact-head deploy intent, pre-queue intent overlaps, request-stage timing alerts, queue state, exact-main CI, deployment, and pipeline control state. |
 | `deploybot plan [--json]` | Read the ordered queue, dependencies, blockers, and source-overlap groups. |
 | `deploybot inspect [PR] [--json]` | Evaluate one exact PR head without granting merge authority. |
 | `deploybot metrics [--limit N] [--json]` | Summarize p50, p95, and maximum delivery timings for recent merged PRs. The default limit is 25. |

examples/github-workflow.yml

@@ -77,5 +77,5 @@ jobs:
         with:
           ref: ${{ github.event.repository.default_branch }}
           persist-credentials: false
-      # v0.2.19 implementation; keep the full commit for privileged workflows.
-      - uses: Forward-Future/DeployBot@1b6379a258a0b6f743ce77c2d108dfd7e83d582b
+      # v0.2.20 implementation; keep the full commit for privileged workflows.
+      - uses: Forward-Future/DeployBot@3bf238140fba3d4d5fbd2d739b9f5422e99567dd

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "deploybot-merge-queue"
-version = "0.2.19"
+version = "0.2.20"
 description = "DeployBot: a provider-neutral GitHub merge queue for coding agents"
 readme = "README.md"
 license = "MIT"

skills/deploybot/SKILL.md

@@ -49,6 +49,11 @@ the destination for repair handoffs and the final deployment receipt. A later
 deploy, repair, integration, or coordinator thread must never claim ownership
 of an already-open PR.
 
+Before the PR-opening thread finishes its response, call `pipeline_status` and
+confirm that the exact PR appears in `pull_request_thread_owners` and not in
+`unbound_pull_requests`. A missing binding means the PR-opening task is still
+incomplete: publish the binding from this thread and verify it before stopping.
+
 Require the user's exact `deploy` instruction before calling
 `request_deployment` or `deploybot request` for that conversation's pull
 request. DeployBot resolves the previously recorded PR-opening thread; never

skills/manage-merge-queue/SKILL.md

@@ -19,6 +19,9 @@ the repository's policy; never assume a particular review vendor.
 6. Immediately after opening the PR, call `update_agent_thread` in `pr-draft`,
    `pr-review`, or `ready` phase with its number so this opening thread becomes
    the immutable repair and deployment-receipt destination.
+7. Before finishing the PR-opening response, call `pipeline_status` and verify
+   that the exact PR is in `pull_request_thread_owners`, not
+   `unbound_pull_requests`. Bind and recheck it if missing.
 
 Do not merge merely because review is complete.
 

src/agent_merge_queue/__init__.py

@@ -1,3 +1,3 @@
 """DeployBot: a provider-neutral GitHub merge queue for coding agents."""
 
-__version__ = "0.2.19"
+__version__ = "0.2.20"

src/agent_merge_queue/cli.py

@@ -2344,9 +2344,22 @@ def snapshot(
             INTEGRATION_MARKER,
             coordinator_logins(self),
         )
-        if integration and any(
-            checks.get(name) != "passed" for name in self.config.required_checks
-        ):
+        # GitHub's PR rollup can briefly retain a cancelled or failed run after
+        # a replacement run for the same exact head has started. Reconcile a
+        # reported failure with the commit check-runs endpoint before creating
+        # a repair block. Integration PRs keep the broader fallback because
+        # their controller-dispatched checks may be absent from the rollup.
+        rollup_has_required_failure = any(
+            checks.get(name) == "failed" for name in self.config.required_checks
+        )
+        integration_needs_exact_checks = bool(
+            integration
+            and any(
+                checks.get(name) != "passed"
+                for name in self.config.required_checks
+            )
+        )
+        if rollup_has_required_failure or integration_needs_exact_checks:
             checks = merge_known_check_states(
                 check_states(check_rollup + self.commit_check_runs(head_sha)),
                 known_checks,
@@ -2910,6 +2923,18 @@ def inspect(
             for _, value in sorted(thread_owners.items())
             if value.pull_request in open_number_set
         ],
+        "unbound_pull_requests": [
+            {
+                "pull_request": value["number"],
+                "pipeline_stage": stage,
+                "head_sha": value["head_sha"],
+                "title": value["title"],
+                "url": value["url"],
+            }
+            for stage, values in stages.items()
+            for value in values
+            if value["number"] not in thread_owners
+        ],
         "notifications": client.deployment_notifications(),
         "pull_requests": stages,
         "queue": [
@@ -2934,6 +2959,7 @@ def print_pipeline_status(value: dict[str, Any], *, json_output: bool) -> None:
         "threads: "
         f"{len(value['threads'])} active; "
         f"notifications: {len(value.get('notifications') or [])} pending; "
+        f"unbound PRs: {len(value.get('unbound_pull_requests') or [])}; "
         "deploy requests: "
         f"{sum(1 for entries in stages.values() for entry in entries if entry.get('deploy_intent'))}; "
         f"queue: {len(value['queue'])}; "

