llmist 15.18.0 → 15.19.0

package/dist/index.cjs

@@ -1832,6 +1832,9 @@ function isRetryableError(error) {
   if (message.includes("hf bad request")) {
     return true;
   }
+  if (message.includes("incomplete json segment") || message.includes("exception parsing stream chunk") || message.includes("json error injected into sse stream") || message.includes("sending request") && message.includes("exception") || message.includes("invalid chunk") && message.includes("finish reason") || message.includes("unexpected line format")) {
+    return true;
+  }
   if (message.includes("401") || message.includes("403") || message.includes("400") || message.includes("404") || message.includes("authentication") || message.includes("unauthorized") || message.includes("forbidden") || message.includes("invalid") || message.includes("content policy") || name === "AuthenticationError" || name === "BadRequestError" || name === "NotFoundError" || name === "PermissionDeniedError") {
     return false;
   }
@@ -1937,6 +1940,9 @@ Documentation: ${docsUrl}`;
   if (message.toLowerCase().includes("content policy") || message.toLowerCase().includes("safety")) {
     return "Content policy violation - the request was blocked";
   }
+  if (message.toLowerCase().includes("incomplete json segment") || message.toLowerCase().includes("exception parsing stream chunk") || message.toLowerCase().includes("json error injected into sse stream") || message.toLowerCase().includes("sending request") && message.toLowerCase().includes("exception") || message.toLowerCase().includes("invalid chunk") && message.toLowerCase().includes("finish reason") || message.toLowerCase().includes("unexpected line format")) {
+    return "Stream interrupted - the API returned corrupted or incomplete data";
+  }
   try {
     const parsed = JSON.parse(message);
     const extractedMessage = parsed?.error?.message || parsed?.message;
@@ -2759,6 +2765,17 @@ var init_gadget = __esm({
        * ```
        */
       maxConcurrent;
+      /**
+       * If true, this gadget must execute alone — no other gadgets in the same
+       * LLM response can run in parallel. When an exclusive gadget arrives and
+       * other gadgets are already in-flight, it is deferred until they complete.
+       *
+       * Use for gadgets that terminate the agent loop (e.g., Finish), where
+       * sibling tool results must be visible to the LLM before the loop ends.
+       *
+       * This is a safety floor: external config cannot weaken it.
+       */
+      exclusive;
       /**
        * Throws an AbortException if the execution has been aborted.
        *
@@ -12779,6 +12796,7 @@ function Gadget(config) {
     timeoutMs = config.timeoutMs;
     examples = config.examples;
     maxConcurrent = config.maxConcurrent;
+    exclusive = config.exclusive;
     /**
      * Type helper property for accessing inferred parameter type.
      * This is used in the execute method signature: `execute(params: this['params'])`
@@ -13519,6 +13537,9 @@ var init_stream_processor = __esm({
       activeCountByGadget = /* @__PURE__ */ new Map();
       /** Queue of gadgets waiting for a concurrency slot (per gadget name) */
       concurrencyQueue = /* @__PURE__ */ new Map();
+      // Exclusive gadget support
+      /** Queue of exclusive gadgets deferred until in-flight gadgets complete */
+      exclusiveQueue = [];
       // Cross-iteration dependency tracking
       /** Invocation IDs completed in previous iterations (read-only reference from Agent) */
       priorCompletedInvocations;
@@ -13872,6 +13893,16 @@ var init_stream_processor = __esm({
         if (limitResult.value === true) {
           return;
         }
+        const gadget = this.registry.get(call.gadgetName);
+        if (gadget?.exclusive && this.inFlightExecutions.size > 0) {
+          this.logger.debug("Deferring exclusive gadget until in-flight gadgets complete", {
+            gadgetName: call.gadgetName,
+            invocationId: call.invocationId,
+            inFlightCount: this.inFlightExecutions.size
+          });
+          this.exclusiveQueue.push(call);
+          return;
+        }
         const limit = this.getConcurrencyLimit(call.gadgetName);
         const activeCount = this.activeCountByGadget.get(call.gadgetName) ?? 0;
         if (limit > 0 && activeCount >= limit) {
@@ -14157,7 +14188,7 @@ var init_stream_processor = __esm({
        * Clears the inFlightExecutions map after all gadgets complete.
        */
       async *waitForInFlightExecutions() {
-        if (this.inFlightExecutions.size === 0 && !this.hasQueuedGadgets()) {
+        if (this.inFlightExecutions.size === 0 && !this.hasQueuedGadgets() && this.exclusiveQueue.length === 0) {
           return;
         }
         this.logger.debug("Waiting for in-flight gadget executions", {
@@ -14178,6 +14209,18 @@ var init_stream_processor = __esm({
           }
         }
         this.inFlightExecutions.clear();
+        if (this.exclusiveQueue.length > 0) {
+          this.logger.debug("Processing deferred exclusive gadgets", {
+            count: this.exclusiveQueue.length
+          });
+          const queue = this.exclusiveQueue;
+          this.exclusiveQueue = [];
+          for (const call of queue) {
+            for await (const evt of this.executeGadgetGenerator(call)) {
+              yield evt;
+            }
+          }
+        }
       }
       /**
        * Check if there are any gadgets waiting in concurrency queues.