fix: drain sdk follow-up turns (#219)

src/agent-runner.test.ts

@@ -852,4 +852,53 @@ describe('agent-runner Codex goals', () => {
       | undefined;
     expect(spawnEnv?.CODEX_GOALS).toBe('true');
   });
+
+  it('marks SDK codex processes so queued IPC follow-ups drain after the active run', async () => {
+    const previousCodexGoals = process.env.CODEX_GOALS;
+    delete process.env.CODEX_GOALS;
+    const onProcess = vi.fn();
+    const codexGroup: RegisteredGroup = {
+      ...testGroup,
+      agentType: 'codex',
+    };
+
+    try {
+      const resultPromise = runAgentProcess(
+        codexGroup,
+        {
+          ...testInput,
+          roomRoleContext: {
+            serviceId: 'codex-main',
+            role: 'owner',
+            ownerServiceId: 'codex-main',
+            reviewerServiceId: 'claude',
+            failoverOwner: false,
+          },
+        },
+        onProcess,
+        async () => {},
+        {
+          CODEX_RUNTIME: 'sdk',
+          CODEX_RUNTIME_SDK_ROLES: 'owner',
+        },
+      );
+
+      fakeProc.emit('close', 0);
+      const result = await resultPromise;
+      expect(result.status).toBe('success');
+
+      expect(onProcess).toHaveBeenCalledWith(
+        fakeProc,
+        expect.any(String),
+        expect.any(String),
+        { drainFollowUpsAfterRun: true },
+      );
+    } finally {
+      if (previousCodexGoals === undefined) {
+        delete process.env.CODEX_GOALS;
+      } else {
+        process.env.CODEX_GOALS = previousCodexGoals;
+      }
+    }
+  });
 });

src/agent-runner.ts

@@ -61,6 +61,30 @@ export interface AgentOutput {
   };
 }
 
+interface ActiveProcessOptions {
+  drainFollowUpsAfterRun?: boolean;
+}
+
+function isCodexSdkRuntimeMode(
+  env: Record<string, string>,
+  input: AgentInput,
+): boolean {
+  const runtime = env.CODEX_RUNTIME?.trim().toLowerCase();
+  if (runtime !== 'sdk') return false;
+  if (input.prompt.trim() === '/compact') return false;
+  if (input.codexGoals === true || env.CODEX_GOALS === 'true') return false;
+
+  const rawRoles = env.CODEX_RUNTIME_SDK_ROLES?.trim();
+  if (!rawRoles) return true;
+  const role = input.roomRoleContext?.role?.trim().toLowerCase();
+  if (!role) return false;
+  return rawRoles
+    .split(',')
+    .map((item) => item.trim().toLowerCase())
+    .filter(Boolean)
+    .includes(role);
+}
+
 function readRoomSkillOverridesForRunner(
   chatJid: string,
 ): StoredRoomSkillOverride[] {
@@ -126,6 +150,7 @@ export async function runAgentProcess(
     proc: ChildProcess,
     processName: string,
     runtimeIpcDir: string,
+    options?: ActiveProcessOptions,
   ) => void,
   onOutput?: (output: AgentOutput) => Promise<void>,
   envOverrides?: Record<string, string>,
@@ -186,6 +211,11 @@ export async function runAgentProcess(
   const safeName = group.folder.replace(/[^a-zA-Z0-9-]/g, '-');
   const processSuffix = input.runId || `${Date.now()}`;
   const processName = `ejclaw-${safeName}-${processSuffix}`;
+  const activeProcessOptions: ActiveProcessOptions = {
+    drainFollowUpsAfterRun:
+      (group.agentType || 'claude-code') === 'codex' &&
+      isCodexSdkRuntimeMode(env, input),
+  };
   const finalizeCodexAuthSessionOnce = createCodexAuthSessionFinalizer(
     () => codexSessionAuth,
   );
@@ -228,7 +258,7 @@ export async function runAgentProcess(
       env,
     });
 
-    onProcess(proc, processName, env[EJCLAW_ENV.ipcDir]);
+    onProcess(proc, processName, env[EJCLAW_ENV.ipcDir], activeProcessOptions);
 
     const runnerInput: AgentInput = {
       ...input,

src/group-queue-state.ts

@@ -37,6 +37,7 @@ export interface GroupState {
   pendingTasks: QueuedTask[];
   process: ChildProcess | null;
   processName: string | null;
+  drainFollowUpsAfterRun: boolean;
   ipcDir: string | null;
   retryCount: number;
   retryTimer: ReturnType<typeof setTimeout> | null;
@@ -60,6 +61,7 @@ export function createGroupState(): GroupState {
     pendingTasks: [],
     process: null,
     processName: null,
+    drainFollowUpsAfterRun: false,
     ipcDir: null,
     retryCount: 0,
     retryTimer: null,
@@ -145,6 +147,7 @@ export function resetRunState(state: GroupState, groupJid: string): void {
   state.startedAt = null;
   state.process = null;
   state.processName = null;
+  state.drainFollowUpsAfterRun = false;
   state.ipcDir = null;
   state.directTerminalDeliveries.clear();
   transitionRunPhase(state, groupJid, 'idle');
@@ -160,7 +163,8 @@ export function assertRunPhaseInvariants(
         state.currentRunId != null ||
         state.runningTaskId != null ||
         state.process != null ||
-        state.processName != null
+        state.processName != null ||
+        state.drainFollowUpsAfterRun
       ) {
         logger.error(
           {
@@ -170,6 +174,7 @@ export function assertRunPhaseInvariants(
             runningTaskId: state.runningTaskId,
             hasProcess: state.process != null,
             processName: state.processName,
+            drainFollowUpsAfterRun: state.drainFollowUpsAfterRun,
           },
           'Invariant violation: idle phase has stale run/task ID or process',
         );

src/group-queue.test.ts

@@ -40,8 +40,6 @@ describe('GroupQueue', () => {
     vi.useRealTimers();
   });
 