tests/test_cli.py

@@ -492,6 +492,13 @@ def test_status_exposes_intent_head_overlap_and_request_delay(self) -> None:
             result["pull_request_thread_owners"][0]["thread_id"],
             "opening-thread",
         )
+        self.assertEqual(
+            {
+                (value["pull_request"], value["pipeline_stage"])
+                for value in result["unbound_pull_requests"]
+            },
+            {(2, "blocked"), (3, "ready")},
+        )
 
     def test_overlap_mode_holds_only_ready_members_of_near_ready_components(
         self,
@@ -1597,6 +1604,64 @@ def test_integration_snapshot_uses_exact_commit_check_fallback(self) -> None:
         self.assertEqual(value.checks["CI"], "passed")
         client.commit_check_runs.assert_called_once_with(head_sha)
 
+    def test_failed_rollup_uses_exact_head_replacement_check(self) -> None:
+        head_sha = "a" * 40
+        client = object.__new__(GitHub)
+        client.config = CONFIG
+        client.repository = "example/repo"
+        client.trusted_logins = {"trusted"}
+        client.coordinator_logins = {"trusted"}
+        client.comments = Mock(return_value=[])
+        client.changed_paths = Mock(return_value=([], []))
+        client.commit_check_runs = Mock(
+            return_value=[
+                {
+                    "name": "CI",
+                    "conclusion": "cancelled",
+                    "started_at": "2026-06-20T00:00:00Z",
+                },
+                {
+                    "name": "CI",
+                    "status": "in_progress",
+                    "started_at": "2026-06-20T00:01:00Z",
+                },
+            ]
+        )
+        client._json = Mock(
+            return_value={
+                "baseRefName": "main",
+                "body": "",
+                "headRefOid": head_sha,
+                "isDraft": False,
+                "labels": [],
+                "mergeStateStatus": "UNSTABLE",
+                "mergeable": "MERGEABLE",
+                "number": 39,
+                "state": "OPEN",
+                "statusCheckRollup": [
+                    {
+                        "__typename": "CheckRun",
+                        "name": "CI",
+                        "conclusion": "CANCELLED",
+                        "startedAt": "2026-06-20T00:00:00Z",
+                    }
+                ],
+                "title": "Replacement check is running",
+                "url": "https://example.test/39",
+            }
+        )
+
+        value = client.snapshot(
+            39,
+            require_marker=False,
+            allow_blocked_label=True,
+        )
+
+        self.assertEqual(value.state, "waiting")
+        self.assertEqual(value.checks["CI"], "pending")
+        self.assertNotIn("CI failed", value.reasons)
+        client.commit_check_runs.assert_called_once_with(head_sha)
+
     def test_required_checks_accept_passing_terminal_conclusions(self) -> None:
         states = check_states(
             [

tests/test_skill.py

@@ -8,7 +8,7 @@
 
 ROOT = Path(__file__).resolve().parents[1]
 CANONICAL = ROOT / "skills" / "deploybot" / "SKILL.md"
-RELEASE_COMMIT = "1b6379a258a0b6f743ce77c2d108dfd7e83d582b"
+RELEASE_COMMIT = "3bf238140fba3d4d5fbd2d739b9f5422e99567dd"
 CHECKOUT_COMMIT = "9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0"
 
 
@@ -81,6 +81,8 @@ def test_status_guidance_is_read_only(self) -> None:
         self.assertIn("acknowledge_thread_deployment", skill)
         self.assertIn("heartbeat automation", skill)
         self.assertIn("notification_handoff.required_action", skill)
+        self.assertIn("unbound_pull_requests", skill)
+        self.assertIn("pull_request_thread_owners", skill)
         self.assertIn("human-facing release receipt", skill)
         self.assertIn("acknowledge silently", skill)
         self.assertIn("untrusted display-only", skill)