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
38 changed files with 1106 additions and 1328 deletions
+2 -92
View File
@@ -41,7 +41,6 @@ import { PlanCard } from "@/components/plan-card";
import type { StreamItem } from "@/types/stream"; import type { StreamItem } from "@/types/stream";
import type { PendingPermission } from "@/types/shared"; import type { PendingPermission } from "@/types/shared";
import type { import type {
AgentPlanAction,
AgentPermissionAction, AgentPermissionAction,
AgentPermissionResponse, AgentPermissionResponse,
} from "@getpaseo/protocol/agent-types"; } from "@getpaseo/protocol/agent-types";
@@ -114,86 +113,6 @@ function renderPendingPermissionsNode(input: {
); );
} }
function PlanTimelineCard({
item,
agentId,
client,
}: {
item: Extract<StreamItem, { kind: "plan" }>;
agentId: string;
client: DaemonClient | null;
}) {
const [respondingActionId, setRespondingActionId] = useState<string | null>(null);
const respondToPlan = useMutation({
mutationFn: async (action: AgentPlanAction) => {
if (!client) {
throw new Error("No daemon connection");
}
setRespondingActionId(action.id);
const result = await client.respondToPlan(agentId, item.planId, { actionId: action.id });
if (!result.ok) {
throw new Error(result.error ?? "Failed to respond to plan");
}
},
onSettled: () => setRespondingActionId(null),
});
const actions = item.actions ?? [];
return (
<View>
<PlanCard title="Plan" text={item.text} testID="timeline-plan-card" />
{actions.length > 0 ? (
<View style={permissionStyles.optionsContainer}>
{actions.map((action) => (
<PlanActionButton
key={action.id}
action={action}
respondingActionId={respondingActionId}
isResponding={respondToPlan.isPending}
onRespond={respondToPlan.mutate}
/>
))}
</View>
) : null}
</View>
);
}
function PlanActionButton({
action,
respondingActionId,
isResponding,
onRespond,
}: {
action: AgentPlanAction;
respondingActionId: string | null;
isResponding: boolean;
onRespond: (action: AgentPlanAction) => void;
}) {
const Icon = action.variant === "danger" ? ThemedXIcon : ThemedCheckIcon;
const permissionAction = useMemo<AgentPermissionAction>(
() => ({
...action,
behavior: action.variant === "danger" ? "deny" : "allow",
}),
[action],
);
const handlePress = useCallback(() => onRespond(action), [action, onRespond]);
return (
<PermissionActionButton
action={permissionAction}
isRespondingAction={respondingActionId === action.id}
isResponding={isResponding}
isPrimary={action.variant === "primary"}
Icon={Icon}
testID={`plan-action-${action.id}`}
onPress={handlePress}
/>
);
}
function renderStreamItemWithTurnFooter(input: { function renderStreamItemWithTurnFooter(input: {
content: ReactNode; content: ReactNode;
layoutItem: StreamLayoutItem; layoutItem: StreamLayoutItem;
@@ -504,6 +423,7 @@ const AgentStreamViewComponent = forwardRef<AgentStreamViewHandle, AgentStreamVi
client={client} client={client}
isFirstInGroup={layoutItem.isFirstInUserGroup} isFirstInGroup={layoutItem.isFirstInUserGroup}
isLastInGroup={layoutItem.isLastInUserGroup} isLastInGroup={layoutItem.isLastInUserGroup}
deliveryHint={item.deliveryHint}
/> />
); );
}, },
@@ -630,9 +550,6 @@ const AgentStreamViewComponent = forwardRef<AgentStreamViewHandle, AgentStreamVi
case "todo_list": case "todo_list":
return <TodoListCard items={item.items} />; return <TodoListCard items={item.items} />;
case "plan":
return <PlanTimelineCard item={item} agentId={agentId} client={client} />;
case "compaction": case "compaction":
return ( return (
<CompactionMarker <CompactionMarker
@@ -646,14 +563,7 @@ const AgentStreamViewComponent = forwardRef<AgentStreamViewHandle, AgentStreamVi
return null; return null;
} }
}, },
[ [renderUserMessageItem, renderAssistantMessageItem, renderThoughtItem, renderToolCallItem],
agentId,
client,
renderUserMessageItem,
renderAssistantMessageItem,
renderThoughtItem,
renderToolCallItem,
],
); );
const bottomTurnFooterHost = streamLayout.auxiliaryTurnFooter; const bottomTurnFooterHost = streamLayout.auxiliaryTurnFooter;
+21 -1
View File
@@ -63,7 +63,11 @@ import Animated, {
import Svg, { Defs, LinearGradient as SvgLinearGradient, Rect, Stop } from "react-native-svg"; import Svg, { Defs, LinearGradient as SvgLinearGradient, Rect, Stop } from "react-native-svg";
import { createMarkdownStyles } from "@/styles/markdown-styles"; import { createMarkdownStyles } from "@/styles/markdown-styles";
import { Fonts } from "@/constants/theme"; 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 { AgentAttachment } from "@getpaseo/protocol/messages";
import type { ToolCallDetail } from "@getpaseo/protocol/agent-types"; import type { ToolCallDetail } from "@getpaseo/protocol/agent-types";
import { buildToolCallPresentation } from "@/tool-calls/presentation"; import { buildToolCallPresentation } from "@/tool-calls/presentation";
@@ -124,6 +128,7 @@ interface UserMessageProps {
isFirstInGroup?: boolean; isFirstInGroup?: boolean;
isLastInGroup?: boolean; isLastInGroup?: boolean;
disableOuterSpacing?: boolean; disableOuterSpacing?: boolean;
deliveryHint?: UserMessageDeliveryHint;
} }
const MessageOuterSpacingContext = createContext(false); const MessageOuterSpacingContext = createContext(false);
@@ -434,6 +439,12 @@ const userMessageStylesheet = StyleSheet.create((theme) => ({
color: theme.colors.foregroundMuted, color: theme.colors.foregroundMuted,
fontSize: STREAM_METADATA_FONT_SIZE, 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 }) { function UserMessageAttachmentThumbnail({ image }: { image: UserMessageImageAttachment }) {
@@ -475,6 +486,7 @@ export const UserMessage = memo(function UserMessage({
isFirstInGroup = true, isFirstInGroup = true,
isLastInGroup = true, isLastInGroup = true,
disableOuterSpacing, disableOuterSpacing,
deliveryHint,
}: UserMessageProps) { }: UserMessageProps) {
const isCompact = useIsCompactFormFactor(); const isCompact = useIsCompactFormFactor();
const [isHovered, setIsHovered] = useState(false); const [isHovered, setIsHovered] = useState(false);
@@ -589,6 +601,14 @@ export const UserMessage = memo(function UserMessage({
