Compare commits

...
Author SHA1 Message Date
Mohamed Boudra d1278aa856 Update tests for steering capability and send-path routing
Test drift surfaced by rebasing onto main:

- client SDK agent fixture now expects supportsSteering: false (the steering feature adds this capability flag, schema-defaulted).

- permission-response fake AgentManager gains startAgentRun, since the shared startAgentRun helper now routes through AgentManager.startAgentRun (the steer-vs-replace-vs-stream dispatch) instead of calling streamAgent/replaceAgentRun directly.
2026-05-31 10:27:13 +07:00
Mohamed Boudra 091350544d Fix finish notification steering dispatch 2026-05-31 10:13:31 +07:00
Mohamed Boudra bc825af843 Add steering follow-up support 2026-05-31 10:13:31 +07:00
23 changed files with 926 additions and 44 deletions
+1
View File
@@ -423,6 +423,7 @@ const AgentStreamViewComponent = forwardRef<AgentStreamViewHandle, AgentStreamVi
client={client}
isFirstInGroup={layoutItem.isFirstInUserGroup}
isLastInGroup={layoutItem.isLastInUserGroup}
deliveryHint={item.deliveryHint}
/>
);
},
+21 -1
View File
@@ -63,7 +63,11 @@ import Animated, {
import Svg, { Defs, LinearGradient as SvgLinearGradient, Rect, Stop } from "react-native-svg";
import { createMarkdownStyles } from "@/styles/markdown-styles";
import { Fonts } from "@/constants/theme";
import type { TodoEntry, UserMessageImageAttachment } from "@/types/stream";
import type {
TodoEntry,
UserMessageDeliveryHint,
UserMessageImageAttachment,
} from "@/types/stream";
import type { AgentAttachment } from "@getpaseo/protocol/messages";
import type { ToolCallDetail } from "@getpaseo/protocol/agent-types";
import { buildToolCallPresentation } from "@/tool-calls/presentation";
@@ -124,6 +128,7 @@ interface UserMessageProps {
isFirstInGroup?: boolean;
isLastInGroup?: boolean;
disableOuterSpacing?: boolean;
deliveryHint?: UserMessageDeliveryHint;
}
const MessageOuterSpacingContext = createContext(false);
@@ -434,6 +439,12 @@ const userMessageStylesheet = StyleSheet.create((theme) => ({
color: theme.colors.foregroundMuted,
fontSize: STREAM_METADATA_FONT_SIZE,
},
deliveryHintLabel: {
alignSelf: "flex-end",
marginTop: theme.spacing[1],
color: theme.colors.foregroundMuted,
fontSize: theme.fontSize.xs,
},
}));
function UserMessageAttachmentThumbnail({ image }: { image: UserMessageImageAttachment }) {
@@ -475,6 +486,7 @@ export const UserMessage = memo(function UserMessage({
isFirstInGroup = true,
isLastInGroup = true,
disableOuterSpacing,
deliveryHint,
}: UserMessageProps) {
const isCompact = useIsCompactFormFactor();
const [isHovered, setIsHovered] = useState(false);
@@ -589,6 +601,14 @@ export const UserMessage = memo(function UserMessage({
/>
</View>
) : null}
{deliveryHint === "steering" ? (
<Text
accessibilityLabel="Steering conversation"
style={userMessageStylesheet.deliveryHintLabel}
>
Steering conversation
</Text>
) : null}
</View>
</View>
);
+38
View File
@@ -417,6 +417,44 @@ describe("dispatchComposerAgentMessage", () => {
expect(client.calls[0]?.options.images).toEqual([]);
});
it("attaches a steering deliveryHint to the optimistic user_message when provided", async () => {
const client = createFakeSendClient();
const stream = createFakeStream();
await dispatchComposerAgentMessage({
client,
agentId: "agent",
text: "steer this",
attachments: [],
encodeImages: passthroughEncodeImages,
stream,
deliveryHint: "steering",
});
const tail = stream.tail.get("agent");
expect(tail).toHaveLength(1);
const userMessage = tail?.[0] as Extract<StreamItem, { kind: "user_message" }>;
expect(userMessage.deliveryHint).toBe("steering");
});
it("omits deliveryHint when none is supplied (no marker for non-steering sends)", async () => {
const client = createFakeSendClient();
const stream = createFakeStream();
await dispatchComposerAgentMessage({
client,
agentId: "agent",
text: "no steering",
attachments: [],
encodeImages: passthroughEncodeImages,
stream,
});
const tail = stream.tail.get("agent");
const userMessage = tail?.[0] as Extract<StreamItem, { kind: "user_message" }>;
expect(userMessage.deliveryHint).toBeUndefined();
});
it("serializes browser_element workspace attachments as text attachments at the wire boundary", async () => {
const client = createFakeSendClient();
const stream = createFakeStream();
+3
View File
@@ -14,6 +14,7 @@ import {
buildOptimisticUserMessage,
generateMessageId,
type StreamItem,
type UserMessageDeliveryHint,
type UserMessageItem,
} from "@/types/stream";
import type { PickedImageAttachmentInput } from "@/hooks/image-attachment-picker";
@@ -130,6 +131,7 @@ export interface DispatchComposerAgentMessageInput {
images: AttachmentMetadata[],
) => Promise<Array<{ data: string; mimeType: string }> | undefined>;
stream: AgentStreamWriter;
deliveryHint?: UserMessageDeliveryHint;
}
export async function dispatchComposerAgentMessage(
@@ -143,6 +145,7 @@ export async function dispatchComposerAgentMessage(
timestamp: new Date(),
images: wirePayload.images,
attachments: wirePayload.attachments,
deliveryHint: input.deliveryHint,
});
appendUserMessageToStream(input.agentId, userMessage, input.stream);
const imagesData = await input.encodeImages(wirePayload.images);
+4
View File
@@ -1070,6 +1070,9 @@ export function Composer({
setHead: (updater) => setAgentStreamHead(serverId, updater),
setTail: (updater) => setAgentStreamTail(serverId, updater),
};
const targetAgent = useSessionStore.getState().sessions[serverId]?.agents?.get(targetAgentId);
const isSteering =
targetAgent?.status === "running" && targetAgent.capabilities.supportsSteering === true;
await dispatchComposerAgentMessage({
client,
agentId: targetAgentId,
@@ -1077,6 +1080,7 @@ export function Composer({
attachments: sendAttachments,
encodeImages,
stream,
...(isSteering ? { deliveryHint: "steering" as const } : {}),
