@mastra/memory 1.18.0-alpha.3 → 1.18.0

package/CHANGELOG.md

@@ -1,5 +1,38 @@
 # @mastra/memory
 
+## 1.18.0
+
+### Minor Changes
+
+- Add metadata filtering support to semantic recall. ([#9256](https://github.com/mastra-ai/mastra/pull/9256))
+
+### Patch Changes
+
+- Fixed an issue where tool results containing AI SDK v5 `image-data` content blocks (returned via `toModelOutput`) were stringified into the observational memory prompt as raw base64 text. The base64 data overflowed the observer's context, causing token-limit errors and degenerate output. ([#16117](https://github.com/mastra-ai/mastra/pull/16117))
+
+  Image and file blocks (`image-data`, `image-url`, `file-data`, `file-url`, and `media`) inside tool results are now hoisted into the observer's input as proper attachments, the same way image and file message parts already are. The text body shows a placeholder like `[Image #1: image/png]` so the observer keeps positional context without seeing the bytes.
+
+- Default top-level observational memory early activation settings to observations only, while allowing per-phase overrides under `observation` and `reflection`. ([#16367](https://github.com/mastra-ai/mastra/pull/16367))
+
+- Auto-recover from transient transport errors (e.g. undici `terminated`, `fetch failed`, `UND_ERR_*`, 5xx, 429) in the OM observer and reflector LLM calls. Adds an internal retry wrapper with exponential backoff (1s, 2s, 4s, 8s, 16s, 32s, 64s, 120s — per-attempt delay capped at 2 minutes, ~4 minute total budget per call) so a single network blip from any provider no longer kills the actor turn during long-running sessions. Non-transient errors (auth, validation, etc.) and user-initiated aborts still fail fast. No public API changes. ([#16454](https://github.com/mastra-ai/mastra/pull/16454))
+
+- Added extra defensive checks to prevent edge cases where system messages may have already been stored in message history. ([#15787](https://github.com/mastra-ai/mastra/pull/15787))
+
+- Fixed read-only observational memory so existing context is still loaded. ([#16059](https://github.com/mastra-ai/mastra/pull/16059))
+
+- Updated dependencies [[`9f17410`](https://github.com/mastra-ai/mastra/commit/9f1741080def23d42ee50b39887a385ae316a3c6), [`7ad5585`](https://github.com/mastra-ai/mastra/commit/7ad55856406f1de398dc713f6a9eaa78b2784bb6), [`ac47842`](https://github.com/mastra-ai/mastra/commit/ac478427aa7a5f5fdaed633a911218689b438c60), [`cc189cc`](https://github.com/mastra-ai/mastra/commit/cc189cc0128eb7af233476b5e421ec6888bffde7), [`d1fdbd0`](https://github.com/mastra-ai/mastra/commit/d1fdbd012add5623cb7e6b7f882b605ab358bbb4), [`210ea7a`](https://github.com/mastra-ai/mastra/commit/210ea7af559791b73a44fc9c12179908aaa3183f), [`7c275a8`](https://github.com/mastra-ai/mastra/commit/7c275a810595e1a6c41ccc39720531ab65734700), [`bae019e`](https://github.com/mastra-ai/mastra/commit/bae019ecb6694da96909f7ec7b9eb3a0a33aa887), [`890b24c`](https://github.com/mastra-ai/mastra/commit/890b24cc7d32ed6aa4dfe253e54dc6bf4099f690), [`f984b4d`](https://github.com/mastra-ai/mastra/commit/f984b4d6c60bf2ae2a9b156f0e8c35a66fe96c91), [`6742347`](https://github.com/mastra-ai/mastra/commit/6742347d71955d7639adc9ddf6ff8282de7ee3ba), [`b59316f`](https://github.com/mastra-ai/mastra/commit/b59316ffa0f7688165b0f9c81ccdf85da461e5b2), [`0f48ebf`](https://github.com/mastra-ai/mastra/commit/0f48ebfc7ac7897b2092a189f45751924cf56d1c), [`37c0dc5`](https://github.com/mastra-ai/mastra/commit/37c0dc5697d343db98628bf867bf71ce6deec6d7), [`087e413`](https://github