@azumag/opencode-rate-limit-fallback 1.0.1 → 1.0.3

package/.claude/settings.local.json

@@ -10,7 +10,9 @@
       "Bash(npm token create:*)",
       "Bash(npm view:*)",
       "Bash(git add:*)",
-      "Bash(git commit -m \"$\\(cat <<''EOF''\nPublish as scoped package @azumag/opencode-rate-limit-fallback\n\n- Rename package to @azumag/opencode-rate-limit-fallback\n- Add npm installation instructions to README\n- Add npm version badge\n\nCo-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>\nEOF\n\\)\")"
+      "Bash(git commit -m \"$\\(cat <<''EOF''\nPublish as scoped package @azumag/opencode-rate-limit-fallback\n\n- Rename package to @azumag/opencode-rate-limit-fallback\n- Add npm installation instructions to README\n- Add npm version badge\n\nCo-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>\nEOF\n\\)\")",
+      "Bash(git push:*)",
+      "Bash(git commit:*)"
     ]
   }
 }

package/README.md

@@ -14,19 +14,17 @@ OpenCode plugin that automatically switches to fallback models when rate limited
 
 ## Installation
 
-### Via npm
+Add the plugin to your `opencode.json`:
 
-```bash
-npm install @azumag/opencode-rate-limit-fallback
+```json
+{
+  "plugins": ["@azumag/opencode-rate-limit-fallback"]
+}
 ```
 
-Then copy to the plugins directory:
+OpenCode will automatically install the plugin on startup.
 
-```bash
-cp node_modules/@azumag/opencode-rate-limit-fallback/index.ts ~/.config/opencode/plugins/rate-limit-fallback.ts
-```
-
-### Manual Installation
+### Manual Installation (Alternative)
 
 Copy `index.ts` to your OpenCode plugins directory:
 
@@ -36,8 +34,6 @@ curl -o ~/.config/opencode/plugins/rate-limit-fallback.ts \
   https://raw.githubusercontent.com/azumag/opencode-rate-limit-fallback/main/index.ts
 ```
 
-Dependencies (`@opencode-ai/plugin`) will be automatically installed by OpenCode on startup.
-
 Restart OpenCode to load the plugin.
 
 ## Configuration

package/index.ts

@@ -76,6 +76,9 @@ function isRateLimitError(error: any): boolean {
     "quota exceeded",
     "resource exhausted",
     "usage limit",
+    "high concurrency usage of this api",
+    "high concurrency",
+    "reduce concurrency",
     "429",
   ];
 
@@ -96,6 +99,8 @@ export const RateLimitFallback: Plugin = async ({ client, directory }) => {
 
   const rateLimitedModels = new Map<string, number>();
   const retryState = new Map<string, { attemptedModels: Set<string>; lastAttemptTime: number }>();
+  const currentSessionModel = new Map<string, { providerID: string; modelID: string }>();
+  const fallbackInProgress = new Map<string, number>(); // sessionID -> timestamp
 
   function isModelRateLimited(providerID: string, modelID: string): boolean {
     const key = getModelKey(providerID, modelID);
@@ -139,12 +144,28 @@ export const RateLimitFallback: Plugin = async ({ client, directory }) => {
 
   async function handleRateLimitFallback(sessionID: string, currentProviderID: string, currentModelID: string) {
     try {
+      // Prevent duplicate fallback processing within 5 seconds
+      const lastFallback = fallbackInProgress.get(sessionID);
+      if (lastFallback && Date.now() - lastFallback < 5000) {
+        return;
+      }
+      fallbackInProgress.set(sessionID, Date.now());
+
+      // If no model info provided, try to get from tracked session model
+      if (!currentProviderID || !currentModelID) {
+        const tracked = currentSessionModel.get(sessionID);
+        if (tracked) {
+          currentProviderID = tracked.providerID;
+          currentModelID = tracked.modelID;
+        }
+      }
+
       await client.session.abort({ path: { id: sessionID } });
 
       await client.tui.showToast({
         body: {
           title: "Rate Limit Detected",
-          message: "Switching to fallback model...",
+          message: `Switching from ${currentModelID || 'current model'}...`,
           variant: "warning",
           duration: 3000,
         },
@@ -170,7 +191,17 @@ export const RateLimitFallback: Plugin = async ({ client, directory }) => {
         state.attemptedModels.add(getModelKey(currentProviderID, currentModelID));
       }
 
-      const nextModel = findNextAvailableModel(currentProviderID || "", currentModelID || "", state.attemptedModels);
+      let nextModel = findNextAvailableModel(currentProviderID || "", currentModelID || "", state.attemptedModels);
+
+      // If no model found and we've attempted models, reset and try again from the beginning
+      if (!nextModel && state.attemptedModels.size > 0) {
+        state.attemptedModels.clear();
+        // Keep the current model marked as attempted to avoid immediate retry
+        if (currentProviderID && currentModelID) {
+          state.attemptedModels.add(getModelKey(currentProviderID, currentModelID));
+        }
+        nextModel = findNextAvailableModel("", "", state.attemptedModels);
+      }
 
       if (!nextModel) {
         await client.tui.showToast({
@@ -208,6 +239,9 @@ export const RateLimitFallback: Plugin = async ({ client, directory }) => {
         },
       });
 
+      // Track the new model for this session
+      currentSessionModel.set(sessionID, { providerID: nextModel.providerID, modelID: nextModel.modelID });
+
       await client.session.prompt({
         path: { id: sessionID },
         body: {
@@ -226,8 +260,11 @@ export const RateLimitFallback: Plugin = async ({ client, directory }) => {
       });
 
       retryState.delete(stateKey);
+      // Clear fallback flag to allow next fallback if needed
+      fallbackInProgress.delete(sessionID);
     } catch (err) {
-      // Fallback failed silently
+      // Fallback failed, clear the flag
+      fallbackInProgress.delete(sessionID);
     }
   }
 
@@ -253,10 +290,15 @@ export const RateLimitFallback: Plugin = async ({ client, directory }) => {
 
         if (status?.type === "retry" && status?.message) {
           const message = status.message.toLowerCase();
-          if (message.includes("usage limit") || message.includes("rate limit")) {
-            if (status.attempt === 1) {
-              await handleRateLimitFallback(props.sessionID, "", "");
-            }
+          const isRateLimitRetry =
+            message.includes("usage limit") ||
+            message.includes("rate limit") ||
+            message.includes("high concurrency") ||
+            message.includes("reduce concurrency");
+
+          if (isRateLimitRetry) {
+            // Try fallback on any attempt, handleRateLimitFallback will manage state
+            await handleRateLimitFallback(props.sessionID, "", "");
           }
         }
       }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@azumag/opencode-rate-limit-fallback",
-  "version": "1.0.1",
+  "version": "1.0.3",
   "description": "OpenCode plugin that automatically switches to fallback models when rate limited",
   "main": "index.ts",
   "type": "module",