SDK/mount: content-hash client cache (If-None-Match) + jittered Retry-After on all 429 paths

.trajectories/completed/2026-09/traj_0d21k8ac1mdg.json

@@ -0,0 +1,53 @@
+{
+  "id": "traj_0d21k8ac1mdg",
+  "version": 1,
+  "task": {
+    "title": "Repair SDK/mount cache and retry changes from branch check failures"
+  },
+  "status": "completed",
+  "startedAt": "2026-09-27T20:02:30.999Z",
+  "completedAt": "2026-09-27T20:11:31.271Z",
+  "agents": [
+    {
+      "name": "default",
+      "role": "lead",
+      "joinedAt": "2026-09-27T20:08:43.980Z"
+    }
+  ],
+  "chapters": [
+    {
+      "id": "chap_tntqc5ip89bb",
+      "title": "Work",
+      "agentName": "default",
+      "startedAt": "2026-09-27T20:08:43.980Z",
+      "endedAt": "2026-09-27T20:11:31.271Z",
+      "events": [
+        {
+          "ts": 1790539723981,
+          "type": "decision",
+          "content": "Poll for foreground mount state after exit 0 until readiness timeout: Poll for foreground mount state after exit 0 until readiness timeout",
+          "raw": {
+            "question": "Poll for foreground mount state after exit 0 until readiness timeout",
+            "chosen": "Poll for foreground mount state after exit 0 until readiness timeout",
+            "alternatives": [],
+            "reasoning": "The child can exit immediately after publishing state; one overlapping read can observe an absent/partial file and previously caused a false early-exit failure."
+          },
+          "significance": "high"
+        }
+      ]
+    }
+  ],
+  "retrospective": {
+    "summary": "Provisioned CI toolchains locally, fixed foreground mount readiness publication race, and passed the full repository check.",
+    "approach": "Standard approach",
+    "confidence": 0.95
+  },
+  "commits": [],
+  "filesChanged": [],
+  "projectId": "AgentWorkforce/relayfile",
+  "tags": [],
+  "_trace": {
+    "startRef": "cafa926c6f07be6dfcf4a04fadd026db93902b84",
+    "endRef": "cafa926c6f07be6dfcf4a04fadd026db93902b84"
+  }
+}

.trajectories/completed/2026-09/traj_0d21k8ac1mdg.md

@@ -0,0 +1,31 @@
+# Trajectory: Repair SDK/mount cache and retry changes from branch check failures
+
+> **Status:** ✅ Completed
+> **Confidence:** 95%
+> **Started:** September 27, 2026 at 08:02 PM
+> **Completed:** September 27, 2026 at 08:11 PM
+
+---
+
+## Summary
+
+Provisioned CI toolchains locally, fixed foreground mount readiness publication race, and passed the full repository check.
+
+**Approach:** Standard approach
+
+---
+
+## Key Decisions
+
+### Poll for foreground mount state after exit 0 until readiness timeout
+- **Chose:** Poll for foreground mount state after exit 0 until readiness timeout
+- **Reasoning:** The child can exit immediately after publishing state; one overlapping read can observe an absent/partial file and previously caused a false early-exit failure.
+
+---
+
+## Chapters
+
+### 1. Work
+*Agent: default*
+
+- Poll for foreground mount state after exit 0 until readiness timeout: Poll for foreground mount state after exit 0 until readiness timeout

