@azumag/opencode-rate-limit-fallback 1.0.21 → 1.0.22

package/README.md

@@ -89,16 +89,14 @@ If no configuration is provided, the following models are used:
 
 ## How It Works
 
-1. **Detection**: The plugin listens for rate limit errors via:
-   - `session.error` events
-   - `message.updated` events with errors
-   - `session.status` events with `type: "retry"`
+ 1. **Detection**: The plugin listens for rate limit errors via:
+    - `session.error` events
+    - `message.updated` events with errors
+    - `session.status` events with `type: "retry"`
 
-2. **Abort**: When a rate limit is detected, the current session is aborted to stop OpenCode's internal retry mechanism.
+2. **Fallback**: The plugin selects the next available model from the fallback list and resends the last user message using the `promptAsync` API.
 
-3. **Fallback**: The plugin selects the next available model from the fallback list and resends the last user message.
-
-4. **Cooldown**: Rate-limited models are tracked and skipped for the configured cooldown period.
+3. **Cooldown**: Rate-limited models are tracked and skipped for the configured cooldown period.
 
 ## License
 

package/dist/index.js

@@ -316,13 +316,12 @@ export const RateLimitFallback = async ({ client, directory }) => {
                 parts: parts,
                 model: { providerID: nextModel.providerID, modelID: nextModel.modelID },
             };
-            // abort → promptAsync: abort the old request first, then send fallback prompt
-            // This prevents the abort from killing the new fallback prompt.
-            //
-            // For session.error events, the request is already in error state, so we
-            // only send promptAsync (skipAbort = true). For retry status, we send both.
-            //
             // promptAsync: HTTP POST /session/{id}/prompt_async → 204 (SDK sdk.gen.js).
+            //
+            // For session.error and retry status events, we skip abort (skipAbort = true)
+            // because the request is already in an error or delayed state. Calling abort
+            // in these cases can trigger interrupted events that cancel the new fallback
+            // prompt.
             const t0 = Date.now();
             if (!skipAbort) {
                 try {
@@ -458,8 +457,10 @@ export const RateLimitFallback = async ({ client, directory }) => {
                 catch {
                     console.log("[rate-limit-fallback] session.status:", status);
                 }
-                // Note: interrupted status is handled by the fallback sequence (abort → promptAsync).
-                // We don't need to handle it separately here as it would cause duplicate prompt sends.
+                // Note: interrupted status is ignored here. Since we no longer call abort
+                // for retry status events, interrupted events should not be triggered by
+                // our fallback logic. If they occur (e.g., from user action), we let them
+                // be handled by the system.
                 if (status?.type === "retry" && status?.message) {
                     const message = status.message.toLowerCase();
                     const isRateLimitRetry = message.includes("usage limit") ||
@@ -494,7 +495,9 @@ export const RateLimitFallback = async ({ client, directory }) => {
                         catch {
                             console.log("[rate-limit-fallback] Attempting fallback for rate limit retry");
                         }
-                        await handleRateLimitFallback(props.sessionID, "", "");
+                        // skipAbort = true for retry status to avoid triggering interrupted events
+                        // that could cancel the new fallback prompt
+                        await handleRateLimitFallback(props.sessionID, "", "", true);
                     }
                 }
             }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@azumag/opencode-rate-limit-fallback",
-  "version": "1.0.21",
+  "version": "1.0.22",
   "description": "OpenCode plugin that automatically switches to fallback models when rate limited",
   "type": "module",
   "main": "./dist/index.js",