.com/mastra-ai/mastra/commit/087e4133e5d6efa36619e9556c16750e4179c047), [`83218c8`](https://github.com/mastra-ai/mastra/commit/83218c88b37773c9424fbe733b37be556e55e94d), [`ef6b584`](https://github.com/mastra-ai/mastra/commit/ef6b5847ac33c0a7e80af3a86e8801e2933dd3ee), [`c6eb39e`](https://github.com/mastra-ai/mastra/commit/c6eb39ea6dca381c6563cb240237fbe608e02f93), [`7b0ad1f`](https://github.com/mastra-ai/mastra/commit/7b0ad1f5c53dc118c6da12ae82ae2587037dc2b8), [`d91ebe2`](https://github.com/mastra-ai/mastra/commit/d91ebe28ee065d8f2ed6df741c3c07f58d359529), [`62666c3`](https://github.com/mastra-ai/mastra/commit/62666c367eaeac3941ead454b1d38810cc855721), [`33f5061`](https://github.com/mastra-ai/mastra/commit/33f5061cd1c0335020c3faae61ce96de822854fa), [`4af2160`](https://github.com/mastra-ai/mastra/commit/4af2160322f4718cac421930cce85641e9512389), [`087e413`](https://github.com/mastra-ai/mastra/commit/087e4133e5d6efa36619e9556c16750e4179c047), [`265ec9f`](https://github.com/mastra-ai/mastra/commit/265ec9f887b5c81255c873a76ff7796f16e4f99b), [`ce01024`](https://github.com/mastra-ai/mastra/commit/ce010242eee9bdfc09e4c26725b9d37998679a8d), [`6ce80bf`](https://github.com/mastra-ai/mastra/commit/6ce80bf4872a891e0bddf8b80561a80584efb14b), [`f984b4d`](https://github.com/mastra-ai/mastra/commit/f984b4d6c60bf2ae2a9b156f0e8c35a66fe96c91), [`136c959`](https://github.com/mastra-ai/mastra/commit/136c9592fb0eeb0cd212f28629d8a29b7557a2fc), [`9268531`](https://github.com/mastra-ai/mastra/commit/9268531e7ec4be98beeba3b3ae8be0a7ea380662), [`13ead79`](https://github.com/mastra-ai/mastra/commit/13ead79149486b88144db7e11e6ff551caef5be1), [`dccd8f1`](https://github.com/mastra-ai/mastra/commit/dccd8f1f8b8f1ad203b77556207e5529567c616d), [`4df7cc7`](https://github.com/mastra-ai/mastra/commit/4df7cc79342fd065fe7fdeef93c094db14b12bcd), [`f180e49`](https://github.com/mastra-ai/mastra/commit/f180e4990e71b04c9a475b523584071712f0048f), [`9260e01`](https://github.com/mastra-ai/mastra/commit/9260e015276fb1b500f7878ee452b47476bf1583), [`2f6c54e`](https://github.com/mastra-ai/mastra/commit/2f6c54e17c041cac1def54baaa6b771647836414), [`aca3121`](https://github.com/mastra-ai/mastra/commit/aca31211233dac25459f140ea4fcfb3a5af64c18), [`e06a159`](https://github.com/mastra-ai/mastra/commit/e06a1598ca07a6c3778aefc2a2d288363c6294ff), [`4dd900d`](https://github.com/mastra-ai/mastra/commit/4dd900d75dfe9be89f8c15188b368a8622aa1e18), [`b560d6f`](https://github.com/mastra-ai/mastra/commit/b560d6f88b9b904b15c10f75c949eb145bc27684), [`99869ec`](https://github.com/mastra-ai/mastra/commit/99869ecb1f2aa6dfcc44fa4e843e5ee0344efa64), [`900d086`](https://github.com/mastra-ai/mastra/commit/900d086bb737b9cf2fcf68f11b0389b801a2738c), [`c50ebc3`](https://github.com/mastra-ai/mastra/commit/c50ebc34da71044558315735e69bfb94fcfb74bf), [`4c0e286`](https://github.com/mastra-ai/mastra/commit/4c0e28637c9cfb4f416549b55e97ebfa13319dfc), [`55f1e2d`](https://github.com/mastra-ai/mastra/commit/55f1e2d65425b95a49ae788053b266f256e38c96), [`4ff5bdf`](https://github.com/mastra-ai/mastra/commit/4ff5bdfe170cba6dfb5260c6af0f4ba668430772), [`9cdf38e`](https://github.com/mastra-ai/mastra/commit/9cdf38e58506e1109c8b38f97cd7770978a4218e), [`087e413`](https://github.com/mastra-ai/mastra/commit/087e4133e5d6efa36619e9556c16750e4179c047), [`db34bc6`](https://github.com/mastra-ai/mastra/commit/db34bc6fb36cf125bda0c46be4d3fdc774b70cc4), [`990851e`](https://github.com/mastra-ai/mastra/commit/990851edcb0e30be5c2c18b6532f1a876cc2d335), [`bbcd93c`](https://github.com/mastra-ai/mastra