});
onAttentionPromptSend?.();
};
@@ -1700,6 +1700,9 @@ function SessionProviderInternal({ children, serverId, client }: SessionProvider
attachments?: AgentAttachment[],
) => {
const messageId = generateMessageId();
const targetAgent = useSessionStore.getState().sessions[serverId]?.agents?.get(agentId);
const isSteering =
targetAgent?.status === "running" && targetAgent.capabilities.supportsSteering === true;
const userMessage: StreamItem = {
kind: "user_message",
id: messageId,
@@ -1708,6 +1711,7 @@ function SessionProviderInternal({ children, serverId, client }: SessionProvider
optimistic: true,
...(images && images.length > 0 ? { images } : {}),
...(attachments && attachments.length > 0 ? { attachments } : {}),
...(isSteering ? { deliveryHint: "steering" as const } : {}),
};
// Append to head if streaming (keeps the user message with the current
+46
View File
@@ -765,6 +765,52 @@ describe("stream reducer canonical tool calls", () => {
assert.strictEqual(message.timestamp.getTime(), optimisticTimestamp.getTime());
});
it("preserves optimistic steering deliveryHint when authoritative user message arrives", () => {
const messageId = "msg-user-steering";
const initialState: StreamItem[] = [
{
kind: "user_message",
id: messageId,
text: "Steer to a different approach",
timestamp: new Date("2025-01-01T11:10:00Z"),
deliveryHint: "steering",
},
];
const event: AgentStreamEventPayload = {
type: "timeline",
provider: "claude",
item: {
type: "user_message",
text: "Steer to a different approach",
messageId,
},
};
const state = reduceStreamUpdate(initialState, event, new Date("2025-01-01T11:10:01Z"));
const message = state.find((item) => item.kind === "user_message");
assert.ok(message);
assert.strictEqual(message.deliveryHint, "steering");
});
it("does not invent a deliveryHint for canonical user messages without prior optimistic state", () => {
const event: AgentStreamEventPayload = {
type: "timeline",
provider: "claude",
item: {
type: "user_message",
text: "Hello there",
messageId: "canonical-only",
},
};
const state = reduceStreamUpdate([], event, new Date("2025-01-01T11:11:00Z"));
const message = state.find((item) => item.kind === "user_message");
assert.ok(message);
assert.strictEqual(message.deliveryHint, undefined);
});
it("keeps canonical assistant/user/assistant order during replay", () => {
const state: StreamItem[] = [
{
+12
View File
@@ -54,6 +54,8 @@ export type StreamItem =
export type UserMessageImageAttachment = AttachmentMetadata;
export type UserMessageDeliveryHint = "steering";
export interface UserMessageItem {
kind: "user_message";
id: string;
@@ -62,6 +64,13 @@ export interface UserMessageItem {
optimistic?: true;
images?: UserMessageImageAttachment[];
attachments?: AgentAttachment[];
/**
* Local-only marker set when this user message was sent optimistically
* while the agent was running and the provider supports steering. Never
* sent or received over the wire — preserved across canonical replays so
* the UI can keep showing the marker.
*/
deliveryHint?: UserMessageDeliveryHint;
}
export interface OptimisticUserMessageInput {
@@ -70,6 +79,7 @@ export interface OptimisticUserMessageInput {
timestamp: Date;
images?: UserMessageImageAttachment[];
attachments?: AgentAttachment[];
deliveryHint?: UserMessageDeliveryHint;
}
export type OptimisticUserMessagePlacement = "tail" | "active-head";
@@ -213,6 +223,7 @@ function buildUserMessageItem(input: {
...(input.optimistic.attachments && input.optimistic.attachments.length > 0
? { attachments: input.optimistic.attachments }
: {}),
...(input.optimistic.deliveryHint ? { deliveryHint: input.optimistic.deliveryHint } : {}),
};
}
@@ -235,6 +246,7 @@ export function buildOptimisticUserMessage(input: OptimisticUserMessageInput): U
...(input.attachments && input.attachments.length > 0
? { attachments: input.attachments }
: {}),
...(input.deliveryHint ? { deliveryHint: input.deliveryHint } : {}),
};
}
+1
View File
@@ -152,6 +152,7 @@ function createAgent(input: Partial<PaseoAgent> = {}): PaseoAgent {
supportsRewindBoth: false,
supportsRewindConversation: false,
supportsRewindFiles: false,
supportsSteering: false,
supportsToolInvocations: true,
},
currentModeId: null,
+1
View File
@@ -125,6 +125,7 @@ export interface AgentCapabilityFlags {
supportsRewindConversation?: boolean;
supportsRewindFiles?: boolean;
supportsRewindBoth?: boolean;
supportsSteering?: boolean;
}
export interface AgentPersistenceHandle {
@@ -1,5 +1,6 @@
import { describe, expect, test } from "vitest";
import {
AgentSnapshotPayloadSchema,
GetProvidersSnapshotResponseMessageSchema,
ProviderSnapshotEntrySchema,
ProvidersSnapshotUpdateMessageSchema,
@@ -16,6 +17,34 @@ describe("provider snapshot message schemas", () => {
expect(parsed.enabled).toBe(true);
});
test("defaults missing agent capability steering support to false", () => {
const parsed = AgentSnapshotPayloadSchema.parse({
id: "agent-1",
provider: "codex",
cwd: "/tmp/repo",
model: null,
createdAt: "2026-04-24T00:00:00.000Z",
updatedAt: "2026-04-24T00:00:00.000Z",
lastUserMessageAt: null,
status: "idle",
capabilities: {
supportsStreaming: true,
supportsSessionPersistence: true,
supportsDynamicModes: false,
supportsMcpServers: true,
supportsReasoningStream: true,
supportsToolInvocations: true,
},
currentModeId: null,
availableModes: [],
pendingPermissions: [],
persistence: null,
title: null,
});
expect(parsed.capabilities.supportsSteering).toBe(false);
});
test("preserves disabled provider snapshot entries", () => {
const parsed = ProviderSnapshotEntrySchema.parse({
provider: "claude",
+2
View File
@@ -251,6 +251,8 @@ const AgentCapabilityFlagsSchema: z.ZodType<AgentCapabilityFlags> = z.object({
supportsRewindFiles: z.boolean().optional().default(false),
// COMPAT(rewind): added in v0.1.X, drop when floor >= v0.1.X.