/> />
</View> </View>
) : null} ) : null}
{deliveryHint === "steering" ? (
<Text
accessibilityLabel="Steering conversation"
style={userMessageStylesheet.deliveryHintLabel}
>
Steering conversation
</Text>
) : null}
</View> </View>
</View> </View>
); );
+38
View File
@@ -417,6 +417,44 @@ describe("dispatchComposerAgentMessage", () => {
expect(client.calls[0]?.options.images).toEqual([]); 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 () => { it("serializes browser_element workspace attachments as text attachments at the wire boundary", async () => {
const client = createFakeSendClient(); const client = createFakeSendClient();
const stream = createFakeStream(); const stream = createFakeStream();
+3
View File
@@ -14,6 +14,7 @@ import {
buildOptimisticUserMessage, buildOptimisticUserMessage,
generateMessageId, generateMessageId,
type StreamItem, type StreamItem,
type UserMessageDeliveryHint,
type UserMessageItem, type UserMessageItem,
} from "@/types/stream"; } from "@/types/stream";
import type { PickedImageAttachmentInput } from "@/hooks/image-attachment-picker"; import type { PickedImageAttachmentInput } from "@/hooks/image-attachment-picker";
@@ -130,6 +131,7 @@ export interface DispatchComposerAgentMessageInput {
images: AttachmentMetadata[], images: AttachmentMetadata[],
) => Promise<Array<{ data: string; mimeType: string }> | undefined>; ) => Promise<Array<{ data: string; mimeType: string }> | undefined>;
stream: AgentStreamWriter; stream: AgentStreamWriter;
deliveryHint?: UserMessageDeliveryHint;
} }
export async function dispatchComposerAgentMessage( export async function dispatchComposerAgentMessage(
@@ -143,6 +145,7 @@ export async function dispatchComposerAgentMessage(
timestamp: new Date(), timestamp: new Date(),
images: wirePayload.images, images: wirePayload.images,
attachments: wirePayload.attachments, attachments: wirePayload.attachments,
deliveryHint: input.deliveryHint,
}); });
appendUserMessageToStream(input.agentId, userMessage, input.stream); appendUserMessageToStream(input.agentId, userMessage, input.stream);
const imagesData = await input.encodeImages(wirePayload.images); const imagesData = await input.encodeImages(wirePayload.images);
+4
View File
@@ -1070,6 +1070,9 @@ export function Composer({
setHead: (updater) => setAgentStreamHead(serverId, updater), setHead: (updater) => setAgentStreamHead(serverId, updater),
setTail: (updater) => setAgentStreamTail(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({ await dispatchComposerAgentMessage({
client, client,
agentId: targetAgentId, agentId: targetAgentId,
@@ -1077,6 +1080,7 @@ export function Composer({
attachments: sendAttachments, attachments: sendAttachments,
encodeImages, encodeImages,
stream, stream,
...(isSteering ? { deliveryHint: "steering" as const } : {}),
}); });
onAttentionPromptSend?.(); onAttentionPromptSend?.();
}; };
@@ -1700,6 +1700,9 @@ function SessionProviderInternal({ children, serverId, client }: SessionProvider
attachments?: AgentAttachment[], attachments?: AgentAttachment[],
) => { ) => {
const messageId = generateMessageId(); const messageId = generateMessageId();
const targetAgent = useSessionStore.getState().sessions[serverId]?.agents?.get(agentId);
const isSteering =
targetAgent?.status === "running" && targetAgent.capabilities.supportsSteering === true;
const userMessage: StreamItem = { const userMessage: StreamItem = {
kind: "user_message", kind: "user_message",
id: messageId, id: messageId,
@@ -1708,6 +1711,7 @@ function SessionProviderInternal({ children, serverId, client }: SessionProvider
optimistic: true, optimistic: true,
...(images && images.length > 0 ? { images } : {}), ...(images && images.length > 0 ? { images } : {}),
...(attachments && attachments.length > 0 ? { attachments } : {}), ...(attachments && attachments.length > 0 ? { attachments } : {}),
...(isSteering ? { deliveryHint: "steering" as const } : {}),
}; };
// Append to head if streaming (keeps the user message with the current // Append to head if streaming (keeps the user message with the current
+46 -33
View File
@@ -99,19 +99,6 @@ function todoTimeline(items: { text: string; completed: boolean }[]): AgentStrea
}; };
} }
function planTimeline(): AgentStreamEventPayload {
return {
type: "timeline",
provider: "codex",
item: {
type: "plan",
planId: "plan-1",
text: "# Plan\n\n- Implement it",
actions: [{ id: "implement", label: "Implement", variant: "primary" }],
},
};
}
function findToolByCallId(state: StreamItem[], callId: string): AgentToolCallItem | undefined { function findToolByCallId(state: StreamItem[], callId: string): AgentToolCallItem | undefined {
return state.find( return state.find(
(item): item is AgentToolCallItem => (item): item is AgentToolCallItem =>
@@ -706,26 +693,6 @@ describe("stream reducer canonical tool calls", () => {
assert.strictEqual(todos.items[1]?.completed, true); assert.strictEqual(todos.items[1]?.completed, true);
}); });
it("converts plan timeline updates to plan items", () => {
const state = hydrateStreamState([
{
event: planTimeline(),
timestamp: new Date("2025-01-01T10:55:00Z"),
},
]);
const plan = state.find(
(item): item is Extract<StreamItem, { kind: "plan" }> => item.kind === "plan",
);
assert.ok(plan);
assert.strictEqual(plan.planId, "plan-1");
assert.strictEqual(plan.text, "# Plan\n\n- Implement it");
assert.deepStrictEqual(plan.actions, [
{ id: "implement", label: "Implement", variant: "primary" },
]);
});
it("renders Claude TodoWrite as todo_list and suppresses tool call badge", () => { it("renders Claude TodoWrite as todo_list and suppresses tool call badge", () => {
const state = hydrateStreamState([ const state = hydrateStreamState([
{ {
@@ -798,6 +765,52 @@ describe("stream reducer canonical tool calls", () => {
assert.strictEqual(message.timestamp.getTime(), optimisticTimestamp.getTime()); 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", () => { it("keeps canonical assistant/user/assistant order during replay", () => {
const state: StreamItem[] = [ const state: StreamItem[] = [
{ {
+13 -46
View File
@@ -1,8 +1,4 @@