/commit/bbcd93cf7d8aa1007d6d84bfd033b8015c912087), [`8373ff4`](https://github.com/mastra-ai/mastra/commit/8373ff46745d77af79f183c4470f80fa2727a6b2), [`d48a705`](https://github.com/mastra-ai/mastra/commit/d48a705ff3dfbdc7a996e07ecd8293b5effd9a2a), [`308bd07`](https://github.com/mastra-ai/mastra/commit/308bd074f35cef0c75d82fc1eb19382fe04ecf6f), [`6068a6c`](https://github.com/mastra-ai/mastra/commit/6068a6c42950fad3ebfc92346417896ba60803d2), [`36b3bbf`](https://github.com/mastra-ai/mastra/commit/36b3bbf5a8d59f7e23d47e29340e76c681b4929c), [`d86f031`](https://github.com/mastra-ai/mastra/commit/d86f031eb6b0b2570145afafea664e59bf688962), [`b275631`](https://github.com/mastra-ai/mastra/commit/b275631dc10541a482b2e2d4a3e3cfa843bd5fa1), [`00106be`](https://github.com/mastra-ai/mastra/commit/00106bede59b81e5b0e9cd6aad8d3b5dbc336387), [`bd36d8e`](https://github.com/mastra-ai/mastra/commit/bd36d8eb6de8c9a0310352649dbd4b06703c2299), [`11c1528`](https://github.com/mastra-ai/mastra/commit/11c152848c5d0ef227184853b5040f5b41ee7b1e), [`4999667`](https://github.com/mastra-ai/mastra/commit/49996678b68356cad7f088430009690406c50fbd), [`e2a079c`](https://github.com/mastra-ai/mastra/commit/e2a079cc3755b1895f7bd5dc36e9be81b11c7c22), [`8ac9141`](https://github.com/mastra-ai/mastra/commit/8ac9141439caa8fdd674944c4d84f29b3c730296), [`25184ff`](https://github.com/mastra-ai/mastra/commit/25184ffaf1293ec95119426eb1a1f8d38831b96c), [`534a456`](https://github.com/mastra-ai/mastra/commit/534a456a25e4df1e5407e7e632f4cb3b1fa14f9d), [`105e454`](https://github.com/mastra-ai/mastra/commit/105e454c95af06a7c741c15969d8f9b0f02463a7), [`aebde9c`](https://github.com/mastra-ai/mastra/commit/aebde9cfacf56592c6b6350cae721740fe090b8a), [`36bae07`](https://github.com/mastra-ai/mastra/commit/36bae07c0e70b1b3006f2fd20830e8883dcbd066), [`5688881`](https://github.com/mastra-ai/mastra/commit/5688881669c7ed157f31ac77f6fc5f8d95ceea32)]:
+  - @mastra/core@1.33.0
+  - @mastra/schema-compat@1.2.10
+
+## 1.18.0-alpha.4
+
+### Patch Changes
+
+- Auto-recover from transient transport errors (e.g. undici `terminated`, `fetch failed`, `UND_ERR_*`, 5xx, 429) in the OM observer and reflector LLM calls. Adds an internal retry wrapper with exponential backoff (1s, 2s, 4s, 8s, 16s, 32s, 64s, 120s — per-attempt delay capped at 2 minutes, ~4 minute total budget per call) so a single network blip from any provider no longer kills the actor turn during long-running sessions. Non-transient errors (auth, validation, etc.) and user-initiated aborts still fail fast. No public API changes. ([#16454](https://github.com/mastra-ai/mastra/pull/16454))
+
+- Updated dependencies [[`b59316f`](https://github.com/mastra-ai/mastra/commit/b59316ffa0f7688165b0f9c81ccdf85da461e5b2), [`55f1e2d`](https://github.com/mastra-ai/mastra/commit/55f1e2d65425b95a49ae788053b266f256e38c96), [`d48a705`](https://github.com/mastra-ai/mastra/commit/d48a705ff3dfbdc7a996e07ecd8293b5effd9a2a)]:
+  - @mastra/core@1.33.0-alpha.12
+
 ## 1.18.0-alpha.3
 
 ### Patch Changes

package/dist/{chunk-XVVCS6R6.js → chunk-LPMZNXSF.js}

@@ -1013,8 +1013,6 @@ var ObservationStrategy = class _ObservationStrategy {
     this.scope = deps.scope;
     this.retrieval = deps.retrieval;
   }
-  deps;
-  opts;
   storage;
   messageHistory;
   tokenCounter;
@@ -2043,8 +2041,6 @@ var ObservationStep = class {
     this.turn = turn;
     this.stepNumber = stepNumber;
   }
-  turn;
-  stepNumber;
   _prepared = false;
   _context;
   /** Whether this step has been prepared. */
@@ -3791,6 +3787,135 @@ function optimizeObservationsForContext(observations) {
