@azumag/opencode-rate-limit-fallback 1.0.13 → 1.0.14

package/dist/index.js

@@ -324,27 +324,18 @@ export const RateLimitFallback = async ({ client, directory }) => {
             catch (abortErr) {
                 logToFile(`abort failed (non-critical): ${abortErr}`);
             }
-            // await toast AFTER abort — provides natural delay (~50ms) for abort flag to clear
-            // before prompting. Without this delay, the new prompt gets immediately aborted.
-            await toast("Retrying", `Using ${nextModel.providerID}/${nextModel.modelID}`, "info");
-            // Try prompt (sync) first — reliably triggers generation in TUI mode.
-            // If it fails (e.g. run mode where server shuts down after abort),
-            // fall back to promptAsync which fires before server can exit.
-            try {
-                await client.session.prompt({
-                    path: { id: sessionID },
-                    body: promptBody,
-                });
-                logToFile(`prompt completed successfully for session ${sessionID} with model ${nextModel.providerID}/${nextModel.modelID}`);
-            }
-            catch (promptErr) {
-                logToFile(`prompt failed (${promptErr}), falling back to promptAsync`);
-                await client.session.promptAsync({
-                    path: { id: sessionID },
-                    body: promptBody,
-                });
-                logToFile(`promptAsync sent successfully for session ${sessionID} with model ${nextModel.providerID}/${nextModel.modelID}`);
-            }
+            // Wait for abort to fully complete before sending new prompt.
+            // Toast alone (~50ms) is insufficient — the abort flag may still be set,
+            // causing the new prompt to be immediately interrupted.
+            await new Promise(resolve => setTimeout(resolve, 500));
+            toast("Retrying", `Using ${nextModel.providerID}/${nextModel.modelID}`, "info").catch(() => { });
+            // Use promptAsync — returns immediately, compatible with both TUI and headless modes.
+            // prompt (sync) causes race condition: abort flag not yet cleared → prompt interrupted.
+            await client.session.promptAsync({
+                path: { id: sessionID },
+                body: promptBody,
+            });
+            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

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@azumag/opencode-rate-limit-fallback",
-  "version": "1.0.13",
+  "version": "1.0.14",
   "description": "OpenCode plugin that automatically switches to fallback models when rate limited",
   "type": "module",
   "main": "./dist/index.js",