supportsRewindBoth: z.boolean().optional().default(false),
// COMPAT(steering): added in v0.1.X, drop when floor >= v0.1.X.
supportsSteering: z.boolean().optional().default(false),
});
const AgentUsageSchema: z.ZodType<AgentUsage> = z.object({
@@ -18,6 +18,7 @@ import type {
AgentLaunchContext,
AgentProvider,
AgentPersistenceHandle,
AgentPromptInput,
AgentRunResult,
AgentSession,
AgentSessionConfig,
@@ -3010,6 +3011,314 @@ test("replaceAgentRun does not emit idle or resolve waiters between interrupted
unsubscribe();
});
test("startAgentRun applies the replace-vs-stream policy in AgentManager", async () => {
const workdir = mkdtempSync(join(tmpdir(), "agent-manager-start-policy-"));
const storagePath = join(workdir, "agents");
const storage = new AgentStorage(storagePath, logger);
const manager = new AgentManager({
clients: {
codex: new TestAgentClient(),
},
registry: storage,
logger,
idFactory: () => "00000000-0000-4000-8000-000000000126",
});
const snapshot = await manager.createAgent({
provider: "codex",
cwd: workdir,
});
const replacementEvents = (async function* replacementEvents() {})();
const streamEvents = (async function* streamEvents() {})();
const hasInFlightRunSpy = vi.spyOn(manager, "hasInFlightRun").mockReturnValue(true);
const replaceAgentRunSpy = vi
.spyOn(manager, "replaceAgentRun")
.mockReturnValue(replacementEvents);
const streamAgentSpy = vi.spyOn(manager, "streamAgent").mockReturnValue(streamEvents);
const replacement = manager.startAgentRun(snapshot.id, "replace me", {
replaceRunning: true,
});
expect(replacement).toEqual({ outOfBand: false, events: replacementEvents });
expect(hasInFlightRunSpy).toHaveBeenCalledWith(snapshot.id);
expect(replaceAgentRunSpy).toHaveBeenCalledWith(snapshot.id, "replace me", undefined);
expect(streamAgentSpy).not.toHaveBeenCalled();
replaceAgentRunSpy.mockClear();
streamAgentSpy.mockClear();
const streamed = manager.startAgentRun(snapshot.id, "stream me", {
replaceRunning: false,
});
expect(streamed).toEqual({ outOfBand: false, events: streamEvents });
expect(streamAgentSpy).toHaveBeenCalledWith(snapshot.id, "stream me", undefined);
expect(replaceAgentRunSpy).not.toHaveBeenCalled();
});
test("startAgentRun steers a running foreground turn when the provider supports steering", async () => {
const workdir = mkdtempSync(join(tmpdir(), "agent-manager-steer-run-"));
const storagePath = join(workdir, "agents");
const storage = new AgentStorage(storagePath, logger);
const allowFirstRunToEnd = deferred<void>();
const steeredPrompts: AgentPromptInput[] = [];
let capturedSession: SteerableSession | null = null;
const steeringCapabilities = {
...TEST_CAPABILITIES,
supportsSteering: true,
} as const;
class SteerableSession implements AgentSession {
readonly provider = "codex" as const;
readonly capabilities = steeringCapabilities;
readonly id = randomUUID();
readonly interrupt = vi.fn(async () => undefined);
private subscribers = new Set<(event: AgentStreamEvent) => void>();
private turnIdCounter = 0;
async run(): Promise<AgentRunResult> {
return {
sessionId: this.id,
finalText: "",
timeline: [],
};
}
async startTurn(): Promise<{ turnId: string }> {
const turnId = `turn-${++this.turnIdCounter}`;
void (async () => {
this.pushEvent({ type: "turn_started", provider: this.provider, turnId });
await allowFirstRunToEnd.promise;
this.pushEvent({ type: "turn_completed", provider: this.provider, turnId });
})();
return { turnId };
}
async steerTurn(prompt: AgentPromptInput): Promise<void> {
steeredPrompts.push(prompt);
}
subscribe(callback: (event: AgentStreamEvent) => void): () => void {
this.subscribers.add(callback);
return () => {
this.subscribers.delete(callback);
};
}
private pushEvent(event: AgentStreamEvent): void {
for (const callback of this.subscribers) {
callback(event);
}
}
async *streamHistory(): AsyncGenerator<AgentStreamEvent> {}
async getRuntimeInfo() {
return { provider: this.provider, sessionId: this.id, model: null, modeId: null };
}
async getAvailableModes() {
return [];
}
async getCurrentMode() {
return null;
}
async setMode(): Promise<void> {}
getPendingPermissions() {
return [];
}
async respondToPermission(): Promise<void> {}
describePersistence() {
return { provider: this.provider, sessionId: this.id };
}
async close(): Promise<void> {}
}
class SteerableClient extends TestAgentClient {
readonly capabilities = steeringCapabilities;
override async createSession(_config: AgentSessionConfig): Promise<AgentSession> {
capturedSession = new SteerableSession();
return capturedSession;
}
}
const manager = new AgentManager({
clients: {
codex: new SteerableClient(),
},
registry: storage,
logger,
idFactory: () => "00000000-0000-4000-8000-000000000127",
});
const snapshot = await manager.createAgent({
provider: "codex",
cwd: workdir,
});
const firstRun = manager.streamAgent(snapshot.id, "first run");
const firstRunDrain = (async () => {
for await (const _event of firstRun) {
// Drain the original foreground run.
}
})();
await manager.waitForAgentRunStart(snapshot.id);
const steered = manager.startAgentRun(snapshot.id, "steer this", {
replaceRunning: true,
});
expect(steered.outOfBand).toBe(false);
if (!steered.outOfBand) {
for await (const _event of steered.events) {
// Steering produces no foreground stream events.