   optimized = optimized.replace(/\n{3,}/g, "\n\n");
   return optimized.trim();
 }
+
+// src/processors/observational-memory/retry.ts
+var RETRY_CONFIG = {
+  /** Maximum number of retry *attempts* (total tries = maxRetries + 1). */
+  maxRetries: 8,
+  /** Initial backoff delay in milliseconds. */
+  initialDelayMs: 1e3,
+  /** Multiplier applied to the delay after each failed attempt. */
+  backoffFactor: 2,
+  /** Cap on per-attempt delay (ms). */
+  maxDelayMs: 12e4,
+  /** Random jitter as a fraction of the computed delay (e.g. 0.2 = ±20%). */
+  jitter: 0.2
+};
+var TRANSIENT_MESSAGE_SUBSTRINGS = [
+  "terminated",
+  "fetch failed",
+  "econnreset",
+  "econnrefused",
+  "enotfound",
+  "eai_again",
+  "socket hang up",
+  "network error",
+  "request timed out",
+  "request timeout",
+  "connection reset",
+  "connection closed"
+];
+function isRecord2(value) {
+  return typeof value === "object" && value !== null;
+}
+function isAbortError(error) {
+  if (!isRecord2(error)) return false;
+  if (error.name === "AbortError") return true;
+  if (typeof error.code === "string" && error.code === "ABORT_ERR") return true;
+  return false;
+}
+function hasTransientMessage(value) {
+  if (!isRecord2(value)) return false;
+  const message = typeof value.message === "string" ? value.message.toLowerCase() : "";
+  if (message && TRANSIENT_MESSAGE_SUBSTRINGS.some((sub) => message.includes(sub))) return true;
+  if (typeof value.code === "string" && value.code.toUpperCase().startsWith("UND_ERR_")) return true;
+  return false;
+}
+function hasRetryableHttpStatus(value) {
+  if (!isRecord2(value)) return false;
+  const status = typeof value.statusCode === "number" ? value.statusCode : void 0;
+  if (status === void 0) return false;
+  if (status === 408 || status === 425 || status === 429) return true;
+  if (status >= 500 && status <= 599) return true;
+  return false;
+}
+function hasIsRetryableFlag(value) {
+  if (!isRecord2(value)) return false;
+  return value.isRetryable === true;
+}
+function isTransientLLMError(error) {
+  if (isAbortError(error)) return false;
+  const visited = /* @__PURE__ */ new WeakSet();
+  function visit(candidate) {
+    if (isRecord2(candidate)) {
+      if (visited.has(candidate)) return false;
+      visited.add(candidate);
+    }
+    if (hasTransientMessage(candidate)) return true;
+    if (hasRetryableHttpStatus(candidate)) return true;
+    if (hasIsRetryableFlag(candidate)) return true;
+    if (isRecord2(candidate)) {
+      if (visit(candidate.cause)) return true;
+      if (visit(candidate.error)) return true;
+    }
+    return false;
+  }
+  return visit(error);
+}
+function computeDelay(attempt) {
+  const base = RETRY_CONFIG.initialDelayMs * Math.pow(RETRY_CONFIG.backoffFactor, attempt);
+  const capped = Math.min(base, RETRY_CONFIG.maxDelayMs);
+  const jitterRange = capped * RETRY_CONFIG.jitter;
+  const offset = (Math.random() * 2 - 1) * jitterRange;
+  return Math.max(0, Math.round(capped + offset));
+}
+function sleep(ms, abortSignal) {
+  if (ms <= 0) return Promise.resolve();
+  return new Promise((resolve, reject) => {
+    if (abortSignal?.aborted) {
+      reject(new Error("The operation was aborted."));
+      return;
+    }
+    const timer = setTimeout(() => {
+      abortSignal?.removeEventListener("abort", onAbort);
+      resolve();
+    }, ms);
+    const onAbort = () => {
+      clearTimeout(timer);
+      abortSignal?.removeEventListener("abort", onAbort);
+      reject(new Error("The operation was aborted."));