import type { import type { AgentProvider, ToolCallDetail } from "@getpaseo/protocol/agent-types";
AgentPlanAction,
AgentProvider,
ToolCallDetail,
} from "@getpaseo/protocol/agent-types";
import type { AgentAttachment, AgentStreamEventPayload } from "@getpaseo/protocol/messages"; import type { AgentAttachment, AgentStreamEventPayload } from "@getpaseo/protocol/messages";
import type { AttachmentMetadata } from "@/attachments/types"; import type { AttachmentMetadata } from "@/attachments/types";
import { extractTaskEntriesFromToolCall } from "../utils/tool-call-parsers"; import { extractTaskEntriesFromToolCall } from "../utils/tool-call-parsers";
@@ -52,13 +48,14 @@ export type StreamItem =
| AssistantMessageItem | AssistantMessageItem
| ThoughtItem | ThoughtItem
| ToolCallItem | ToolCallItem
| PlanItem
| TodoListItem | TodoListItem
| ActivityLogItem | ActivityLogItem
| CompactionItem; | CompactionItem;
export type UserMessageImageAttachment = AttachmentMetadata; export type UserMessageImageAttachment = AttachmentMetadata;
export type UserMessageDeliveryHint = "steering";
export interface UserMessageItem { export interface UserMessageItem {
kind: "user_message"; kind: "user_message";
id: string; id: string;
@@ -67,6 +64,13 @@ export interface UserMessageItem {
optimistic?: true; optimistic?: true;
images?: UserMessageImageAttachment[]; images?: UserMessageImageAttachment[];
attachments?: AgentAttachment[]; 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 { export interface OptimisticUserMessageInput {
@@ -75,6 +79,7 @@ export interface OptimisticUserMessageInput {
timestamp: Date; timestamp: Date;
images?: UserMessageImageAttachment[]; images?: UserMessageImageAttachment[];
attachments?: AgentAttachment[]; attachments?: AgentAttachment[];
deliveryHint?: UserMessageDeliveryHint;
} }
export type OptimisticUserMessagePlacement = "tail" | "active-head"; export type OptimisticUserMessagePlacement = "tail" | "active-head";
@@ -173,16 +178,6 @@ export interface TodoListItem {
items: TodoEntry[]; items: TodoEntry[];
} }
export interface PlanItem {
kind: "plan";
id: string;
timestamp: Date;
provider: AgentProvider;
planId: string;
text: string;
actions?: AgentPlanAction[];
}
export type StreamUpdateSource = "live" | "canonical"; export type StreamUpdateSource = "live" | "canonical";
function isRecord(value: unknown): value is Record<string, unknown> { function isRecord(value: unknown): value is Record<string, unknown> {
@@ -228,6 +223,7 @@ function buildUserMessageItem(input: {
...(input.optimistic.attachments && input.optimistic.attachments.length > 0 ...(input.optimistic.attachments && input.optimistic.attachments.length > 0
? { attachments: input.optimistic.attachments } ? { attachments: input.optimistic.attachments }
: {}), : {}),
...(input.optimistic.deliveryHint ? { deliveryHint: input.optimistic.deliveryHint } : {}),
}; };
} }
@@ -250,6 +246,7 @@ export function buildOptimisticUserMessage(input: OptimisticUserMessageInput): U
...(input.attachments && input.attachments.length > 0 ...(input.attachments && input.attachments.length > 0
? { attachments: input.attachments } ? { attachments: input.attachments }
: {}), : {}),
...(input.deliveryHint ? { deliveryHint: input.deliveryHint } : {}),
}; };
} }
@@ -668,34 +665,6 @@ function appendTodoList(
return [...state, entry]; return [...state, entry];
} }
function appendPlan(
state: StreamItem[],
provider: AgentProvider,
plan: { planId: string; text: string; actions?: AgentPlanAction[] },
timestamp: Date,
): StreamItem[] {
const existingIndex = state.findIndex(
(item) => item.kind === "plan" && item.provider === provider && item.planId === plan.planId,
);
const entry: PlanItem = {
kind: "plan",
id: `plan_${plan.planId}`,
timestamp,
provider,
planId: plan.planId,
text: plan.text,
actions: plan.actions,
};
if (existingIndex >= 0) {
const next = [...state];
next[existingIndex] = entry;
return next;
}
return [...state, entry];
}
function reduceTimelineToolCall( function reduceTimelineToolCall(
state: StreamItem[], state: StreamItem[],
event: Extract<AgentStreamEventPayload, { type: "timeline" }>, event: Extract<AgentStreamEventPayload, { type: "timeline" }>,
@@ -817,8 +786,6 @@ function reduceTimelineEvent(
})); }));
return finalizeActiveThoughts(appendTodoList(state, event.provider, items, timestamp)); return finalizeActiveThoughts(appendTodoList(state, event.provider, items, timestamp));
} }
case "plan":
return finalizeActiveThoughts(appendPlan(state, event.provider, item, timestamp));
case "error": { case "error": {
const activity: ActivityLogItem = { const activity: ActivityLogItem = {
kind: "activity_log", kind: "activity_log",
@@ -255,7 +255,6 @@ test("advertises client capabilities in hello", async () => {
protocolVersion: 1, protocolVersion: 1,
capabilities: { capabilities: {
custom_mode_icons: true, custom_mode_icons: true,
first_class_plans: true,
reasoning_merge_enum: true, reasoning_merge_enum: true,
}, },
}); });
-38
View File
@@ -84,7 +84,6 @@ import type {
import type { import type {
AgentPermissionRequest, AgentPermissionRequest,
AgentPermissionResponse, AgentPermissionResponse,
AgentPlanResponse,
AgentPersistenceHandle, AgentPersistenceHandle,
AgentProvider, AgentProvider,
AgentSessionConfig, AgentSessionConfig,
@@ -364,10 +363,6 @@ type DictationFinishAcceptedPayload = Extract<
{ type: "dictation_stream_finish_accepted" } { type: "dictation_stream_finish_accepted" }
>["payload"]; >["payload"];
type AgentPermissionResolvedPayload = AgentPermissionResolvedMessage["payload"]; type AgentPermissionResolvedPayload = AgentPermissionResolvedMessage["payload"];
type AgentPlanRespondPayload = Extract<
SessionOutboundMessage,
{ type: "agent.plan.respond.response" }
>["payload"];
type ListTerminalsPayload = ListTerminalsResponse["payload"]; type ListTerminalsPayload = ListTerminalsResponse["payload"];
type CreateTerminalPayload = CreateTerminalResponse["payload"]; type CreateTerminalPayload = CreateTerminalResponse["payload"];
