@azumag/opencode-rate-limit-fallback 1.0.10 → 1.0.12

package/dist/index.js

@@ -225,8 +225,7 @@ export const RateLimitFallback = async ({ client, directory }) => {
                     currentModelID = tracked.modelID;
                 }
             }
-            await client.session.abort({ path: { id: sessionID } });
-            await toast("Rate Limit Detected", `Switching from ${currentModelID || 'current model'}...`, "warning");
+            // Fetch messages BEFORE abort — session must still be alive
             const messagesResult = await client.session.messages({ path: { id: sessionID } });
             if (!messagesResult.data)
                 return;
@@ -234,6 +233,7 @@ export const RateLimitFallback = async ({ client, directory }) => {
             const lastUserMessage = [...messages].reverse().find(m => m.info.role === "user");
             if (!lastUserMessage)
                 return;
+            toast("Rate Limit Detected", `Switching from ${currentModelID || 'current model'}...`, "warning").catch(() => { });
             const stateKey = `${sessionID}:${lastUserMessage.info.id}`;
             let state = retryState.get(stateKey);
             if (!state || Date.now() - state.lastAttemptTime > 30000) {
@@ -259,12 +259,11 @@ export const RateLimitFallback = async ({ client, directory }) => {
                     // Try the last model in the list once, then reset on next prompt
                     const lastModel = config.fallbackModels[config.fallbackModels.length - 1];
                     if (lastModel) {
-                        const lastKey = getModelKey(lastModel.providerID, lastModel.modelID);
                         const isLastModelCurrent = currentProviderID === lastModel.providerID && currentModelID === lastModel.modelID;
                         if (!isLastModelCurrent && !isModelRateLimited(lastModel.providerID, lastModel.modelID)) {
                             // Use the last model for one more try
                             nextModel = lastModel;
-                            await toast("Last Resort", `Trying ${lastModel.modelID} one more time...`, "warning");
+                            toast("Last Resort", `Trying ${lastModel.modelID} one more time...`, "warning").catch(() => { });
                         }
                         else {
                             // Last model also failed, reset for next prompt
@@ -279,9 +278,9 @@ export const RateLimitFallback = async ({ client, directory }) => {
                 // "stop" mode: nextModel remains null, will show error below
             }
             if (!nextModel) {
-                await toast("No Fallback Available", config.fallbackMode === "stop"
+                toast("No Fallback Available", config.fallbackMode === "stop"
                     ? "All fallback models exhausted"
-                    : "All models are rate limited", "error");
+                    : "All models are rate limited", "error").catch(() => { });
                 retryState.delete(stateKey);
                 fallbackInProgress.delete(sessionID);
                 return;
@@ -300,22 +299,35 @@ export const RateLimitFallback = async ({ client, directory }) => {
                 .filter(Boolean);
             if (parts.length === 0)
                 return;
-            await toast("Retrying", `Using ${nextModel.providerID}/${nextModel.modelID}`, "info");
+            toast("Retrying", `Using ${nextModel.providerID}/${nextModel.modelID}`, "info").catch(() => { });
             // Track the new model for this session
             currentSessionModel.set(sessionID, { providerID: nextModel.providerID, modelID: nextModel.modelID });
-            await client.session.prompt({
+            const promptBody = {
+                parts: parts,
+                model: { providerID: nextModel.providerID, modelID: nextModel.modelID },
+            };
+            // Abort first to cancel the retry loop, then promptAsync immediately
+            // The abort→promptAsync gap is minimal, so even in headless mode
+            // the server won't shut down before promptAsync fires
+            try {
+                await client.session.abort({ path: { id: sessionID } });
+                logToFile(`abort succeeded for session ${sessionID}`);
+            }
+            catch (abortErr) {
+                logToFile(`abort failed (non-critical): ${abortErr}`);
+            }
+            await client.session.promptAsync({
                 path: { id: sessionID },
-                body: {
-                    parts: parts,
-                    model: { providerID: nextModel.providerID, modelID: nextModel.modelID },
-                },
+                body: promptBody,
             });
-            await toast("Fallback Successful", `Now using ${nextModel.modelID}`, "success");
+            logToFile(`promptAsync sent successfully for session ${sessionID} with model ${nextModel.providerID}/${nextModel.modelID}`);
+            toast("Fallback Successful", `Now using ${nextModel.modelID}`, "success").catch(() => { });
             retryState.delete(stateKey);
             // Clear fallback flag to allow next fallback if needed
             fallbackInProgress.delete(sessionID);
         }
         catch (err) {
+            logToFile(`handleRateLimitFallback error: ${err}`);
             // Fallback failed, clear the flag
             fallbackInProgress.delete(sessionID);
         }
@@ -384,6 +396,13 @@ export const RateLimitFallback = async ({ client, directory }) => {
                 catch {
                     console.log("[rate-limit-fallback] message.updated:", info);
                 }
+                // Track assistant message model info for later use in fallback
+                if (info?.role === "assistant" && info?.sessionID && info?.providerID && info?.modelID) {
+                    currentSessionModel.set(info.sessionID, {
+                        providerID: info.providerID,
+                        modelID: info.modelID,
+                    });
+                }
                 if (info?.error && isRateLimitError(info.error)) {
                     logToFile("Rate limit error in message, attempting fallback");
                     try {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@azumag/opencode-rate-limit-fallback",
-  "version": "1.0.10",
+  "version": "1.0.12",
   "description": "OpenCode plugin that automatically switches to fallback models when rate limited",
   "type": "module",
   "main": "./dist/index.js",