@azumag/opencode-rate-limit-fallback 1.66.0 → 1.68.0

package/dist/index.js

@@ -158,6 +158,11 @@ export const RateLimitFallback = async ({ client, directory, worktree }) => {
     if (!config.enabled) {
         return {};
     }
+    // Disable fallback in headless mode — headless sessions should use their configured model only
+    if (isHeadless) {
+        logger.info("Headless mode detected — model fallback disabled");
+        return {};
+    }
     // Initialize error pattern registry
     const errorPatternRegistry = new ErrorPatternRegistry(logger);
     if (config.errorPatterns?.custom) {

package/dist/src/fallback/FallbackHandler.d.ts

@@ -43,13 +43,9 @@ export declare class FallbackHandler {
         modelID: string;
     } | null;
     /**
-     * Abort current session with error handling
-     */
-    private abortSession;
-    /**
-     * Queue prompt asynchronously (non-blocking), then abort retry loop.
-     * promptAsync FIRST queues pending work so that server doesn't dispose on idle.
-     * abort SECOND cancels the retry loop; the server sees the queued prompt and processes it.
+     * Queue prompt asynchronously (non-blocking) to schedule fallback.
+     * The server's retry loop finishes naturally; it then picks up the queued prompt.
+     * We do NOT call abort — its AbortController signal persists and kills the new stream.
      */
     retryWithModel(targetSessionID: string, model: FallbackModel, parts: MessagePart[], hierarchy: SessionHierarchy | null): Promise<void>;
     /**

package/dist/src/fallback/FallbackHandler.js

@@ -92,21 +92,9 @@ export class FallbackHandler {
         return tracked ? { providerID: tracked.providerID, modelID: tracked.modelID } : null;
     }
     /**
-     * Abort current session with error handling
-     */
-    async abortSession(sessionID) {
-        try {
-            await this.client.session.abort({ path: { id: sessionID } });
-        }
-        catch (abortError) {
-            // Silently ignore abort errors and continue with fallback
-            this.logger.debug(`Failed to abort session ${sessionID}`, { error: abortError });
-        }
-    }
-    /**
-     * Queue prompt asynchronously (non-blocking), then abort retry loop.
-     * promptAsync FIRST queues pending work so that server doesn't dispose on idle.
-     * abort SECOND cancels the retry loop; the server sees the queued prompt and processes it.
+     * Queue prompt asynchronously (non-blocking) to schedule fallback.
+     * The server's retry loop finishes naturally; it then picks up the queued prompt.
+     * We do NOT call abort — its AbortController signal persists and kills the new stream.
      */
     async retryWithModel(targetSessionID, model, parts, hierarchy) {
         // Record model usage for dynamic prioritization
@@ -152,11 +140,13 @@ export class FallbackHandler {
                 model: { providerID: model.providerID, modelID: model.modelID },
             },
         });
-        // 2. abort: cancel the retry loop; server sees queued prompt and processes it
-        await this.abortSession(targetSessionID);
+        // Do NOT call abort after promptAsync.
+        // The AbortController signal persists and kills the newly queued stream too,
+        // causing "interrupted" in TUI mode and server disposal in headless mode.
+        // Let the server's retry loop finish naturally; it will pick up the queued prompt.
         await safeShowToast(this.client, {
             body: {
-                title: "Fallback Successful",
+                title: "Fallback Queued",
                 message: `Now using ${model.modelID}`,
                 variant: "success",
                 duration: 3000,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@azumag/opencode-rate-limit-fallback",
-  "version": "1.66.0",
+  "version": "1.68.0",
   "description": "OpenCode plugin that automatically switches to fallback models when rate limited",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",