@azumag/opencode-rate-limit-fallback 1.0.13 → 1.0.15

package/dist/index.js

@@ -316,7 +316,15 @@ export const RateLimitFallback = async ({ client, directory }) => {
                 parts: parts,
                 model: { providerID: nextModel.providerID, modelID: nextModel.modelID },
             };
-            // Abort to cancel the retry loop
+            toast("Retrying", `Using ${nextModel.providerID}/${nextModel.modelID}`, "info").catch(() => { });
+            // Abort to cancel the retry loop, then promptAsync immediately.
+            // CRITICAL: no delay between abort and promptAsync.
+            // - In headless (opencode run), server disposes ~8ms after session idle.
+            //   Any delay (even 50ms) means promptAsync arrives after server is dead.
+            // - promptAsync returns immediately (just queues), so it fires before
+            //   the server can shut down.
+            // - Do NOT use prompt (sync) here — it causes abort flag race condition
+            //   in TUI mode (prompt immediately interrupted).
             try {
                 await client.session.abort({ path: { id: sessionID } });
                 logToFile(`abort succeeded for session ${sessionID}`);
@@ -324,27 +332,11 @@ 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}`);
-            }
+            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": "1.0.15",
   "description": "OpenCode plugin that automatically switches to fallback models when rate limited",
   "type": "module",
   "main": "./dist/index.js",