export type RenameTerminalResult = z.infer<typeof RenameTerminalResponseSchema>["payload"]; export type RenameTerminalResult = z.infer<typeof RenameTerminalResponseSchema>["payload"];
@@ -3618,38 +3613,6 @@ export class DaemonClient {
}); });
} }
async respondToPlan(
agentId: string,
planId: string,
response: AgentPlanResponse,
requestId = `plan-response-${Date.now()}-${Math.random().toString(36).slice(2)}`,
timeout = 15000,
): Promise<AgentPlanRespondPayload> {
const message = SessionInboundMessageSchema.parse({
type: "agent.plan.respond.request",
agentId,
planId,
actionId: response.actionId,
...(response.feedback !== undefined ? { feedback: response.feedback } : {}),
requestId,
});
return this.sendRequest({
requestId,
message,
timeout,
options: { skipQueue: true },
select: (msg) => {
if (msg.type !== "agent.plan.respond.response") {
return null;
}
if (msg.payload.requestId !== requestId) {
return null;
}
return msg.payload;
},
});
}
// ============================================================================ // ============================================================================
// Waiting / Streaming Helpers // Waiting / Streaming Helpers
// ============================================================================ // ============================================================================
@@ -4315,7 +4278,6 @@ export class DaemonClient {
protocolVersion: 1, protocolVersion: 1,
capabilities: { capabilities: {
[CLIENT_CAPS.customModeIcons]: true, [CLIENT_CAPS.customModeIcons]: true,
[CLIENT_CAPS.firstClassPlans]: true,
[CLIENT_CAPS.reasoningMergeEnum]: true, [CLIENT_CAPS.reasoningMergeEnum]: true,
}, },
...(this.config.appVersion ? { appVersion: this.config.appVersion } : {}), ...(this.config.appVersion ? { appVersion: this.config.appVersion } : {}),
+1
View File
@@ -152,6 +152,7 @@ function createAgent(input: Partial<PaseoAgent> = {}): PaseoAgent {
supportsRewindBoth: false, supportsRewindBoth: false,
supportsRewindConversation: false, supportsRewindConversation: false,
supportsRewindFiles: false, supportsRewindFiles: false,
supportsSteering: false,
supportsToolInvocations: true, supportsToolInvocations: true,
}, },
currentModeId: null, currentModeId: null,
+1 -20
View File
@@ -125,6 +125,7 @@ export interface AgentCapabilityFlags {
supportsRewindConversation?: boolean; supportsRewindConversation?: boolean;
supportsRewindFiles?: boolean; supportsRewindFiles?: boolean;
supportsRewindBoth?: boolean; supportsRewindBoth?: boolean;
supportsSteering?: boolean;
} }
export interface AgentPersistenceHandle { export interface AgentPersistenceHandle {
@@ -313,32 +314,12 @@ export interface CompactionTimelineItem {
preTokens?: number; preTokens?: number;
} }
export interface AgentPlanAction {
id: string;
label: string;
variant?: "primary" | "secondary" | "danger";
}
export interface PlanTimelineItem {
[key: string]: unknown;
type: "plan";
planId: string;
text: string;
actions?: AgentPlanAction[];
}
export interface AgentPlanResponse {
actionId: string;
feedback?: string;
}
export type AgentTimelineItem = export type AgentTimelineItem =
| { type: "user_message"; text: string; messageId?: string } | { type: "user_message"; text: string; messageId?: string }
| { type: "assistant_message"; text: string; messageId?: string } | { type: "assistant_message"; text: string; messageId?: string }
| { type: "reasoning"; text: string } | { type: "reasoning"; text: string }
| ToolCallTimelineItem | ToolCallTimelineItem
| { type: "todo"; items: { text: string; completed: boolean }[] } | { type: "todo"; items: { text: string; completed: boolean }[] }
| PlanTimelineItem
| { type: "error"; message: string } | { type: "error"; message: string }
| CompactionTimelineItem; | CompactionTimelineItem;
@@ -1,5 +1,4 @@
export const CLIENT_CAPS = { export const CLIENT_CAPS = {
firstClassPlans: "first_class_plans",
reasoningMergeEnum: "reasoning_merge_enum", reasoningMergeEnum: "reasoning_merge_enum",
// COMPAT(customModeIcons): added in v0.1.84. Old clients pin AgentModeIcon to // COMPAT(customModeIcons): added in v0.1.84. Old clients pin AgentModeIcon to
// a closed enum and crash rendering unknown values; daemon downgrades icons // a closed enum and crash rendering unknown values; daemon downgrades icons
@@ -179,13 +179,11 @@ describe("checkout PR schemas", () => {
features: { features: {
providersSnapshot: true, providersSnapshot: true,
checkoutGithubSetAutoMerge: true, checkoutGithubSetAutoMerge: true,
firstClassPlans: true,
}, },
}).features, }).features,
).toEqual({ ).toEqual({
providersSnapshot: true, providersSnapshot: true,
checkoutGithubSetAutoMerge: true, checkoutGithubSetAutoMerge: true,
firstClassPlans: true,
}); });
}); });
}); });
@@ -1,5 +1,6 @@
import { describe, expect, test } from "vitest"; import { describe, expect, test } from "vitest";
import { import {
AgentSnapshotPayloadSchema,
GetProvidersSnapshotResponseMessageSchema, GetProvidersSnapshotResponseMessageSchema,
ProviderSnapshotEntrySchema, ProviderSnapshotEntrySchema,
ProvidersSnapshotUpdateMessageSchema, ProvidersSnapshotUpdateMessageSchema,
@@ -16,6 +17,34 @@ describe("provider snapshot message schemas", () => {
expect(parsed.enabled).toBe(true); 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", () => { test("preserves disabled provider snapshot entries", () => {
const parsed = ProviderSnapshotEntrySchema.parse({ const parsed = ProviderSnapshotEntrySchema.parse({
provider: "claude", provider: "claude",
+2 -38
View File
@@ -153,7 +153,6 @@ import type {
AgentPermissionRequest, AgentPermissionRequest,
AgentPermissionResponse, AgentPermissionResponse,
AgentPersistenceHandle, AgentPersistenceHandle,
AgentPlanAction,
ProviderStatus, ProviderStatus,
AgentRuntimeInfo, AgentRuntimeInfo,
AgentTimelineItem, AgentTimelineItem,
@@ -252,6 +251,8 @@ const AgentCapabilityFlagsSchema: z.ZodType<AgentCapabilityFlags> = z.object({
supportsRewindFiles: z.boolean().optional().default(false), supportsRewindFiles: z.boolean().optional().default(false),
// COMPAT(rewind): added in v0.1.X, drop when floor >= v0.1.X. // COMPAT(rewind): added in v0.1.X, drop when floor >= v0.1.X.
