Merge pull request #4072 from PeterPonyu/fix/3645-runtime-fallback-git-silent-fail

fix(runtime-fallback): fall back to synthetic continuation when session messages are empty (#3645)

src/hooks/runtime-fallback/auto-retry.ts

@@ -155,8 +155,20 @@ export function createAutoRetryHelpers(deps: HookDeps) {
         query: { directory: ctx.directory },
       })
       const retryPayload = getLastUserRetryPayload(messagesResp, sessionID)
-      const retryParts = retryPayload.retryParts
+      const fetchedParts = retryPayload.retryParts
-      if (retryParts.length > 0) {
+      const retryParts =
+        fetchedParts.length > 0
+          ? fetchedParts
+          : (() => {
+              log(
+                `[${HOOK_NAME}] No user message parts found for auto-retry (${source}); using synthetic continuation`,
+                {
+                  sessionID,
+                  hint: "This can occur when the working directory contains .git and messages are not yet persisted",
+                },
+              )
+              return [{ type: "text" as const, text: "continue" }]
+            })()
       log(`[${HOOK_NAME}] Auto-retrying with fallback model (${source})`, {
         sessionID,
         model: newModel,
@@ -214,9 +226,6 @@ export function createAutoRetryHelpers(deps: HookDeps) {
         state.pendingFallbackPromptMayHaveBeenAccepted = false
       }
       retryDispatched = true
-      } else {
-        log(`[${HOOK_NAME}] No user message found for auto-retry (${source})`, { sessionID })
-      }
     } catch (retryError) {
       log(`[${HOOK_NAME}] Auto-retry failed (${source})`, { sessionID, error: String(retryError) })
     } finally {

src/hooks/runtime-fallback/index.test.ts

@@ -380,6 +380,9 @@ describe("runtime-fallback", () => {
           type: "session.error",
           properties: {
             sessionID,
+            // model at the top level so the awaiting-fallback gate recognises this
+            // as an error from the fallback model we just dispatched
+            model: "anthropic/claude-opus-4.7",
             error: { name: "UnknownError", data: { message: "Model not found: anthropic/claude-opus-4.7." } },
           },
         },
@@ -432,6 +435,9 @@ describe("runtime-fallback", () => {
           type: "session.error",
           properties: {
             sessionID,
+            // model at the top level so the awaiting-fallback gate recognises this
+            // as an error from the fallback model we just dispatched
+            model: "anthropic/claude-opus-4.7",
             error: {
               name: "ProviderModelNotFoundError",
               data: {
@@ -2764,11 +2770,15 @@ describe("runtime-fallback", () => {
         },
       })
 
+      // Simulate the fallback session completing before the next error arrives
+      await hook.event({ event: { type: "session.idle", properties: { sessionID } } })
+
       //#when - second error occurs immediately; tries to switch back to original model but should be in cooldown
       await hook.event({
         event: {
           type: "session.error",
-          properties: { sessionID, error: { statusCode: 429 } },
+          // model matches pendingFallbackModel so the awaiting-fallback gate lets this through
+          properties: { sessionID, model: "openai/gpt-5.4", error: { statusCode: 429 } },
         },
       })
 
@@ -3158,14 +3168,21 @@ describe("runtime-fallback", () => {
         },
       })
 
-      const autoRetryLog = logCalls.find((call) => call.msg.includes("No user message found for auto-retry"))
+      const autoRetryLog = logCalls.find((call) =>
+        call.msg.includes("No user message parts found for auto-retry") &&
+        call.msg.includes("using synthetic continuation"),
+      )
       expect(autoRetryLog).toBeDefined()
 
+      // Simulate the fallback session completing before the next error arrives
+      await hook.event({ event: { type: "session.idle", properties: { sessionID } } })
+
       //#when - second error fires after retry completed (retryInFlight cleared)
       await hook.event({
         event: {
           type: "session.error",
-          properties: { sessionID, error: { statusCode: 429, message: "Rate limit again" } },
+          // model matches pendingFallbackModel so the awaiting-fallback gate lets this through
+          properties: { sessionID, model: "provider-a/model-a", error: { statusCode: 429, message: "Rate limit again" } },
         },
       })