}
}
expect(steeredPrompts).toEqual(["steer this"]);
expect(capturedSession?.interrupt).not.toHaveBeenCalled();
expect(manager.getAgent(snapshot.id)?.lifecycle).toBe("running");
allowFirstRunToEnd.resolve();
await firstRunDrain;
});
test("startAgentRun replaces a running foreground turn when steering is unsupported", async () => {
const workdir = mkdtempSync(join(tmpdir(), "agent-manager-unsupported-steer-run-"));
const storagePath = join(workdir, "agents");
const storage = new AgentStorage(storagePath, logger);
const allowFirstRunToCancel = deferred<void>();
const allowSecondRunToEnd = deferred<void>();
let startTurnCount = 0;
let interruptCount = 0;
class UnsupportedSteeringSession implements AgentSession {
readonly provider = "codex" as const;
readonly capabilities = TEST_CAPABILITIES;
readonly id = randomUUID();
private subscribers = new Set<(event: AgentStreamEvent) => void>();
async run(): Promise<AgentRunResult> {
return {
sessionId: this.id,
finalText: "",
timeline: [],
};
}
async startTurn(): Promise<{ turnId: string }> {
startTurnCount += 1;
const turnId = `turn-${startTurnCount}`;
const turnNumber = startTurnCount;
void (async () => {
this.pushEvent({ type: "turn_started", provider: this.provider, turnId });
if (turnNumber === 1) {
await allowFirstRunToCancel.promise;
this.pushEvent({
type: "turn_canceled",
provider: this.provider,
reason: "interrupted",
turnId,
});
return;
}
await allowSecondRunToEnd.promise;
this.pushEvent({ type: "turn_completed", provider: this.provider, turnId });
})();
return { turnId };
}
subscribe(callback: (event: AgentStreamEvent) => void): () => void {
this.subscribers.add(callback);
return () => {
this.subscribers.delete(callback);
};
}
private pushEvent(event: AgentStreamEvent): void {
for (const callback of this.subscribers) {
callback(event);
}
}
async *streamHistory(): AsyncGenerator<AgentStreamEvent> {}
async getRuntimeInfo() {
return { provider: this.provider, sessionId: this.id, model: null, modeId: null };
}
async getAvailableModes() {
return [];
}
async getCurrentMode() {
return null;
}
async setMode(): Promise<void> {}
getPendingPermissions() {
return [];
}
async respondToPermission(): Promise<void> {}
describePersistence() {
return { provider: this.provider, sessionId: this.id };
}
async interrupt(): Promise<void> {
interruptCount += 1;
allowFirstRunToCancel.resolve();
}
async close(): Promise<void> {}
}
class UnsupportedSteeringClient extends TestAgentClient {
override async createSession(_config: AgentSessionConfig): Promise<AgentSession> {
return new UnsupportedSteeringSession();
}
}
const manager = new AgentManager({
clients: {
codex: new UnsupportedSteeringClient(),
},
registry: storage,
logger,
idFactory: () => "00000000-0000-4000-8000-000000000128",
});
const snapshot = await manager.createAgent({
provider: "codex",
cwd: workdir,
});
const firstRun = manager.streamAgent(snapshot.id, "first run");
const firstRunDrain = (async () => {
for await (const _event of firstRun) {
// Drain the original foreground run.
}
})();
await manager.waitForAgentRunStart(snapshot.id);
const replacement = manager.startAgentRun(snapshot.id, "replace this", {
replaceRunning: true,
});
expect(replacement.outOfBand).toBe(false);
const replacementDrain =
replacement.outOfBand === false
? (async () => {
for await (const _event of replacement.events) {
// Drain the replacement foreground run.
}
})()
: Promise.resolve();
await manager.waitForAgentRunStart(snapshot.id);
expect(interruptCount).toBe(1);
expect(startTurnCount).toBe(2);
allowSecondRunToEnd.resolve();
await firstRunDrain;
await replacementDrain;
});
test("replaceAgentRun stays running when a stale old terminal arrives before the replacement turn is current", async () => {
const workdir = mkdtempSync(join(tmpdir(), "agent-manager-replace-stale-terminal-"));
const storagePath = join(workdir, "agents");
@@ -205,6 +205,15 @@ export interface WaitForAgentStartOptions {
signal?: AbortSignal;
}
export interface StartAgentRunOptions {
replaceRunning?: boolean;
runOptions?: AgentRunOptions;
}
export type StartAgentRunResult =
| { outOfBand: true }
| { outOfBand: false; events: AsyncGenerator<AgentStreamEvent> };
type AttentionState =
| { requiresAttention: false }
| {
@@ -1482,6 +1491,55 @@ export class AgentManager {
return true;
}
startAgentRun(
agentId: string,
prompt: AgentPromptInput,
options?: StartAgentRunOptions,
): StartAgentRunResult {
// Out-of-band commands (e.g. /goal pause) must run WITHOUT canceling an
// in-flight turn. Keeping this policy here makes every send surface share
// the same replace-vs-stream decision.