supportsRewindBoth: z.boolean().optional().default(false), 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({ const AgentUsageSchema: z.ZodType<AgentUsage> = z.object({
@@ -335,12 +336,6 @@ export const AgentPermissionResponseSchema: z.ZodType<AgentPermissionResponse> =
}), }),
]); ]);
const AgentPlanActionSchema: z.ZodType<AgentPlanAction> = z.object({
id: z.string(),
label: z.string(),
variant: z.enum(["primary", "secondary", "danger"]).optional(),
});
export const AgentPermissionRequestPayloadSchema: z.ZodType< export const AgentPermissionRequestPayloadSchema: z.ZodType<
AgentPermissionRequest, AgentPermissionRequest,
z.ZodTypeDef, z.ZodTypeDef,
@@ -556,12 +551,6 @@ export const AgentTimelineItemPayloadSchema: z.ZodType<AgentTimelineItem, z.ZodT
}), }),
), ),
}), }),
z.object({
type: z.literal("plan"),
planId: z.string(),
text: z.string(),
actions: z.array(AgentPlanActionSchema).optional(),
}),
z.object({ z.object({
type: z.literal("error"), type: z.literal("error"),
message: z.string(), message: z.string(),
@@ -1337,15 +1326,6 @@ export const AgentPermissionResponseMessageSchema = z.object({
response: AgentPermissionResponseSchema, response: AgentPermissionResponseSchema,
}); });
export const AgentPlanRespondRequestMessageSchema = z.object({
type: z.literal("agent.plan.respond.request"),
agentId: z.string(),
planId: z.string(),
actionId: z.string(),
feedback: z.string().optional(),
requestId: z.string(),
});
const CheckoutErrorCodeSchema = z.enum([ const CheckoutErrorCodeSchema = z.enum([
"NOT_GIT_REPO", "NOT_GIT_REPO",
"NOT_ALLOWED", "NOT_ALLOWED",
@@ -1923,7 +1903,6 @@ export const SessionInboundMessageSchema = z.discriminatedUnion("type", [
SetAgentFeatureRequestMessageSchema, SetAgentFeatureRequestMessageSchema,
AgentRewindRequestMessageSchema, AgentRewindRequestMessageSchema,
AgentPermissionResponseMessageSchema, AgentPermissionResponseMessageSchema,
AgentPlanRespondRequestMessageSchema,
CheckoutStatusRequestSchema, CheckoutStatusRequestSchema,
SubscribeCheckoutDiffRequestSchema, SubscribeCheckoutDiffRequestSchema,
UnsubscribeCheckoutDiffRequestSchema, UnsubscribeCheckoutDiffRequestSchema,
@@ -2161,8 +2140,6 @@ export const ServerInfoStatusPayloadSchema = z
.object({ .object({
providersSnapshot: z.boolean().optional(), providersSnapshot: z.boolean().optional(),
checkoutGithubSetAutoMerge: z.boolean().optional(), checkoutGithubSetAutoMerge: z.boolean().optional(),
// COMPAT(firstClassPlans): added in v0.1.82, remove gate after 2026-11-28.
firstClassPlans: z.boolean().optional(),
// COMPAT(daemonStatusRpc): added in v0.1.76, remove gate after 2026-11-18. // COMPAT(daemonStatusRpc): added in v0.1.76, remove gate after 2026-11-18.
daemonStatusRpc: z.boolean().optional(), daemonStatusRpc: z.boolean().optional(),
// COMPAT(terminalRestoreModes): added in v0.1.81, remove gate after 2026-11-23. // COMPAT(terminalRestoreModes): added in v0.1.81, remove gate after 2026-11-23.
@@ -2729,17 +2706,6 @@ export const SendAgentMessageResponseMessageSchema = z.object({
}), }),
}); });
export const AgentPlanRespondResponseMessageSchema = z.object({
type: z.literal("agent.plan.respond.response"),
payload: z.object({
requestId: z.string(),
agentId: z.string(),
planId: z.string(),
ok: z.boolean(),
error: z.string().nullable().optional(),
}),
});
export const WaitForFinishResponseMessageSchema = z.object({ export const WaitForFinishResponseMessageSchema = z.object({
type: z.literal("wait_for_finish_response"), type: z.literal("wait_for_finish_response"),
payload: z.object({ payload: z.object({
@@ -3729,7 +3695,6 @@ export const SessionOutboundMessageSchema = z.discriminatedUnion("type", [
CancelAgentResponseMessageSchema, CancelAgentResponseMessageSchema,
ClearAgentAttentionResponseMessageSchema, ClearAgentAttentionResponseMessageSchema,
SendAgentMessageResponseMessageSchema, SendAgentMessageResponseMessageSchema,
AgentPlanRespondResponseMessageSchema,
SetVoiceModeResponseMessageSchema, SetVoiceModeResponseMessageSchema,
DaemonGetStatusResponseSchema, DaemonGetStatusResponseSchema,
DaemonGetPairingOfferResponseSchema, DaemonGetPairingOfferResponseSchema,
@@ -4143,7 +4108,6 @@ export const WSHelloMessageSchema = z.object({
.object({ .object({
voice: z.boolean().optional(), voice: z.boolean().optional(),
pushNotifications: z.boolean().optional(), pushNotifications: z.boolean().optional(),
[CLIENT_CAPS.firstClassPlans]: z.boolean().optional(),
[CLIENT_CAPS.reasoningMergeEnum]: z.boolean().optional(), [CLIENT_CAPS.reasoningMergeEnum]: z.boolean().optional(),
[CLIENT_CAPS.customModeIcons]: z.boolean().optional(), [CLIENT_CAPS.customModeIcons]: z.boolean().optional(),
}) })
@@ -18,6 +18,7 @@ import type {
AgentLaunchContext, AgentLaunchContext,
AgentProvider, AgentProvider,
AgentPersistenceHandle, AgentPersistenceHandle,
AgentPromptInput,
AgentRunResult, AgentRunResult,
AgentSession, AgentSession,
AgentSessionConfig, AgentSessionConfig,
@@ -3010,6 +3011,314 @@ test("replaceAgentRun does not emit idle or resolve waiters between interrupted
unsubscribe(); 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 () => { 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 workdir = mkdtempSync(join(tmpdir(), "agent-manager-replace-stale-terminal-"));
const storagePath = join(workdir, "agents"); const storagePath = join(workdir, "agents");
@@ -19,8 +19,6 @@ import {
type AgentLaunchContext, type AgentLaunchContext,
type AgentSlashCommand, type AgentSlashCommand,
type AgentMode, type AgentMode,
type AgentPlanResponse,
type AgentPlanResult,
type AgentPermissionRequest, type AgentPermissionRequest,
type AgentPermissionResponse, type AgentPermissionResponse,
type AgentPermissionResult, type AgentPermissionResult,
@@ -207,6 +205,15 @@ export interface WaitForAgentStartOptions {
signal?: AbortSignal; signal?: AbortSignal;
} }
export interface StartAgentRunOptions {
replaceRunning?: boolean;
runOptions?: AgentRunOptions;
