fix(attach): isolate registration stalls and back off on database overload

.agentworkforce/trajectories/compacted/compact_msichwe8jqzl_2026-09-08.json

@@ -0,0 +1,37 @@
+{
+  "id": "compact_msichwe8jqzl",
+  "version": 1,
+  "type": "compacted",
+  "compactedAt": "2026-09-08T10:22:43.328Z",
+  "sourceTrajectories": [
+    "traj_pg9aglu4aqlw"
+  ],
+  "dateRange": {
+    "start": "2026-09-08T10:04:45.586Z",
+    "end": "2026-09-08T10:21:50.230Z"
+  },
+  "summary": {
+    "totalDecisions": 1,
+    "totalEvents": 1,
+    "uniqueAgents": [
+      "default"
+    ]
+  },
+  "decisionGroups": [
+    {
+      "category": "api",
+      "decisions": [
+        {
+          "question": "Move HTTP-spawn cloud preparation out of the broker actor and bound it",
+          "chosen": "Move HTTP-spawn cloud preparation out of the broker actor and bound it",
+          "reasoning": "Wrangler traced a 35.7s duplicate registration blocking local attach. Preserve node tokens, merge only completed identity/cursor state, reserve local names, reject pending releases, and skip launch when the caller is gone. Retry only explicit pre-allocation D1 overload for fleet attach.",
+          "fromTrajectory": "traj_pg9aglu4aqlw"
+        }
+      ]
+    }
+  ],
+  "keyLearnings": [],
+  "keyFindings": [],
+  "filesAffected": [],
+  "commits": []
+}

.agentworkforce/trajectories/compacted/compact_msichwe8jqzl_2026-09-08.md

@@ -0,0 +1,18 @@
+# Trajectory Compaction: Sep 8, 2026 - Sep 8, 2026
+
+## Summary
+- Sessions: 1
+- Decisions: 1
+- Events: 1
+- Agents: default
+- Files: 0
+- Commits: 0
+
+## Api
+- Move HTTP-spawn cloud preparation out of the broker actor and bound it -> Move HTTP-spawn cloud preparation out of the broker actor and bound it (traj_pg9aglu4aqlw)
+
+## Key Learnings
+- None
+
+## Key Findings
+- None

.agentworkforce/trajectories/completed/2026-09/traj_pg9aglu4aqlw/compaction.json

@@ -0,0 +1,5 @@
+{
+  "trajectoryId": "traj_pg9aglu4aqlw",
+  "compactedInto": "compact_msichwe8jqzl",
+  "compactedAt": "2026-09-08T10:22:43.428Z"
+}

.agentworkforce/trajectories/completed/2026-09/traj_pg9aglu4aqlw/summary.md

@@ -0,0 +1,31 @@
+# Trajectory: Keep local attach responsive during slow cloud registration
+
+> **Status:** ✅ Completed
+> **Confidence:** 90%
+> **Started:** September 8, 2026 at 12:04 PM
+> **Completed:** September 8, 2026 at 12:21 PM
+
+---
+
+## Summary
+
+Isolated HTTP-spawn cloud registration from local attach; bounded registration and overload retries; accurate health/attach diagnostics. 1065 broker tests passed (4 ignored), 43 attach tests passed, CLI typecheck passed. Same real-pipe regression fails on baseline 11.10.4 and passes with fix. Production broker untouched; rollout pending.
+
+**Approach:** Standard approach
+
+---
+
+## Key Decisions
+
+### Move HTTP-spawn cloud preparation out of the broker actor and bound it
+- **Chose:** Move HTTP-spawn cloud preparation out of the broker actor and bound it
+- **Reasoning:** Wrangler traced a 35.7s duplicate registration blocking local attach. Preserve node tokens, merge only completed identity/cursor state, reserve local names, reject pending releases, and skip launch when the caller is gone. Retry only explicit pre-allocation D1 overload for fleet attach.
+
+---
+
+## Chapters
+
+### 1. Work
+*Agent: default*
+
+- Move HTTP-spawn cloud preparation out of the broker actor and bound it: Move HTTP-spawn cloud preparation out of the broker actor and bound it