.trajectories/completed/2026-09/traj_6qsrb702mx1s.json

@@ -0,0 +1,65 @@
+{
+  "id": "traj_6qsrb702mx1s",
+  "version": 1,
+  "task": {
+    "title": "Implement content-hash client cache and jittered Retry-After handling"
+  },
+  "status": "completed",
+  "startedAt": "2026-09-27T19:55:27.193Z",
+  "completedAt": "2026-09-27T20:01:45.765Z",
+  "agents": [
+    {
+      "name": "default",
+      "role": "lead",
+      "joinedAt": "2026-09-27T19:55:27.421Z"
+    }
+  ],
+  "chapters": [
+    {
+      "id": "chap_yvjv91bjz5w9",
+      "title": "Work",
+      "agentName": "default",
+      "startedAt": "2026-09-27T19:55:27.421Z",
+      "endedAt": "2026-09-27T20:01:45.765Z",
+      "events": [
+        {
+          "ts": 1790538927421,
+          "type": "decision",
+          "content": "Use content hashes rather than the existing revision ETag for conditional cache identity: Use content hashes rather than the existing revision ETag for conditional cache identity",
+          "raw": {
+            "question": "Use content hashes rather than the existing revision ETag for conditional cache identity",
+            "chosen": "Use content hashes rather than the existing revision ETag for conditional cache identity",
+            "alternatives": [],
+            "reasoning": "Current ETag semantics are revision-based while contentHash is already in response bodies and tree entries."
+          },
+          "significance": "high"
+        },
+        {
+          "ts": 1790538927631,
+          "type": "decision",
+          "content": "Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops: Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops",
+          "raw": {
+            "question": "Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops",
+            "chosen": "Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops",
+            "alternatives": [],
+            "reasoning": "FUSE reproduces the shipped reconnect-storm path; setup retries, agents reconnect, outbox writes, and polite polling are outside the read fan-in failure mode."
+          },
+          "significance": "high"
+        }
+      ]
+    }
+  ],
+  "retrospective": {
+    "summary": "Added SDK content-hash caches, persistent mount object reuse, polite full-jitter retries, bounded bootstrap fan-out, and cursor-safe listen reconnects",
+    "approach": "Standard approach",
+    "confidence": 0.85
+  },
+  "commits": [],
+  "filesChanged": [],
+  "projectId": "AgentWorkforce/relayfile",
+  "tags": [],
+  "_trace": {
+    "startRef": "2d32cfddbfc06e9e371462a50bebd0373e7343a6",
+    "endRef": "2d32cfddbfc06e9e371462a50bebd0373e7343a6"
+  }
+}

.trajectories/completed/2026-09/traj_6qsrb702mx1s.md

@@ -0,0 +1,36 @@
+# Trajectory: Implement content-hash client cache and jittered Retry-After handling
+
+> **Status:** ✅ Completed
+> **Confidence:** 85%
+> **Started:** September 27, 2026 at 07:55 PM
+> **Completed:** September 27, 2026 at 08:01 PM
+
+---
+
+## Summary
+
+Added SDK content-hash caches, persistent mount object reuse, polite full-jitter retries, bounded bootstrap fan-out, and cursor-safe listen reconnects
+
+**Approach:** Standard approach
+
+---
+
+## Key Decisions
+
+### Use content hashes rather than the existing revision ETag for conditional cache identity
+- **Chose:** Use content hashes rather than the existing revision ETag for conditional cache identity
+- **Reasoning:** Current ETag semantics are revision-based while contentHash is already in response bodies and tree entries.
+
+### Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops
+- **Chose:** Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops
+- **Reasoning:** FUSE reproduces the shipped reconnect-storm path; setup retries, agents reconnect, outbox writes, and polite polling are outside the read fan-in failure mode.
+
+---
+
+## Chapters
+
+### 1. Work
+*Agent: default*
+
+- Use content hashes rather than the existing revision ETag for conditional cache identity: Use content hashes rather than the existing revision ETag for conditional cache identity
+- Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops: Include the FUSE websocket invalidator but defer unrelated write and control-plane retry loops

.trajectories/index.json

@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "lastUpdated": "2026-09-18T14:07:19.400Z",
+  "lastUpdated": "2026-09-27T20:11:31.329Z",
   "trajectories": {
     "traj_4pvrlmqfnzng": {
       "title": "Review PR #278 in AgentWorkforce/relayfile",
@@ -407,6 +407,20 @@
       "startedAt": "2026-09-18T13:58:18.225Z",
       "completedAt": "2026-09-18T14:07:19.346Z",
       "path": "/home/khaliqgant/Projects/AgentWorkforce/relayfile/.trajectories/completed/2026-09/traj_p2o6b2q8epj1.json"
+    },
+    "traj_6qsrb702mx1s": {
+      "title": "Implement content-hash client cache and jittered Retry-After handling",
+      "status": "completed",
+      "startedAt": "2026-09-27T19:55:27.193Z",
+      "completedAt": "2026-09-27T20:01:45.765Z",
+      "path": "/home/daytona/.relayflow-v2-supervisor/durable/repository/.trajectories/completed/2026-09/traj_6qsrb702mx1s.json"
+    },
+    "traj_0d21k8ac1mdg": {
+      "title": "Repair SDK/mount cache and retry changes from branch check failures",
+      "status": "completed",
+      "startedAt": "2026-09-27T20:02:30.999Z",
+      "completedAt": "2026-09-27T20:11:31.271Z",
+      "path": "/home/daytona/.relayflow-v2-supervisor/durable/repository/.trajectories/completed/2026-09/traj_0d21k8ac1mdg.json"
     }
   }
 }