+    };
+    abortSignal?.addEventListener("abort", onAbort, { once: true });
+  });
+}
+async function withRetry(fn, opts) {
+  const { label, abortSignal } = opts;
+  let attempt = 0;
+  while (true) {
+    if (abortSignal?.aborted) {
+      throw new Error("The operation was aborted.");
+    }
+    try {
+      return await fn();
+    } catch (error) {
+      if (isAbortError(error) || abortSignal?.aborted) throw error;
+      if (attempt >= RETRY_CONFIG.maxRetries || !isTransientLLMError(error)) {
+        if (attempt > 0) {
+          omDebug(
+            `[OM:retry:${label}] giving up after ${attempt} retry/retries: ${error instanceof Error ? error.message : String(error)}`
+          );
+        }
+        throw error;
+      }
+      const delay = computeDelay(attempt);
+      attempt++;
+      omDebug(
+        `[OM:retry:${label}] transient error on attempt ${attempt}, retrying in ${delay}ms: ${error instanceof Error ? error.message : String(error)}`
+      );
+      await sleep(delay, abortSignal);
+    }
+  }
+}
 var PHASE_CONFIG = {
   observer: {
     name: "om.observer",
@@ -3901,32 +4026,35 @@ var ObserverRunner = class {
       buildObserverHistoryMessage(messagesToObserve)
     ];
     const doGenerate = async () => {
-      return withOmTracingSpan({
-        phase: "observer",
-        model: resolvedModel.model,
-        inputTokens,
-        requestContext: options?.requestContext,
-        observabilityContext: options?.observabilityContext,
-        metadata: {
-          omPreviousObserverTokens: this.observationConfig.previousObserverTokens,
-          omThreadTitleEnabled: this.observationConfig.threadTitle,
-          omSkipContinuationHints: options?.skipContinuationHints ?? false,
-          omWasTruncated: options?.wasTruncated ?? false,
-          ...resolvedModel.selectedThreshold !== void 0 ? { omSelectedThreshold: resolvedModel.selectedThreshold } : {},
-          ...resolvedModel.routingStrategy ? { omRoutingStrategy: resolvedModel.routingStrategy } : {},
-          ...resolvedModel.routingThresholds ? { omRoutingThresholds: resolvedModel.routingThresholds } : {}
-        },
-        callback: (childObservabilityContext) => this.withAbortCheck(async () => {
-          const streamResult = await agent.stream(observerMessages, {
-            modelSettings: { ...this.observationConfig.modelSettings },
-            providerOptions: this.observationConfig.providerOptions,
-            ...abortSignal ? { abortSignal } : {},
-            ...options?.requestContext ? { requestContext: options.requestContext } : {},
-            ...childObservabilityContext
-          });
-          return streamResult.getFullOutput();
-        }, abortSignal)
-      });
+      return withRetry(
+        () => withOmTracingSpan({
+          phase: "observer",
+          model: resolvedModel.model,
+          inputTokens,
+          requestContext: options?.requestContext,
+          observabilityContext: options?.observabilityContext,
+          metadata: {
+            omPreviousObserverTokens: this.observationConfig.previousObserverTokens,
+            omThreadTitleEnabled: this.observationConfig.threadTitle,
+            omSkipContinuationHints: options?.skipContinuationHints ?? false,
+            omWasTruncated: options?.wasTruncated ?? false,
+            ...resolvedModel.selectedThreshold !== void 0 ? { omSelectedThreshold: resolvedModel.selectedThreshold } : {},
+            ...resolvedModel.routingStrategy ? { omRoutingStrategy: resolvedModel.routingStrategy } : {},