if (this.tryRunOutOfBand(agentId, prompt)) {
return { outOfBand: true };
}
const shouldReplace = Boolean(options?.replaceRunning && this.hasInFlightRun(agentId));
const runOptions = options?.runOptions;
if (shouldReplace && this.canSteerActiveForegroundRun(agentId)) {
const events = this.steerAgentRun(agentId, prompt, runOptions);
return { outOfBand: false, events };
}
const events = shouldReplace
? this.replaceAgentRun(agentId, prompt, runOptions)
: this.streamAgent(agentId, prompt, runOptions);
return { outOfBand: false, events };
}
private canSteerActiveForegroundRun(agentId: string): boolean {
const agent = this.agents.get(agentId);
return Boolean(
agent?.activeForegroundTurnId &&
agent.capabilities.supportsSteering === true &&
agent.session.steerTurn,
);
}
private async *steerAgentRun(
agentId: string,
prompt: AgentPromptInput,
options?: AgentRunOptions,
): AsyncGenerator<AgentStreamEvent> {
const agent = this.requireSessionAgent(agentId);
if (!agent.session.steerTurn) {
throw new Error(`Agent ${agentId} does not support steering`);
}
await agent.session.steerTurn(prompt, options);
this.touchUpdatedAt(agent);
this.emitState(agent);
yield* [];
}
async appendTimelineItem(agentId: string, item: AgentTimelineItem): Promise<void> {
const agent = this.requireAgent(agentId);
this.touchUpdatedAt(agent);
@@ -10,6 +10,9 @@ import {
} from "./agent-prompt.js";
import type { AgentManagerEvent, ManagedAgent } from "./agent-manager.js";
const CHILD_AGENT_ID = "11111111-1111-4111-8111-111111111111";
const CALLER_AGENT_ID = "22222222-2222-4222-8222-222222222222";
test("isSystemInjectedEnvelope matches the envelope formatSystemNotificationPrompt produces", () => {
expect(isSystemInjectedEnvelope(formatSystemNotificationPrompt("child finished"))).toBe(true);
expect(isSystemInjectedEnvelope("hello world")).toBe(false);
@@ -19,27 +22,24 @@ it("does not notify archived callers", async () => {
let subscriber: ((event: AgentManagerEvent) => void) | null = null;
const childAgent: ManagedAgent = Object.create(null);
Reflect.set(childAgent, "id", "child-agent");
Reflect.set(childAgent, "id", CHILD_AGENT_ID);
Reflect.set(childAgent, "lifecycle", "idle");
Reflect.set(childAgent, "config", { title: "Child Agent" });
const callerAgent: ManagedAgent = Object.create(null);
Reflect.set(callerAgent, "id", "caller-agent");
Reflect.set(callerAgent, "id", CALLER_AGENT_ID);
Reflect.set(callerAgent, "lifecycle", "idle");
Reflect.set(callerAgent, "config", { title: "Caller Agent" });
const streamAgentSpy = vi.fn(() => (async function* noop() {})());
const replaceAgentRunSpy = vi.fn(() => (async function* noop() {})());
const agentManager: AgentManager = Object.create(AgentManager.prototype);
Reflect.set(
agentManager,
"getAgent",
vi.fn((agentId: string) => {
if (agentId === "child-agent") {
if (agentId === CHILD_AGENT_ID) {
return childAgent;
}
if (agentId === "caller-agent") {
if (agentId === CALLER_AGENT_ID) {
return callerAgent;
}
return null;
@@ -55,12 +55,14 @@ it("does not notify archived callers", async () => {
};
}),
);
Reflect.set(agentManager, "hasInFlightRun", vi.fn().mockReturnValue(false));
Reflect.set(agentManager, "streamAgent", streamAgentSpy);
Reflect.set(agentManager, "replaceAgentRun", replaceAgentRunSpy);
const startAgentRunSpy = vi.fn(() => ({
outOfBand: false,
events: (async function* noop() {})(),
}));
Reflect.set(agentManager, "startAgentRun", startAgentRunSpy);
const agentStorageGetSpy = vi.fn(async (agentId: string) =>
agentId === "caller-agent" ? { archivedAt: "2024-01-01" } : null,
agentId === CALLER_AGENT_ID ? { archivedAt: "2024-01-01" } : null,
);
const agentStorage: AgentStorage = Object.create(AgentStorage.prototype);
Reflect.set(agentStorage, "get", agentStorageGetSpy);
@@ -68,8 +70,8 @@ it("does not notify archived callers", async () => {
setupFinishNotification({
agentManager,
agentStorage,
childAgentId: "child-agent",
callerAgentId: "caller-agent",
childAgentId: CHILD_AGENT_ID,
callerAgentId: CALLER_AGENT_ID,
logger: createTestLogger(),
});
@@ -88,9 +90,92 @@ it("does not notify archived callers", async () => {
});
await vi.waitFor(() => {
expect(agentStorageGetSpy).toHaveBeenCalledWith("caller-agent");
expect(agentStorageGetSpy).toHaveBeenCalledWith(CALLER_AGENT_ID);
});
expect(streamAgentSpy).not.toHaveBeenCalled();
expect(replaceAgentRunSpy).not.toHaveBeenCalled();
expect(startAgentRunSpy).not.toHaveBeenCalled();
});
it("uses AgentManager startAgentRun for finish notifications", async () => {
let subscriber: ((event: AgentManagerEvent) => void) | null = null;
const childAgent: ManagedAgent = Object.create(null);
Reflect.set(childAgent, "id", CHILD_AGENT_ID);
Reflect.set(childAgent, "lifecycle", "idle");
Reflect.set(childAgent, "config", { title: "Child Agent" });
const callerAgent: ManagedAgent = Object.create(null);
Reflect.set(callerAgent, "id", CALLER_AGENT_ID);
Reflect.set(callerAgent, "lifecycle", "idle");
Reflect.set(callerAgent, "config", { title: "Caller Agent" });
const startAgentRunSpy = vi.fn(() => ({
outOfBand: false,
events: (async function* noop() {})(),
}));
const agentManager: AgentManager = Object.create(AgentManager.prototype);
Reflect.set(
agentManager,
"getAgent",
vi.fn((agentId: string) => {
if (agentId === CHILD_AGENT_ID) {
return childAgent;
}
if (agentId === CALLER_AGENT_ID) {
return callerAgent;
}
return null;
}),
);
Reflect.set(
agentManager,
"subscribe",
vi.fn((callback: (event: AgentManagerEvent) => void) => {
subscriber = callback;
return () => {
subscriber = null;
};
}),
);
Reflect.set(agentManager, "startAgentRun", startAgentRunSpy);
const agentStorage: AgentStorage = Object.create(AgentStorage.prototype);
Reflect.set(
agentStorage,
"get",
vi.fn(async (agentId: string) =>
agentId === CHILD_AGENT_ID ? { title: "Child Agent" } : null,
),
);
setupFinishNotification({
agentManager,
agentStorage,
childAgentId: CHILD_AGENT_ID,
callerAgentId: CALLER_AGENT_ID,
logger: createTestLogger(),
});
expect(subscriber).not.toBeNull();
childAgent.lifecycle = "running";
subscriber?.({
type: "agent_state",
agent: childAgent,
});
childAgent.lifecycle = "idle";
subscriber?.({
type: "agent_state",
agent: childAgent,
});
await vi.waitFor(() => {
expect(startAgentRunSpy).toHaveBeenCalledWith(
CALLER_AGENT_ID,
`<paseo-system>\nAgent ${CHILD_AGENT_ID} (Child Agent) finished.\n</paseo-system>`,
{ replaceRunning: true },
);
});
});
@@ -1,19 +1,11 @@
import type { Logger } from "pino";
import type { AgentPromptInput, AgentRunOptions } from "./agent-sdk-types.js";
import type { AgentManager, ManagedAgent } from "./agent-manager.js";
import type { AgentManager, ManagedAgent, StartAgentRunOptions } from "./agent-manager.js";
import type { AgentStorage } from "./agent-storage.js";
import { ensureAgentLoaded } from "./agent-loading.js";
export type AgentRunController = Pick<
AgentManager,
"getAgent" | "tryRunOutOfBand" | "hasInFlightRun" | "replaceAgentRun" | "streamAgent"
>;
export interface StartAgentRunOptions {
replaceRunning?: boolean;
runOptions?: AgentRunOptions;
}
export type AgentRunController = Pick<AgentManager, "getAgent" | "startAgentRun">;
export function startAgentRun(
agentManager: AgentRunController,
@@ -35,23 +27,17 @@ export function startAgentRun(
},
"agent.session.start_stream.request",
);
// Out-of-band commands (e.g. /goal pause) must run WITHOUT canceling an
// in-flight turn — replaceAgentRun would interrupt the running turn. The
// intercept lives at this layer so it covers every prompt entrypoint.