cmd/relayfile-cli/main.go

@@ -8785,25 +8785,34 @@ func runListen(args []string, stdout io.Writer) error {
 	)
 	backoff := baseBackoff
 	everConnected := false
+	lastCursor := ""
 	for {
-		connected, err := runListenSession(rootCtx, cfg)
-		if connected {
+		cfg.dialURL = listenDialURLWithCursor(cfg.dialURL, lastCursor)
+		result, err := runListenSession(rootCtx, cfg)
+		if result.cursor != "" {
+			lastCursor = result.cursor
+		}
+		if result.connected {
 			everConnected = true
 			backoff = baseBackoff
 		}
 		if err == nil {
 			return nil
 		}
-		if !everConnected {
+		if !everConnected && result.status != http.StatusTooManyRequests && result.status != http.StatusServiceUnavailable {
 			return fmt.Errorf("connect to event stream: %w", err)
 		}
+		delay := time.Duration(mathrand.Float64() * float64(backoff))
+		if result.retryAfter > delay {
+			delay = result.retryAfter
+		}
 		if !*daemonized {
-			fmt.Fprintf(os.Stderr, "listen: stream error (%v); reconnecting in %s\n", err, backoff)
+			fmt.Fprintf(os.Stderr, "listen: stream error (%v); reconnecting in %s\n", err, delay)
 		}
 		select {
 		case <-rootCtx.Done():
 			return nil
-		case <-time.After(backoff):
+		case <-time.After(delay):
 		}
 		if backoff < maxBackoff {
 			backoff *= 2
@@ -8814,6 +8823,18 @@ func runListen(args []string, stdout io.Writer) error {
 	}
 }
 
+func listenDialURLWithCursor(rawURL, cursor string) string {
+	u, err := url.Parse(rawURL)
+	if err != nil || strings.TrimSpace(cursor) == "" {
+		return rawURL
+	}
+	q := u.Query()
+	q.Del("from")
+	q.Set("cursor", strings.TrimSpace(cursor))
+	u.RawQuery = q.Encode()
+	return u.String()
+}
+
 // listenSessionConfig carries the per-session inputs for runListenSession so a
 // dropped connection can be re-established without recomputing filters.
 type listenSessionConfig struct {
@@ -8833,15 +8854,27 @@ type listenSessionConfig struct {
 // bool reports whether the websocket was successfully established, so the
 // caller can tell a first-attempt dial failure (fatal) from a mid-flight
 // disconnect (retryable).
-func runListenSession(rootCtx context.Context, cfg listenSessionConfig) (bool, error) {
-	conn, _, err := websocket.Dial(rootCtx, cfg.dialURL, &websocket.DialOptions{
+type listenSessionResult struct {
+	connected  bool
+	cursor     string
+	status     int
+	retryAfter time.Duration
+}
+
+func runListenSession(rootCtx context.Context, cfg listenSessionConfig) (listenSessionResult, error) {
+	conn, response, err := websocket.Dial(rootCtx, cfg.dialURL, &websocket.DialOptions{
 		HTTPClient: cfg.httpClient,
 		HTTPHeader: http.Header{
 			"Authorization": []string{"Bearer " + cfg.token},
 		},
 	})
 	if err != nil {
-		return false, err
+		result := listenSessionResult{}
+		if response != nil {
+			result.status = response.StatusCode
+			result.retryAfter = parseRetryAfter(response.Header.Get("Retry-After"))
+		}
+		return result, err
 	}
 	defer conn.Close(websocket.StatusNormalClosure, "")
 
@@ -8852,6 +8885,7 @@ func runListenSession(rootCtx context.Context, cfg listenSessionConfig) (bool, e
 	conn.SetReadLimit(8 << 20) // 8 MiB
 
 	stdout := cfg.stdout
+	lastCursor := ""
 	for {
 		var raw json.RawMessage
 		if err := wsjson.Read(rootCtx, conn, &raw); err != nil {
@@ -8863,15 +8897,18 @@ func runListenSession(rootCtx context.Context, cfg listenSessionConfig) (bool, e
 			// would exit 0 and a `Restart=on-failure` supervisor unit would
 			// not bring the listener back.
 			if errors.Is(err, context.Canceled) || rootCtx.Err() != nil {
-				return true, nil
+				return listenSessionResult{connected: true, cursor: lastCursor}, nil
 			}
-			return true, fmt.Errorf("event stream error: %w", err)
+			return listenSessionResult{connected: true, cursor: lastCursor}, fmt.Errorf("event stream error: %w", err)
 		}
 
 		var evt listenEvent
 		if err := json.Unmarshal(raw, &evt); err != nil || evt.Type == "" || evt.Type == "pong" {
 			continue
 		}
+		if cursor := strings.TrimSpace(evt.EventID); cursor != "" {
+			lastCursor = cursor
+		}
 		if cfg.typeFilter != "" && evt.Type != cfg.typeFilter {
 			continue
 		}