+            ...resolvedModel.routingThresholds ? { omRoutingThresholds: resolvedModel.routingThresholds } : {}
+          },
+          callback: (childObservabilityContext) => this.withAbortCheck(async () => {
+            const streamResult = await agent.stream(observerMessages, {
+              modelSettings: { ...this.observationConfig.modelSettings },
+              providerOptions: this.observationConfig.providerOptions,
+              ...abortSignal ? { abortSignal } : {},
+              ...options?.requestContext ? { requestContext: options.requestContext } : {},
+              ...childObservabilityContext
+            });
+            return streamResult.getFullOutput();
+          }, abortSignal)
+        }),
+        { label: "observer", abortSignal }
+      );
     };
     let result = await doGenerate();
     let parsed = parseObserverOutput(result.text);
@@ -4000,31 +4128,34 @@ var ObserverRunner = class {
       }
     }
     const doGenerate = async () => {
-      return withOmTracingSpan({
-        phase: "observer-multi-thread",
-        model: resolvedModel.model,
-        inputTokens,
-        requestContext,
-        observabilityContext,
-        metadata: {
-          omThreadCount: threadOrder.length,
-          omPreviousObserverTokens: this.observationConfig.previousObserverTokens,
-          omThreadTitleEnabled: this.observationConfig.threadTitle,
-          ...resolvedModel.selectedThreshold !== void 0 ? { omSelectedThreshold: resolvedModel.selectedThreshold } : {},
-          ...resolvedModel.routingStrategy ? { omRoutingStrategy: resolvedModel.routingStrategy } : {},
-          ...resolvedModel.routingThresholds ? { omRoutingThresholds: resolvedModel.routingThresholds } : {}
-        },
-        callback: (childObservabilityContext) => this.withAbortCheck(async () => {
-          const streamResult = await agent.stream(observerMessages, {
-            modelSettings: { ...this.observationConfig.modelSettings },
-            providerOptions: this.observationConfig.providerOptions,
-            ...abortSignal ? { abortSignal } : {},
-            ...requestContext ? { requestContext } : {},
-            ...childObservabilityContext
-          });
-          return streamResult.getFullOutput();
-        }, abortSignal)
-      });
+      return withRetry(
+        () => withOmTracingSpan({
+          phase: "observer-multi-thread",
+          model: resolvedModel.model,
+          inputTokens,
+          requestContext,
+          observabilityContext,
+          metadata: {
+            omThreadCount: threadOrder.length,
+            omPreviousObserverTokens: this.observationConfig.previousObserverTokens,
+            omThreadTitleEnabled: this.observationConfig.threadTitle,
+            ...resolvedModel.selectedThreshold !== void 0 ? { omSelectedThreshold: resolvedModel.selectedThreshold } : {},
+            ...resolvedModel.routingStrategy ? { omRoutingStrategy: resolvedModel.routingStrategy } : {},
+            ...resolvedModel.routingThresholds ? { omRoutingThresholds: resolvedModel.routingThresholds } : {}
+          },
+          callback: (childObservabilityContext) => this.withAbortCheck(async () => {
+            const streamResult = await agent.stream(observerMessages, {
+              modelSettings: { ...this.observationConfig.modelSettings },
+              providerOptions: this.observationConfig.providerOptions,
+              ...abortSignal ? { abortSignal } : {},
+              ...requestContext ? { requestContext } : {},