if (agentManager.tryRunOutOfBand(agentId, prompt)) {
const result = agentManager.startAgentRun(agentId, prompt, options);
if (result.outOfBand) {
return { outOfBand: true };
}
const shouldReplace = Boolean(options?.replaceRunning && agentManager.hasInFlightRun(agentId));
const runOptions = options?.runOptions;
const iterator = shouldReplace
? agentManager.replaceAgentRun(agentId, prompt, runOptions)
: agentManager.streamAgent(agentId, prompt, runOptions);
const iterator = result.events;
logger.trace(
{
agentId,
provider: snapshot?.provider,
providerSessionId: snapshot?.persistence?.sessionId ?? undefined,
shouldReplace,
},
"agent.session.start_stream.iterator_returned",
);
@@ -153,6 +153,7 @@ export interface AgentCapabilityFlags {
supportsRewindConversation?: boolean;
supportsRewindFiles?: boolean;
supportsRewindBoth?: boolean;
supportsSteering?: boolean;
}
export interface AgentPersistenceHandle {
@@ -558,6 +559,7 @@ export interface AgentSession {
readonly features?: AgentFeature[];
run(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<AgentRunResult>;
startTurn(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<{ turnId: string }>;
steerTurn?(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<void>;
subscribe(callback: (event: AgentStreamEvent) => void): () => void;
streamHistory(): AsyncGenerator<AgentStreamEvent>;
getRuntimeInfo(): Promise<AgentRuntimeInfo>;
@@ -8,6 +8,7 @@ import type {
AgentPermissionResponse,
} from "./agent-sdk-types.js";
import type { AgentStreamEvent } from "../messages.js";
import type { StartAgentRunOptions, StartAgentRunResult } from "./agent-manager.js";
import { respondToAgentPermission } from "./permission-response.js";
class FakePermissionAgentManager {
@@ -61,6 +62,21 @@ class FakePermissionAgentManager {
this.replacementRuns.push({ agentId, prompt, options });
return emptyAgentStream();
}
startAgentRun(
agentId: string,
prompt: AgentPromptInput,
options?: StartAgentRunOptions,
): StartAgentRunResult {
if (this.tryRunOutOfBand()) {
return { outOfBand: true };
}
const shouldReplace = Boolean(options?.replaceRunning && this.hasInFlightRun());
const events = shouldReplace
? this.replaceAgentRun(agentId, prompt, options?.runOptions)
: this.streamAgent(agentId, prompt, options?.runOptions);
return { outOfBand: false, events };
}
}
async function* emptyAgentStream(): AsyncGenerator<AgentStreamEvent> {}
@@ -351,6 +351,7 @@ export function wrapSessionProvider(provider: AgentProvider, inner: AgentSession
},
run: (prompt, options) => inner.run(prompt, options),
startTurn: (prompt, options) => inner.startTurn(prompt, options),
steerTurn: inner.steerTurn?.bind(inner),
subscribe: (callback) => inner.subscribe((event) => callback(mapStreamEvent(provider, event))),
async *streamHistory() {
for await (const event of inner.streamHistory()) {
@@ -2,7 +2,7 @@ import * as fs from "node:fs/promises";
import * as os from "node:os";
import * as path from "node:path";
import { afterEach, describe, expect, test, vi } from "vitest";
import type { SDKMessage } from "@anthropic-ai/claude-agent-sdk";
import type { SDKMessage, SDKUserMessage } from "@anthropic-ai/claude-agent-sdk";
import { createTestLogger } from "../../../../test-utils/test-logger.js";
import * as executableUtils from "../../../../utils/executable.js";
@@ -749,7 +749,10 @@ describe("ClaudeAgentSession context window usage", () => {
return session as unknown as TestClaudeSession;
}
function createQueryFactoryForTurns(turns: Array<Array<Record<string, unknown>>>) {
function createQueryFactoryForTurns(
turns: Array<Array<Record<string, unknown>>>,
onPrompt?: (prompt: unknown) => void,
) {
return vi.fn(({ prompt }: { prompt: AsyncIterable<unknown> }) => {
const queuedMessages: Array<Record<string, unknown>> = [];
const waiters: Array<() => void> = [];
@@ -768,6 +771,7 @@ describe("ClaudeAgentSession context window usage", () => {
void (async () => {
for await (const _prompt of prompt) {
onPrompt?.(_prompt);
const turnMessages = turns[turnIndex] ?? [];
turnIndex += 1;
for (const message of turnMessages) {
@@ -1003,6 +1007,42 @@ describe("ClaudeAgentSession context window usage", () => {
}
});
test("steerTurn pushes a priority next user message into the existing input stream", async () => {
const capturedPrompts: SDKUserMessage[] = [];
const queryFactory = createQueryFactoryForTurns([], (prompt) => {
capturedPrompts.push(prompt as SDKUserMessage);
});
const client = new ClaudeAgentClient({ logger, queryFactory });
const session = await client.createSession({
provider: "claude",
cwd: process.cwd(),
});
try {
await session.startTurn("first prompt");
await vi.waitFor(() => {
expect(capturedPrompts).toHaveLength(1);
});
await expect(session.steerTurn?.("steered prompt")).resolves.toBeUndefined();
await vi.waitFor(() => {
expect(capturedPrompts).toHaveLength(2);
});
expect(capturedPrompts[1]).toMatchObject({
type: "user",
priority: "next",
message: {
role: "user",
content: [{ type: "text", text: "steered prompt" }],
},
});
expect(queryFactory).toHaveBeenCalledTimes(1);
} finally {
await session.close();
}
});
test("convertUsage includes contextWindowMaxTokens and derives used tokens from result usage as initial fallback", async () => {
const session = await createSessionForTest();
@@ -216,6 +216,7 @@ const CLAUDE_CAPABILITIES: AgentCapabilityFlags = {