}
export type StartAgentRunResult =
| { outOfBand: true }
| { outOfBand: false; events: AsyncGenerator<AgentStreamEvent> };
type AttentionState = type AttentionState =
| { requiresAttention: false } | { requiresAttention: false }
| { | {
@@ -1484,6 +1491,55 @@ export class AgentManager {
return true; 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 ${agentId} does not support steering`);
}
await agent.session.steerTurn(prompt, options);
this.touchUpdatedAt(agent);
this.emitState(agent);
yield* [];
}
async appendTimelineItem(agentId: string, item: AgentTimelineItem): Promise<void> { async appendTimelineItem(agentId: string, item: AgentTimelineItem): Promise<void> {
const agent = this.requireAgent(agentId); const agent = this.requireAgent(agentId);
this.touchUpdatedAt(agent); this.touchUpdatedAt(agent);
@@ -1838,31 +1894,6 @@ export class AgentManager {
} }
} }
async respondToPlan(
agentId: string,
planId: string,
response: AgentPlanResponse,
): Promise<AgentPlanResult | void> {
const agent = this.requireAgent(agentId);
if (!agent.session.respondToPlan) {
throw new Error(`Agent provider '${agent.provider}' does not support plan responses`);
}
const result = await agent.session.respondToPlan(planId, response);
try {
await this.refreshSessionState(agent);
} catch {
// Ignore refresh errors - state sync after plan response is best effort.
}
this.touchUpdatedAt(agent);
await this.persistSnapshot(agent);
this.emitState(agent);
return result;
}
async cancelAgentRun(agentId: string): Promise<boolean> { async cancelAgentRun(agentId: string): Promise<boolean> {
const agent = this.requireSessionAgent(agentId); const agent = this.requireSessionAgent(agentId);
const pendingRun = this.foregroundRuns.getPendingRun(agentId); const pendingRun = this.foregroundRuns.getPendingRun(agentId);
@@ -10,6 +10,9 @@ import {
} from "./agent-prompt.js"; } from "./agent-prompt.js";
import type { AgentManagerEvent, ManagedAgent } from "./agent-manager.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", () => { test("isSystemInjectedEnvelope matches the envelope formatSystemNotificationPrompt produces", () => {
expect(isSystemInjectedEnvelope(formatSystemNotificationPrompt("child finished"))).toBe(true); expect(isSystemInjectedEnvelope(formatSystemNotificationPrompt("child finished"))).toBe(true);
expect(isSystemInjectedEnvelope("hello world")).toBe(false); expect(isSystemInjectedEnvelope("hello world")).toBe(false);
@@ -19,27 +22,24 @@ it("does not notify archived callers", async () => {
let subscriber: ((event: AgentManagerEvent) => void) | null = null; let subscriber: ((event: AgentManagerEvent) => void) | null = null;
const childAgent: ManagedAgent = Object.create(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, "lifecycle", "idle");
Reflect.set(childAgent, "config", { title: "Child Agent" }); Reflect.set(childAgent, "config", { title: "Child Agent" });
const callerAgent: ManagedAgent = Object.create(null); 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, "lifecycle", "idle");
Reflect.set(callerAgent, "config", { title: "Caller Agent" }); 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); const agentManager: AgentManager = Object.create(AgentManager.prototype);
Reflect.set( Reflect.set(
agentManager, agentManager,
"getAgent", "getAgent",
vi.fn((agentId: string) => { vi.fn((agentId: string) => {
if (agentId === "child-agent") { if (agentId === CHILD_AGENT_ID) {
return childAgent; return childAgent;
} }
if (agentId === "caller-agent") { if (agentId === CALLER_AGENT_ID) {
return callerAgent; return callerAgent;
} }
return null; return null;
@@ -55,12 +55,14 @@ it("does not notify archived callers", async () => {
}; };
}), }),
); );
Reflect.set(agentManager, "hasInFlightRun", vi.fn().mockReturnValue(false)); const startAgentRunSpy = vi.fn(() => ({
Reflect.set(agentManager, "streamAgent", streamAgentSpy); outOfBand: false,
Reflect.set(agentManager, "replaceAgentRun", replaceAgentRunSpy); events: (async function* noop() {})(),
}));
Reflect.set(agentManager, "startAgentRun", startAgentRunSpy);
const agentStorageGetSpy = vi.fn(async (agentId: string) => 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); const agentStorage: AgentStorage = Object.create(AgentStorage.prototype);
Reflect.set(agentStorage, "get", agentStorageGetSpy); Reflect.set(agentStorage, "get", agentStorageGetSpy);
@@ -68,8 +70,8 @@ it("does not notify archived callers", async () => {
setupFinishNotification({ setupFinishNotification({
agentManager, agentManager,
agentStorage, agentStorage,
childAgentId: "child-agent", childAgentId: CHILD_AGENT_ID,
callerAgentId: "caller-agent", callerAgentId: CALLER_AGENT_ID,
logger: createTestLogger(), logger: createTestLogger(),
}); });
@@ -88,9 +90,92 @@ it("does not notify archived callers", async () => {
}); });
await vi.waitFor(() => { await vi.waitFor(() => {
expect(agentStorageGetSpy).toHaveBeenCalledWith("caller-agent"); expect(agentStorageGetSpy).toHaveBeenCalledWith(CALLER_AGENT_ID);
}); });
expect(streamAgentSpy).not.toHaveBeenCalled(); expect(startAgentRunSpy).not.toHaveBeenCalled();
expect(replaceAgentRunSpy).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 { Logger } from "pino";
import type { AgentPromptInput, AgentRunOptions } from "./agent-sdk-types.js"; 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 type { AgentStorage } from "./agent-storage.js";
import { ensureAgentLoaded } from "./agent-loading.js"; import { ensureAgentLoaded } from "./agent-loading.js";
export type AgentRunController = Pick< export type AgentRunController = Pick<AgentManager, "getAgent" | "startAgentRun">;
AgentManager,
"getAgent" | "tryRunOutOfBand" | "hasInFlightRun" | "replaceAgentRun" | "streamAgent"
>;
export interface StartAgentRunOptions {
replaceRunning?: boolean;
runOptions?: AgentRunOptions;
}
export function startAgentRun( export function startAgentRun(