.agentworkforce/trajectories/completed/2026-09/traj_pg9aglu4aqlw/trajectory.json

@@ -0,0 +1,53 @@
+{
+  "id": "traj_pg9aglu4aqlw",
+  "version": 1,
+  "task": {
+    "title": "Keep local attach responsive during slow cloud registration"
+  },
+  "status": "completed",
+  "startedAt": "2026-09-08T10:04:45.586Z",
+  "completedAt": "2026-09-08T10:21:50.230Z",
+  "agents": [
+    {
+      "name": "default",
+      "role": "lead",
+      "joinedAt": "2026-09-08T10:16:44.611Z"
+    }
+  ],
+  "chapters": [
+    {
+      "id": "chap_k90zf0m041ko",
+      "title": "Work",
+      "agentName": "default",
+      "startedAt": "2026-09-08T10:16:44.611Z",
+      "endedAt": "2026-09-08T10:21:50.230Z",
+      "events": [
+        {
+          "ts": 1788862604611,
+          "type": "decision",
+          "content": "Move HTTP-spawn cloud preparation out of the broker actor and bound it: Move HTTP-spawn cloud preparation out of the broker actor and bound it",
+          "raw": {
+            "question": "Move HTTP-spawn cloud preparation out of the broker actor and bound it",
+            "chosen": "Move HTTP-spawn cloud preparation out of the broker actor and bound it",
+            "alternatives": [],
+            "reasoning": "Wrangler traced a 35.7s duplicate registration blocking local attach. Preserve node tokens, merge only completed identity/cursor state, reserve local names, reject pending releases, and skip launch when the caller is gone. Retry only explicit pre-allocation D1 overload for fleet attach."
+          },
+          "significance": "high"
+        }
+      ]
+    }
+  ],
+  "retrospective": {
+    "summary": "Isolated HTTP-spawn cloud registration from local attach; bounded registration and overload retries; accurate health/attach diagnostics. 1065 broker tests passed (4 ignored), 43 attach tests passed, CLI typecheck passed. Same real-pipe regression fails on baseline 11.10.4 and passes with fix. Production broker untouched; rollout pending.",
+    "approach": "Standard approach",
+    "confidence": 0.9
+  },
+  "commits": [],
+  "filesChanged": [],
+  "projectId": "AgentWorkforce/relay",
+  "tags": [],
+  "_trace": {
+    "startRef": "d754fff143c464c367ac743ccc3c8085bf5ef04d",
+    "endRef": "d754fff143c464c367ac743ccc3c8085bf5ef04d"
+  }
+}

CHANGELOG.md

@@ -5,7 +5,13 @@ All notable changes to Agent Relay will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [Unreleased]
+## [Unreleased - Patch]
+
+### Fixed
+
+- Local agent attach, snapshots, and input remain responsive while HTTP spawns wait for cloud registration; registration now has bounded concurrency and an overall deadline.
+- Fleet attach retries explicit database-overload rejections with server-directed backoff, and attach lookup failures explain that the agent may still be running.
+- Broker health marks runtime status as unavailable instead of claiming zero agents when its status lookup times out.
 
 ## [11.10.4] - 2026-09-08
 

crates/broker/Cargo.toml

@@ -47,6 +47,7 @@ nix = { version = "0.30", features = ["signal", "process", "term", "fs"] }
 libc = "0.2"
 
 [dev-dependencies]
+tokio = { version = "1.44", features = ["test-util"] }
 httpmock = "0.7"
 tempfile = "3.19"
 tower = { version = "0.5", features = ["util"] }

crates/broker/src/listen_api.rs

@@ -612,6 +612,29 @@ async fn listen_api_health(
     let mut payload = listen_api_health_payload(state.default_workspace_id, state.memberships);
     if let Some(status) = fetch_status_for_health(&state.tx).await {
         merge_status_into_health_payload(&mut payload, &status);
+        payload["runtimeResponsive"] = json!(true);
+    } else {
+        // Keep the existing public liveness HTTP contract: a busy actor must
+        // not induce a supervisor restart storm. But never present fabricated
+        // zero-agent/offline values as a successful runtime observation.
+        payload["runtimeResponsive"] = json!(false);
+        if payload["status"] == "ok" {
+            payload["status"] = json!("degraded");
+        }
+        payload["runtimeStatusError"] = json!("runtime_status_unavailable");
+        for key in [
+            "agentCount",
+            "uptimeMs",
+            "pendingDeliveryCount",
+            "deadLetterCount",
+            "wsConnections",
+            "memoryMb",
+            "relaycastConnected",
+            "nodeConnected",
+            "nodeDelivery",
+        ] {
+            payload[key] = Value::Null;
+        }
     }
     axum::Json(payload)
 }
@@ -3880,6 +3903,35 @@ mod auth_tests {
             .expect("request should succeed");
 
         assert_eq!(response.status(), StatusCode::OK);
+        let payload = response_json(response).await;
+        assert_eq!(payload["runtimeResponsive"], false);
+        assert_eq!(payload["runtimeStatusError"], "runtime_status_unavailable");
+        assert!(payload["agentCount"].is_null());
+        assert!(payload["nodeConnected"].is_null());
+    }
+
+    #[tokio::test]
+    async fn health_route_reports_a_responsive_runtime_without_losing_counts() {
+        let (router, mut rx) = test_router(None);
+        let runtime = tokio::spawn(async move {
+            if let Some(super::ListenApiRequest::GetStatus { reply }) = rx.recv().await {
+                let _ = reply.send(Ok(json!({ "agent_count": 1 })));
+            }
+        });
+        let response = router
+            .oneshot(
+                Request::builder()
+                    .uri("/health")
+                    .body(Body::empty())
+                    .unwrap(),
+            )
+            .await
+            .unwrap();
+        assert_eq!(response.status(), StatusCode::OK);
+        let payload = response_json(response).await;
+        assert_eq!(payload["runtimeResponsive"], true);
+        assert_eq!(payload["agentCount"], 1);
+        runtime.await.unwrap();
     }
 
     #[test]