supportsRewindConversation: true,
supportsRewindFiles: true,
supportsRewindBoth: true,
supportsSteering: true,
};
const DEFAULT_MODES: AgentMode[] = [
@@ -1789,6 +1790,20 @@ class ClaudeAgentSession implements AgentSession {
return { turnId };
}
async steerTurn(prompt: AgentPromptInput, _options?: AgentRunOptions): Promise<void> {
if (this.closed) {
throw new Error("Claude session is closed");
}
const sdkMessage = this.toSdkUserMessage(prompt);
sdkMessage.priority = "next";
await this.ensureQuery();
if (!this.input) {
throw new Error("Claude session input stream not initialized");
}
this.input.push(sdkMessage);
}
subscribe(callback: (event: AgentStreamEvent) => void): () => void {
this.subscribers.add(callback);
return () => {
@@ -63,6 +63,7 @@ type CodexTestSession = AgentSession & {
connected: boolean;
currentThreadId: string | null;
activeForegroundTurnId: string | null;
activeAppServerTurnId: string | null;
client: CodexClientLike | null;
};
@@ -99,6 +100,7 @@ function createSession(
session.connected = true;
session.currentThreadId = "test-thread";
session.activeForegroundTurnId = "test-turn";
session.activeAppServerTurnId = "test-turn";
return session;
}
@@ -395,6 +397,12 @@ describe("Codex app-server provider", () => {
);
});
test("advertises steering support for app-server sessions", () => {
const session = createSession();
expect(session.capabilities.supportsSteering).toBe(true);
});
test("passes ephemeral: true to thread/start when constructed as ephemeral", async () => {
const requests: Array<{ method: string; params: unknown }> = [];
const fakeClient: CodexClientLike = {
@@ -880,6 +888,161 @@ describe("Codex app-server provider", () => {
);
});
test("steers against the app-server turn id returned by turn/start", async () => {
const session = createSession();
const request = vi.fn(async (method: string) => {
if (method === "thread/loaded/list") {
return { data: ["test-thread"] };
}
if (method === "turn/start") {
return { turn: { id: "app-server-turn-from-start" } };
}
if (method === "turn/steer") {
return { turnId: "app-server-turn-from-start" };
}
throw new Error(`Unexpected request: ${method}`);
});
session.activeForegroundTurnId = null;
session.client = createStub<CodexClientLike>({ request });
const started = await session.startTurn("Start the turn");
expect(started.turnId).not.toBe("app-server-turn-from-start");
await session.steerTurn!("Steer the active turn");
expect(request).toHaveBeenCalledWith(
"turn/steer",
expect.objectContaining({
expectedTurnId: "app-server-turn-from-start",
}),
);
});
test("waits for the app-server turn id before steering Codex", async () => {
const session = createSession();
const request = vi.fn(async (method: string) => {
if (method === "thread/loaded/list") {
return { data: ["test-thread"] };
}
if (method === "turn/start") {
return {};
}
if (method === "turn/steer") {
return {};
}
throw new Error(`Unexpected request: ${method}`);
});
session.activeForegroundTurnId = null;
session.client = createStub<CodexClientLike>({ request });
await session.startTurn("Start the turn");
await expect(session.steerTurn!("Too early")).rejects.toThrow(
"Cannot steer Codex turn without an active app-server turn",
);
asInternals(session).handleNotification("turn/started", {
turn: { id: "app-server-turn-from-notification" },
});
await session.steerTurn!("Steer after notification");
expect(request).toHaveBeenCalledWith(
"turn/steer",
expect.objectContaining({
expectedTurnId: "app-server-turn-from-notification",
}),
);
});
test("sends turn/steer with thread id, built input, and active foreground turn id", async () => {
const session = createSession();
const request = vi.fn(async (method: string) => {
if (method === "turn/steer") {
return {};
}
throw new Error(`Unexpected request: ${method}`);
});
session.client = createStub<CodexClientLike>({ request });
session.currentThreadId = "thread-for-steer";
session.activeForegroundTurnId = "paseo-foreground-turn-for-steer";
session.activeAppServerTurnId = "app-server-turn-for-steer";
await session.steerTurn!("Use this guidance next.");
expect(request).toHaveBeenCalledTimes(1);
expect(request).toHaveBeenCalledWith("turn/steer", {
threadId: "thread-for-steer",
expectedTurnId: "app-server-turn-for-steer",
input: [
{
type: "text",
text: "Use this guidance next.",
text_elements: [],
},
],
});
});
test("does not start a new turn when steering Codex", async () => {
const session = createSession();
const request = vi.fn(async (method: string) => {
if (method === "turn/steer") {
return {};
}
throw new Error(`Unexpected request: ${method}`);
});
session.client = createStub<CodexClientLike>({ request });
await session.steerTurn!("Keep going with this constraint.");
expect(request.mock.calls.some(([method]) => method === "turn/start")).toBe(false);
expect(session.activeForegroundTurnId).toBe("test-turn");
});
test("fails Codex steering without an initialized client", async () => {
const session = createSession();
session.client = null;
session.connected = true;
await expect(session.steerTurn!("hello")).rejects.toThrow("Codex client not initialized");
});
test("fails Codex steering without a current thread", async () => {
const session = createSession();
session.currentThreadId = null;
session.client = createStub<CodexClientLike>({
request: vi.fn(async () => ({})),
});