agentManager: AgentRunController, agentManager: AgentRunController,
@@ -35,23 +27,17 @@ export function startAgentRun(
}, },
"agent.session.start_stream.request", "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 const result = agentManager.startAgentRun(agentId, prompt, options);
// intercept lives at this layer so it covers every prompt entrypoint. if (result.outOfBand) {
if (agentManager.tryRunOutOfBand(agentId, prompt)) {
return { outOfBand: true }; return { outOfBand: true };
} }
const shouldReplace = Boolean(options?.replaceRunning && agentManager.hasInFlightRun(agentId)); const iterator = result.events;
const runOptions = options?.runOptions;
const iterator = shouldReplace
? agentManager.replaceAgentRun(agentId, prompt, runOptions)
: agentManager.streamAgent(agentId, prompt, runOptions);
logger.trace( logger.trace(
{ {
agentId, agentId,
provider: snapshot?.provider, provider: snapshot?.provider,
providerSessionId: snapshot?.persistence?.sessionId ?? undefined, providerSessionId: snapshot?.persistence?.sessionId ?? undefined,
shouldReplace,
}, },
"agent.session.start_stream.iterator_returned", "agent.session.start_stream.iterator_returned",
); );
@@ -153,6 +153,7 @@ export interface AgentCapabilityFlags {
supportsRewindConversation?: boolean; supportsRewindConversation?: boolean;
supportsRewindFiles?: boolean; supportsRewindFiles?: boolean;
supportsRewindBoth?: boolean; supportsRewindBoth?: boolean;
supportsSteering?: boolean;
} }
export interface AgentPersistenceHandle { export interface AgentPersistenceHandle {
@@ -342,27 +343,12 @@ export interface CompactionTimelineItem {
preTokens?: number; preTokens?: number;
} }
export interface AgentPlanAction {
id: string;
label: string;
variant?: "primary" | "secondary" | "danger";
}
export interface PlanTimelineItem {
[key: string]: unknown;
type: "plan";
planId: string;
text: string;
actions?: AgentPlanAction[];
}
export type AgentTimelineItem = export type AgentTimelineItem =
| { type: "user_message"; text: string; messageId?: string } | { type: "user_message"; text: string; messageId?: string }
| { type: "assistant_message"; text: string; messageId?: string } | { type: "assistant_message"; text: string; messageId?: string }
| { type: "reasoning"; text: string } | { type: "reasoning"; text: string }
| ToolCallTimelineItem | ToolCallTimelineItem
| { type: "todo"; items: { text: string; completed: boolean }[] } | { type: "todo"; items: { text: string; completed: boolean }[] }
| PlanTimelineItem
| { type: "error"; message: string } | { type: "error"; message: string }
| CompactionTimelineItem; | CompactionTimelineItem;
@@ -566,15 +552,6 @@ export interface AgentPermissionResult {
followUpPrompt?: AgentPromptInput; followUpPrompt?: AgentPromptInput;
} }
export interface AgentPlanResponse {
actionId: string;
feedback?: string;
}
export interface AgentPlanResult {
followUpPrompt?: AgentPromptInput;
}
export interface AgentSession { export interface AgentSession {
readonly provider: AgentProvider; readonly provider: AgentProvider;
readonly id: string | null; readonly id: string | null;
@@ -582,6 +559,7 @@ export interface AgentSession {
readonly features?: AgentFeature[]; readonly features?: AgentFeature[];
run(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<AgentRunResult>; run(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<AgentRunResult>;
startTurn(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<{ turnId: string }>; startTurn(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<{ turnId: string }>;
steerTurn?(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<void>;
subscribe(callback: (event: AgentStreamEvent) => void): () => void; subscribe(callback: (event: AgentStreamEvent) => void): () => void;
streamHistory(): AsyncGenerator<AgentStreamEvent>; streamHistory(): AsyncGenerator<AgentStreamEvent>;
getRuntimeInfo(): Promise<AgentRuntimeInfo>; getRuntimeInfo(): Promise<AgentRuntimeInfo>;
@@ -593,7 +571,6 @@ export interface AgentSession {
requestId: string, requestId: string,
response: AgentPermissionResponse, response: AgentPermissionResponse,
): Promise<AgentPermissionResult | void>; ): Promise<AgentPermissionResult | void>;
respondToPlan?(planId: string, response: AgentPlanResponse): Promise<AgentPlanResult | void>;
describePersistence(): AgentPersistenceHandle | null; describePersistence(): AgentPersistenceHandle | null;
interrupt(): Promise<void>; interrupt(): Promise<void>;
close(): Promise<void>; close(): Promise<void>;
@@ -8,6 +8,7 @@ import type {
AgentPermissionResponse, AgentPermissionResponse,
} from "./agent-sdk-types.js"; } from "./agent-sdk-types.js";
import type { AgentStreamEvent } from "../messages.js"; import type { AgentStreamEvent } from "../messages.js";
import type { StartAgentRunOptions, StartAgentRunResult } from "./agent-manager.js";
import { respondToAgentPermission } from "./permission-response.js"; import { respondToAgentPermission } from "./permission-response.js";
class FakePermissionAgentManager { class FakePermissionAgentManager {
@@ -61,6 +62,21 @@ class FakePermissionAgentManager {
this.replacementRuns.push({ agentId, prompt, options }); this.replacementRuns.push({ agentId, prompt, options });
return emptyAgentStream(); 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> {} async function* emptyAgentStream(): AsyncGenerator<AgentStreamEvent> {}
@@ -1,41 +0,0 @@
import { describe, expect, test } from "vitest";
import { isPlanFilePath, planItemFromToolCall } from "./plan-files.js";
describe("plan file detection", () => {
test("accepts only narrow Paseo and OpenCode plan markdown paths", () => {
expect(isPlanFilePath(".paseo/plans/feature.md")).toBe(true);
expect(isPlanFilePath("/Users/me/project/.paseo/plans/feature.markdown")).toBe(true);
expect(isPlanFilePath(".opencode/plans/refactor.md")).toBe(true);
expect(isPlanFilePath("/Users/me/.opencode/plans/refactor.markdown")).toBe(true);