+              ...childObservabilityContext
+            });
+            return streamResult.getFullOutput();
+          }, abortSignal)
+        }),
+        { label: "observer-multi-thread", abortSignal }
+      );
     };
     let result = await doGenerate();
     let parsed = parseMultiThreadObserverOutput(result.text);
@@ -4468,54 +4599,60 @@ var ReflectorRunner = class {
         `[OM:callReflector] ${isRetry ? `retry #${attemptNumber - 1}` : "first attempt"}: level=${currentLevel}, originalTokens=${originalTokens}, targetThreshold=${targetThreshold}, promptLen=${prompt.length}, skipContinuationHints=${skipContinuationHints}`
       );
       let chunkCount = 0;
-      const result = await withOmTracingSpan({
-        phase: "reflector",
-        model: resolvedModel.model,
-        inputTokens: originalTokens,
-        requestContext,
-        observabilityContext,
-        metadata: {
-          omCompressionLevel: currentLevel,
-          omCompressionAttempt: attemptNumber,
-          omTargetThreshold: targetThreshold,
-          omSkipContinuationHints: skipContinuationHints ?? false,
-          ...resolvedModel.selectedThreshold !== void 0 ? { omSelectedThreshold: resolvedModel.selectedThreshold } : {},
-          ...resolvedModel.routingStrategy ? { omRoutingStrategy: resolvedModel.routingStrategy } : {},
-          ...resolvedModel.routingThresholds ? { omRoutingThresholds: resolvedModel.routingThresholds } : {}
-        },
-        callback: (childObservabilityContext) => withAbortCheck(async () => {
-          const streamResult = await agent.stream(prompt, {
-            modelSettings: {
-              ...this.reflectionConfig.modelSettings
-            },
-            providerOptions: this.reflectionConfig.providerOptions,
-            ...abortSignal ? { abortSignal } : {},
-            ...requestContext ? { requestContext } : {},
-            ...childObservabilityContext,
-            ...attemptNumber === 1 ? {
-              onChunk(chunk) {
-                chunkCount++;
-                if (chunkCount === 1 || chunkCount % 50 === 0) {
-                  const preview = chunk.type === "text-delta" ? ` text="${chunk.textDelta?.slice(0, 80)}..."` : chunk.type === "tool-call" ? ` tool=${chunk.toolName}` : "";
-                  omDebug(`[OM:callReflector] chunk#${chunkCount}: type=${chunk.type}${preview}`);
-                }
-              },
-              onFinish(event) {
-                omDebug(
-                  `[OM:callReflector] onFinish: chunks=${chunkCount}, finishReason=${event.finishReason}, inputTokens=${event.usage?.inputTokens}, outputTokens=${event.usage?.outputTokens}, textLen=${event.text?.length}`
-                );
-              },
-              onAbort(event) {
-                omDebug(`[OM:callReflector] onAbort: chunks=${chunkCount}, reason=${event?.reason ?? "unknown"}`);
+      const result = await withRetry(
+        () => withOmTracingSpan({
+          phase: "reflector",
+          model: resolvedModel.model,
+          inputTokens: originalTokens,
+          requestContext,
+          observabilityContext,
+          metadata: {
+            omCompressionLevel: currentLevel,
+            omCompressionAttempt: attemptNumber,
+            omTargetThreshold: targetThreshold,
+            omSkipContinuationHints: skipContinuationHints ?? false,
+            ...resolvedModel.selectedThreshold !== void 0 ? { omSelectedThreshold: resolvedModel.selectedThreshold } : {},
+            ...resolvedModel.routingStrategy ? { omRoutingStrategy: resolvedModel.routingStrategy } : {},
