🍵 feat: Continue Late Steers in Warm Agent Runs (#15357)

* feat: continue late steers in warm agent runs

* chore: bump agents sdk to v3.7.9

* fix: guard terminal steer admission

api/server/controllers/agents/__tests__/client.steerWiring.spec.js

@@ -1,14 +1,20 @@
 const AgentClient = require('../client');
-const { isSteeringSupported, isSteerPreemptSupported } = require('@librechat/api');
+const {
+  isSteeringSupported,
+  isSteerPreemptSupported,
+  isSteerTerminalContinuationSupported,
+} = require('@librechat/api');
 
 jest.mock('@librechat/api', () => ({
   ...jest.requireActual('@librechat/api'),
   isSteeringSupported: jest.fn(() => true),
   isSteerPreemptSupported: jest.fn(() => true),
+  isSteerTerminalContinuationSupported: jest.fn(() => true),
 }));
 
 const mockIsSteeringSupported = isSteeringSupported;
 const mockIsPreemptSupported = isSteerPreemptSupported;
+const mockIsTerminalContinuationSupported = isSteerTerminalContinuationSupported;
 
 /** Minimal `this` for the wiring builder — it only reads these three. */
 function buildWiring(streamId, { jobCreatedAt = 1700000000000 } = {}) {
@@ -25,6 +31,7 @@ describe('AgentClient.buildSteerWiring — preempt capability gating', () => {
     jest.clearAllMocks();
     mockIsSteeringSupported.mockReturnValue(true);
     mockIsPreemptSupported.mockReturnValue(true);
+    mockIsTerminalContinuationSupported.mockReturnValue(true);
   });
 
   it('returns both boundary hooks and the poll when preempt is supported', () => {
@@ -33,6 +40,16 @@ describe('AgentClient.buildSteerWiring — preempt capability gating', () => {
     expect(typeof wiring.hook).toBe('function');
     expect(typeof wiring.preemptHook).toBe('function');
     expect(typeof wiring.preemption?.shouldPreempt).toBe('function');
+    expect(typeof wiring.terminalHook).toBe('function');
+  });
+
+  it('omits only terminal continuation when the SDK lacks Stop continuation', () => {
+    mockIsTerminalContinuationSupported.mockReturnValue(false);
+    const wiring = buildWiring('stream-terminal-unsupported');
+
+    expect(typeof wiring.hook).toBe('function');
+    expect(typeof wiring.preemptHook).toBe('function');
+    expect(wiring.terminalHook).toBeUndefined();
   });
 
   /**

api/server/controllers/agents/__tests__/request.partialDisconnect.spec.js

@@ -51,6 +51,7 @@ jest.mock('@librechat/api', () => ({
   exemptFromConcurrencyLimiter: jest.fn(() => false),
   toPendingSteer: jest.fn((item) => item),
   isSteerPreemptSupported: jest.fn(() => true),
+  isSteerTerminalContinuationSupported: jest.fn(() => false),
   buildRecoveredSteerPayload: jest.fn(() => null),
   deleteAgentCheckpoint: jest.fn(),
   getViolationInfo: jest.fn(() => ({

api/server/controllers/agents/__tests__/request.resumeMetadata.spec.js

@@ -254,6 +254,7 @@ jest.mock('@librechat/api', () => ({
   /** Recorded onto the job so the steer route can honour the OWNING replica's
    *  seal capability rather than its own probe. */
   isSteerPreemptSupported: jest.fn(() => true),
+  isSteerTerminalContinuationSupported: jest.fn(() => false),
   buildRecoveredSteerPayload: jest.fn((text, files) => {
     if (typeof text !== 'string' || (files != null && !Array.isArray(files))) {
       return null;

api/server/controllers/agents/__tests__/resume.spec.js

@@ -132,6 +132,7 @@ jest.mock('@librechat/api', () => ({
   decrementPendingRequest: (...args) => mockDecrementPendingRequest(...args),
   checkAndIncrementPendingRequest: (...args) => mockCheckAndIncrementPendingRequest(...args),
   isSteerPreemptSupported: jest.fn(() => true),
+  isSteerTerminalContinuationSupported: jest.fn(() => false),
   createMCPRuntimeRequestBody: ({ messageId, conversationId, parentMessageId }) => ({
     messageId,
     conversationId,

api/server/controllers/agents/client.js

@@ -65,9 +65,11 @@ const {
   createSteerIndexOffsetHandlers,
   createSteerDrainHook,
   createSteerPreemptBoundaryHook,
+  createSteerTerminalContinuationHook,
   createSteerPreemptPoll,
   isSteeringSupported,
   isSteerPreemptSupported,
+  isSteerTerminalContinuationSupported,
   buildSteerMedia,
   collectSteerStampTargets,
   stampSteerPartMedia,
@@ -557,9 +559,9 @@ class AgentClient extends BaseClient {
 
   /**
    * The `steering` fragment for `createRun`: the run-scoped PostToolBatch
-   * drain hook — plus, when the SDK can seal mid-stream, the PreemptBoundary
-   * twin and the preempt poll built from the SAME drain closures, so both
-   * boundaries inject byte-identical shapes. `undefined` when there is no
+   * drain hook — plus the capability-gated PreemptBoundary and terminal Stop
+   * twins built from the SAME drain closures, so every boundary injects
+   * byte-identical shapes. `undefined` when there is no
    * resumable job surface or the installed SDK cannot inject hook messages
    * (draining would drop them).
    *
@@ -592,6 +594,9 @@ class AgentClient extends BaseClient {
         preemptHook: createSteerPreemptBoundaryHook(drainOptions),
         preemption: createSteerPreemptPoll(streamId),
       }),
+      ...(isSteerTerminalContinuationSupported() && {
+        terminalHook: createSteerTerminalContinuationHook(drainOptions),
+      }),
     };
   }