await expect(session.steerTurn!("hello")).rejects.toThrow(
"Cannot steer Codex turn without an active thread",
);
});
test("fails Codex steering without an active foreground turn", async () => {
const session = createSession();
session.activeForegroundTurnId = null;
session.client = createStub<CodexClientLike>({
request: vi.fn(async () => ({})),
});
await expect(session.steerTurn!("hello")).rejects.toThrow(
"Cannot steer Codex turn without an active foreground turn",
);
});
test("fails Codex steering without an active app-server turn id", async () => {
const session = createSession();
session.activeAppServerTurnId = null;
session.client = createStub<CodexClientLike>({
request: vi.fn(async () => ({})),
});
await expect(session.steerTurn!("hello")).rejects.toThrow(
"Cannot steer Codex turn without an active app-server turn",
);
});
test("resolves Codex skill slash commands into app-server skill input", async () => {
const session = createSession();
const request = vi.fn(async (method: string) => {
@@ -171,6 +171,7 @@ const CODEX_APP_SERVER_CAPABILITIES: AgentCapabilityFlags = {
supportsRewindConversation: true,
supportsRewindFiles: false,
supportsRewindBoth: false,
supportsSteering: true,
};
const CODEX_MODES: AgentMode[] = [
@@ -2888,6 +2889,7 @@ export class CodexAppServerAgentSession implements AgentSession {
private readonly subscribers = new Set<(event: AgentStreamEvent) => void>();
private nextTurnOrdinal = 0;
private activeForegroundTurnId: string | null = null;
private activeAppServerTurnId: string | null = null;
private cachedRuntimeInfo: AgentRuntimeInfo | null = null;
private serviceTier: "fast" | null = null;
private planModeEnabled = false;
@@ -3481,6 +3483,13 @@ export class CodexAppServerAgentSession implements AgentSession {
);
}
private async buildEffectivePromptInput(prompt: AgentPromptInput): Promise<CodexPromptInput> {
const slashCommand = await this.resolveSlashCommandInvocation(prompt);
return slashCommand
? await this.buildCommandPromptInput(slashCommand.commandName, slashCommand.args)
: prompt;
}
async run(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<AgentRunResult> {
return runProviderTurn({
prompt,
@@ -3513,10 +3522,7 @@ export class CodexAppServerAgentSession implements AgentSession {
throw new Error("Codex client not initialized");
}
const slashCommand = await this.resolveSlashCommandInvocation(prompt);
const effectivePrompt = slashCommand
? await this.buildCommandPromptInput(slashCommand.commandName, slashCommand.args)
: prompt;
const effectivePrompt = await this.buildEffectivePromptInput(prompt);
if (this.currentThreadId) {
await this.ensureThreadLoaded();
@@ -3528,6 +3534,7 @@ export class CodexAppServerAgentSession implements AgentSession {
const turnId = this.createTurnId();
this.activeForegroundTurnId = turnId;
this.activeAppServerTurnId = null;
try {
this.logTurnStartSummary({
@@ -3539,9 +3546,15 @@ export class CodexAppServerAgentSession implements AgentSession {
hasDeveloperInstructions: turnStart.hasDeveloperInstructions,
hasCodexConfig: turnStart.hasCodexConfig,
});
await this.client.request("turn/start", turnStart.params, TURN_START_TIMEOUT_MS);
const response = await this.client.request(
"turn/start",
turnStart.params,
TURN_START_TIMEOUT_MS,
);
this.rememberStartedAppServerTurn(response);
} catch (error) {
this.activeForegroundTurnId = null;
this.activeAppServerTurnId = null;
throw error;
}
@@ -3583,6 +3596,36 @@ export class CodexAppServerAgentSession implements AgentSession {
};
}
private rememberStartedAppServerTurn(response: unknown): void {
const record = toObjectRecord(response);
const turn = toObjectRecord(record?.turn);
this.activeAppServerTurnId = typeof turn?.id === "string" ? turn.id : null;
}
async steerTurn(prompt: AgentPromptInput, _options?: AgentRunOptions): Promise<void> {
await this.connect();
if (!this.client) {
throw new Error("Codex client not initialized");
}
if (!this.currentThreadId) {
throw new Error("Cannot steer Codex turn without an active thread");
}
if (!this.activeForegroundTurnId) {
throw new Error("Cannot steer Codex turn without an active foreground turn");
}
if (!this.activeAppServerTurnId) {
throw new Error("Cannot steer Codex turn without an active app-server turn");
}
const effectivePrompt = await this.buildEffectivePromptInput(prompt);
const input = await this.buildUserInput(effectivePrompt);
await this.client.request("turn/steer", {
threadId: this.currentThreadId,
input,
expectedTurnId: this.activeAppServerTurnId,
});
}
subscribe(callback: (event: AgentStreamEvent) => void): () => void {
this.subscribers.add(callback);
return () => {
@@ -3920,6 +3963,7 @@ export class CodexAppServerAgentSession implements AgentSession {
this.resolvedPermissionRequests.clear();
this.subscribers.clear();
this.activeForegroundTurnId = null;
this.activeAppServerTurnId = null;
if (this.client) {
await this.client.dispose();
}
@@ -4541,6 +4585,7 @@ export class CodexAppServerAgentSession implements AgentSession {
return;
}
this.currentTurnId = parsed.turnId;
this.activeAppServerTurnId = parsed.turnId;
this.resetTurnTrackingState();
this.emitEvent({ type: "turn_started", provider: CODEX_PROVIDER });
}
@@ -4578,6 +4623,7 @@ export class CodexAppServerAgentSession implements AgentSession {
});
}
this.activeForegroundTurnId = null;
this.activeAppServerTurnId = null;
this.resetTurnTrackingState();
}