+            ...resolvedModel.routingThresholds ? { omRoutingThresholds: resolvedModel.routingThresholds } : {}
+          },
+          callback: (childObservabilityContext) => withAbortCheck(async () => {
+            chunkCount = 0;
+            const streamResult = await agent.stream(prompt, {
+              modelSettings: {
+                ...this.reflectionConfig.modelSettings
               },
-              onError({ error }) {
-                omError(`[OM:callReflector] onError after ${chunkCount} chunks`, error);
-              }
-            } : {}
-          });
-          return streamResult.getFullOutput();
-        }, abortSignal)
-      });
+              providerOptions: this.reflectionConfig.providerOptions,
+              ...abortSignal ? { abortSignal } : {},
+              ...requestContext ? { requestContext } : {},
+              ...childObservabilityContext,
+              ...attemptNumber === 1 ? {
+                onChunk(chunk) {
+                  chunkCount++;
+                  if (chunkCount === 1 || chunkCount % 50 === 0) {
+                    const preview = chunk.type === "text-delta" ? ` text="${chunk.textDelta?.slice(0, 80)}..."` : chunk.type === "tool-call" ? ` tool=${chunk.toolName}` : "";
+                    omDebug(`[OM:callReflector] chunk#${chunkCount}: type=${chunk.type}${preview}`);
+                  }
+                },
+                onFinish(event) {
+                  omDebug(
+                    `[OM:callReflector] onFinish: chunks=${chunkCount}, finishReason=${event.finishReason}, inputTokens=${event.usage?.inputTokens}, outputTokens=${event.usage?.outputTokens}, textLen=${event.text?.length}`
+                  );
+                },
+                onAbort(event) {
+                  omDebug(
+                    `[OM:callReflector] onAbort: chunks=${chunkCount}, reason=${event?.reason ?? "unknown"}`
+                  );
+                },
+                onError({ error }) {
+                  omError(`[OM:callReflector] onError after ${chunkCount} chunks`, error);
+                }
+              } : {}
+            });
+            return streamResult.getFullOutput();
+          }, abortSignal)
+        }),
+        { label: "reflector", abortSignal }
+      );
       omDebug(
         `[OM:callReflector] attempt #${attemptNumber} returned: textLen=${result.text?.length}, textPreview="${result.text?.slice(0, 120)}...", inputTokens=${result.usage?.inputTokens ?? result.totalUsage?.inputTokens}, outputTokens=${result.usage?.outputTokens ?? result.totalUsage?.outputTokens}`
       );
@@ -9496,5 +9633,5 @@ function getObservationsAsOf(activeObservations, asOf) {
 }
 
 export { ModelByInputTokens, OBSERVER_SYSTEM_PROMPT, ObservationalMemory, ObservationalMemoryProcessor, TokenCounter, buildObserverPrompt, buildObserverSystemPrompt, combineObservationGroupRanges, deriveObservationGroupProvenance, extractCurrentTask, formatMessagesForObserver, formatToolResultForObserver, getObservationsAsOf, hasCurrentTaskSection, injectAnchorIds, optimizeObservationsForContext, parseAnchorId, parseObservationGroups, parseObserverOutput, reconcileObservationGroupsFromReflection, renderObservationGroupsForReflection, resolveToolResultValue, stripEphemeralAnchorIds, stripObservationGroups, truncateStringByTokens, wrapInObservationGroup };
-//# sourceMappingURL=chunk-XVVCS6R6.js.map
-//# sourceMappingURL=chunk-XVVCS6R6.js.map
+//# sourceMappingURL=chunk-LPMZNXSF.js.map
+//# sourceMappingURL=chunk-LPMZNXSF.js.map