fix(runtime-fallback): gate retry prompts

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

@@ -10,6 +10,10 @@ import { buildRetryModelPayload } from "./retry-model-payload"
 import { getLastUserRetryParts } from "./last-user-retry-parts"
 import { extractSessionMessages } from "./session-messages"
 import { resolveRegisteredAgentName } from "../../features/claude-code-session-state"
+import {
+  promptAsyncAfterSessionIdle,
+  releasePromptAsyncReservation,
+} from "../shared/prompt-async-gate"
 
 const SESSION_TTL_MS = 30 * 60 * 1000
 
@@ -33,6 +37,7 @@ export function createAutoRetryHelpers(deps: HookDeps) {
   const abortSessionRequest = async (sessionID: string, source: string): Promise<void> => {
     try {
       await ctx.client.session.abort({ path: { id: sessionID } })
+      releasePromptAsyncReservation(sessionID, `runtime-fallback-abort:${source}`)
       log(`[${HOOK_NAME}] Aborted in-flight session request (${source})`, { sessionID })
     } catch (error) {
       log(`[${HOOK_NAME}] Failed to abort in-flight session request (${source})`, {
@@ -137,15 +142,32 @@ export function createAutoRetryHelpers(deps: HookDeps) {
         sessionAwaitingFallbackResult.add(sessionID)
         scheduleSessionFallbackTimeout(sessionID, retryAgent)
 
-        await ctx.client.session.promptAsync({
-          path: { id: sessionID },
-          body: {
-            ...(launchAgent ? { agent: launchAgent } : {}),
-            ...retryModelPayload,
-            parts: retryParts,
+        const promptResult = await promptAsyncAfterSessionIdle({
+          client: ctx.client,
+          sessionID,
+          source: `runtime-fallback:${source}`,
+          settleMs: 0,
+          postDispatchHoldMs: 0,
+          input: {
+            path: { id: sessionID },
+            body: {
+              ...(launchAgent ? { agent: launchAgent } : {}),
+              ...retryModelPayload,
+              parts: retryParts,
+            },
+            query: { directory: ctx.directory },
           },
-          query: { directory: ctx.directory },
         })
+        if (promptResult.status === "failed") {
+          throw promptResult.error
+        }
+        if (promptResult.status !== "dispatched") {
+          log(`[${HOOK_NAME}] Auto-retry skipped by promptAsync gate (${source})`, {
+            sessionID,
+            status: promptResult.status,
+          })
+          return
+        }
         retryDispatched = true
       } else {
         log(`[${HOOK_NAME}] No user message found for auto-retry (${source})`, { sessionID })

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

@@ -40,6 +40,7 @@ describe("runtime-fallback", () => {
       messages?: (args: unknown) => Promise<unknown>
       promptAsync?: (args: unknown) => Promise<unknown>
       abort?: (args: unknown) => Promise<unknown>
+      status?: () => Promise<unknown>
     }
   }) {
     return unsafeTestValue({
@@ -57,6 +58,7 @@ describe("runtime-fallback", () => {
           messages: overrides?.session?.messages ?? (async () => ({ data: [] })),
           promptAsync: overrides?.session?.promptAsync ?? (async () => ({})),
           abort: overrides?.session?.abort ?? (async () => ({})),
+          ...(overrides?.session?.status ? { status: overrides.session.status } : {}),
         },
       },
       directory: "/test/dir",
@@ -2471,6 +2473,66 @@ describe("runtime-fallback", () => {
       expect(callBody?.agent).toBe("prometheus")
       expect(callBody?.model).toEqual({ providerID: "github-copilot", modelID: "claude-opus-4.7" })
     })
+
+    test("should not dispatch a second fallback prompt while the accepted retry session is still active", async () => {
+      const sessionID = "test-runtime-fallback-active-gate"
+      let sessionStatus = "idle"
+      const promptCalls: Array<Record<string, unknown>> = []
+      const hook = createRuntimeFallbackHook(
+        createMockPluginInput({
+          session: {
+            messages: async () => ({
+              data: [
+                {
+                  info: { role: "user" },
+                  parts: [{ type: "text", text: "retry this" }],
+                },
+              ],
+            }),
+            promptAsync: async (args: unknown) => {
+              promptCalls.push(args as Record<string, unknown>)
+              sessionStatus = "busy"
+              return {}
+            },
+            status: async () => ({ data: { [sessionID]: { type: sessionStatus } } }),
+          },
+        }),
+        {
+          config: createMockConfig({ notify_on_fallback: false }),
+          pluginConfig: createMockPluginConfigWithCategoryFallback([
+            "github-copilot/claude-opus-4.7",
+            "openai/gpt-5.4",
+          ]),
+        },
+      )
+      SessionCategoryRegistry.register(sessionID, "test")
+
+      await hook.event({
+        event: {
+          type: "session.created",
+          properties: { info: { id: sessionID, model: "anthropic/claude-opus-4-7" } },
+        },
+      })
+
+      await hook.event({
+        event: {
+          type: "session.error",
+          properties: { sessionID, error: { statusCode: 503, message: "Service unavailable" } },
+        },
+      })
+      await hook.event({
+        event: {
+          type: "session.error",
+          properties: {
+            sessionID,
+            model: "github-copilot/claude-opus-4.7",
+            error: { statusCode: 503, message: "Service unavailable" },
+          },
+        },
+      })
+
+      expect(promptCalls).toHaveLength(1)
+    })
   })
 
   describe("cooldown mechanism", () => {