-  // --- Single group at a time ---
-
   it('only runs one agent per group at a time', async () => {
     let concurrentCount = 0;
     let maxConcurrent = 0;
@@ -90,6 +88,7 @@ describe('GroupQueue', () => {
 
     releaseRun(true);
     await vi.advanceTimersByTimeAsync(10);
+    expect(processMessages).toHaveBeenCalledTimes(1);
   });
 
   it('does not pipe follow-up messages after stdin close was requested', async () => {
@@ -791,3 +790,46 @@ describe('GroupQueue close reason tracking', () => {
     expect(queue.getCloseReasonForRun('group1@g.us', activeRunId)).toBeNull();
   });
 });
+
+describe('GroupQueue SDK follow-up draining', () => {
+  let queue: GroupQueue;
+
+  beforeEach(() => {
+    vi.useFakeTimers();
+    vi.clearAllMocks();
+    queue = new GroupQueue();
+  });
+
+  afterEach(() => {
+    vi.useRealTimers();
+  });
+
+  it('drains follow-up messages after active processes that cannot consume queued IPC', async () => {
+    const ipcDir = '/tmp/ejclaw-test-data/ipc/group-folder';
+    let releaseRun!: (value: boolean) => void;
+    let activeRun = true;
+    const blocker = new Promise<boolean>((resolve) => {
+      releaseRun = resolve;
+    });
+
+    const processMessages = vi.fn(async () => {
+      if (!activeRun) return true;
+      activeRun = false;
+      return await blocker;
+    });
+
+    queue.setProcessMessagesFn(processMessages);
+    queue.enqueueMessageCheck('group1@g.us', ipcDir);
+    await vi.advanceTimersByTimeAsync(10);
+
+    queue.registerProcess('group1@g.us', {} as any, 'sdk-agent', ipcDir, {
+      drainFollowUpsAfterRun: true,
+    });
+    expect(queue.sendMessage('group1@g.us', '후속 메시지')).toBe(true);
+
+    releaseRun(true);
+    await vi.advanceTimersByTimeAsync(10);
+
+    expect(processMessages).toHaveBeenCalledTimes(2);
+  });
+});

src/group-queue.ts

@@ -65,7 +65,6 @@ export class GroupQueue {
     this.processMessagesFn = fn;
   }
 
-  /** Limit concurrency after restart to avoid API rate-limit storms. */
   enterRecoveryMode(): void {
     this.recoveryMode = true;
     logger.info(
@@ -101,7 +100,6 @@ export class GroupQueue {
 
     const state = this.getGroup(groupJid);
 
-    // Pre-set IPC dir so sendMessage can pipe follow-ups while agent process starts
     if (ipcDir && !state.ipcDir) {
       state.ipcDir = ipcDir;
     }
@@ -157,7 +155,6 @@ export class GroupQueue {
 
     const state = this.getGroup(groupJid);
 
-    // Prevent double-queuing: check both pending and currently-running task
     if (state.runningTaskId === taskId) {
       logger.debug({ groupJid, taskId }, 'Task already running, skipping');
       return;
@@ -196,7 +193,6 @@ export class GroupQueue {
       return;
     }
 
-    // Run immediately
     this.runTask(groupJid, { id: taskId, groupJid, fn }).catch((err) =>
       logger.error({ groupJid, taskId, err }, 'Unhandled error in runTask'),
     );
@@ -207,10 +203,12 @@ export class GroupQueue {
     proc: ChildProcess,
     processName: string,
     ipcDir?: string,
+    options?: { drainFollowUpsAfterRun?: boolean },
   ): void {
     const state = this.getGroup(groupJid);
     state.process = proc;
     state.processName = processName;
+    state.drainFollowUpsAfterRun = options?.drainFollowUpsAfterRun === true;
     if (ipcDir) state.ipcDir = ipcDir;
     logger.info(
       {
@@ -241,6 +239,7 @@ export class GroupQueue {
 
     try {
       const filename = queueFollowUpMessage(state.ipcDir, text);
+      state.pendingMessages ||= state.drainFollowUpsAfterRun;
       logger.info(
         {
           groupJid,
@@ -248,6 +247,7 @@ export class GroupQueue {
           ipcDir: state.ipcDir,
           textLength: text.length,
           filename,
+          pendingMessages: state.pendingMessages,
         },
         'Queued follow-up message for active agent',
       );

src/message-agent-executor-attempt-runner.ts

@@ -100,6 +100,7 @@ interface RunMessageAgentAttemptArgs {
       : never,
     processName: string,
     ipcDir?: string,
+    options?: { drainFollowUpsAfterRun?: boolean },
   ) => void;
   onOutput?: (output: AgentOutput) => Promise<void>;
   pairedExecutionLifecycle: {

src/message-agent-executor.ts

@@ -274,8 +274,8 @@ export async function runAgentForGroup(
         deps.persistSession(sessionFolder, sessionId);
         currentSessionId = sessionId;
       },
-      registerProcess: (proc, processName, ipcDir) =>
+      registerProcess: (proc, processName, ipcDir, options) =>
-        deps.queue.registerProcess(chatJid, proc, processName, ipcDir),
+        deps.queue.registerProcess(chatJid, proc, processName, ipcDir, options),
       onOutput,
       pairedExecutionLifecycle,
       log,