expect(isPlanFilePath("PLAN.md")).toBe(false);
expect(isPlanFilePath("docs/plan.md")).toBe(false);
expect(isPlanFilePath(".paseo/notes/feature.md")).toBe(false);
expect(isPlanFilePath(".paseo/plans/feature.txt")).toBe(false);
});
test("turns successful plan writes into non-actionable plan items", async () => {
const item = await planItemFromToolCall({
cwd: "/workspace",
homeDir: "/Users/me",
item: {
type: "tool_call",
callId: "write-plan",
name: "write",
status: "completed",
error: null,
detail: {
type: "write",
filePath: ".paseo/plans/feature.md",
content: "# Plan\n\n- Implement it",
},
},
});
expect(item).toEqual({
type: "plan",
planId: "plan-file:.paseo/plans/feature.md",
text: "# Plan\n\n- Implement it",
});
});
});
@@ -1,70 +0,0 @@
import path from "node:path";
import fs from "node:fs/promises";
import type { AgentTimelineItem, ToolCallTimelineItem } from "./agent-sdk-types.js";
const PLAN_FILE_EXTENSIONS = new Set([".md", ".markdown"]);
const PLAN_DIRECTORIES = new Set(["/.paseo/plans/", "/.opencode/plans/"]);
export function isPlanFilePath(filePath: string): boolean {
const normalized = normalizePlanPath(filePath);
const ext = path.posix.extname(normalized).toLowerCase();
if (!PLAN_FILE_EXTENSIONS.has(ext)) {
return false;
}
const searchable = normalized.startsWith("/") ? normalized : `/${normalized}`;
return Array.from(PLAN_DIRECTORIES).some((dir) => searchable.includes(dir));
}
export async function planItemFromToolCall(params: {
item: ToolCallTimelineItem;
cwd: string;
homeDir: string;
}): Promise<AgentTimelineItem | null> {
const { item, cwd, homeDir } = params;
if (item.status !== "completed") {
return null;
}
const detail = item.detail;
if (detail.type !== "write" && detail.type !== "edit") {
return null;
}
if (!isPlanFilePath(detail.filePath)) {
return null;
}
const explicitContent = detail.type === "write" ? detail.content : undefined;
const text = explicitContent ?? (await readPlanFile(detail.filePath, cwd, homeDir));
if (!text?.trim()) {
return null;
}
return {
type: "plan",
planId: `plan-file:${normalizePlanPath(detail.filePath)}`,
text: text.trim(),
};
}
function normalizePlanPath(filePath: string): string {
return path.posix.normalize(filePath.replace(/\\/g, "/"));
}
async function readPlanFile(
filePath: string,
cwd: string,
homeDir: string,
): Promise<string | null> {
const candidates = path.isAbsolute(filePath)
? [filePath]
: [path.resolve(cwd, filePath), path.resolve(homeDir, filePath)];
for (const candidate of candidates) {
try {
return await fs.readFile(candidate, "utf8");
} catch {
// Try the next candidate.
}
}
return null;
}
@@ -351,6 +351,7 @@ export function wrapSessionProvider(provider: AgentProvider, inner: AgentSession
}, },
run: (prompt, options) => inner.run(prompt, options), run: (prompt, options) => inner.run(prompt, options),
startTurn: (prompt, options) => inner.startTurn(prompt, options), startTurn: (prompt, options) => inner.startTurn(prompt, options),
steerTurn: inner.steerTurn?.bind(inner),
subscribe: (callback) => inner.subscribe((event) => callback(mapStreamEvent(provider, event))), subscribe: (callback) => inner.subscribe((event) => callback(mapStreamEvent(provider, event))),
async *streamHistory() { async *streamHistory() {
for await (const event of inner.streamHistory()) { for await (const event of inner.streamHistory()) {
@@ -363,7 +364,6 @@ export function wrapSessionProvider(provider: AgentProvider, inner: AgentSession
setMode: (modeId) => inner.setMode(modeId), setMode: (modeId) => inner.setMode(modeId),
getPendingPermissions: () => inner.getPendingPermissions(), getPendingPermissions: () => inner.getPendingPermissions(),
respondToPermission: (requestId, response) => inner.respondToPermission(requestId, response), respondToPermission: (requestId, response) => inner.respondToPermission(requestId, response),
respondToPlan: inner.respondToPlan?.bind(inner),
describePersistence: () => mapPersistenceHandle(provider, inner.describePersistence()), describePersistence: () => mapPersistenceHandle(provider, inner.describePersistence()),
interrupt: () => inner.interrupt(), interrupt: () => inner.interrupt(),
close: () => inner.close(), close: () => inner.close(),
@@ -997,7 +997,7 @@ test("preserves bypass capability across query restarts triggered by thinking ch
} }
}); });
test("plan item exposes a resume-bypass action and can return to bypassPermissions", async () => { test("plan approval exposes a resume-bypass action and can return to bypassPermissions", async () => {
const queryMock = createBaseQueryMock(vi.fn(async () => ({ done: true, value: undefined }))); const queryMock = createBaseQueryMock(vi.fn(async () => ({ done: true, value: undefined })));
sdkQueryFactory.mockImplementation(() => queryMock); sdkQueryFactory.mockImplementation(() => queryMock);
@@ -1023,36 +1023,44 @@ test("plan item exposes a resume-bypass action and can return to bypassPermissio
{}, {},
); );
const planEvent = events.find( const requestEvent = events.find(
(event): event is Extract<AgentStreamEvent, { type: "timeline" }> => (event): event is Extract<AgentStreamEvent, { type: "permission_requested" }> =>
event.type === "timeline" && event.item.type === "plan", event.type === "permission_requested" && event.request.kind === "plan",
); );
expect(planEvent).toBeDefined(); expect(requestEvent).toBeDefined();
expect(planEvent?.item.type === "plan" ? planEvent.item.actions : undefined).toEqual([ expect(requestEvent?.request.actions).toEqual([
{ {
id: "reject", id: "reject",
label: "Reject", label: "Reject",
behavior: "deny",
variant: "danger", variant: "danger",
intent: "dismiss",
}, },
{ {
id: "implement", id: "implement",
label: "Implement", label: "Implement",
behavior: "allow",
variant: "primary", variant: "primary",
intent: "implement",
}, },
{ {
id: "implement_resume", id: "implement_resume",
label: "Implement with Bypass", label: "Implement with Bypass",
behavior: "allow",
variant: "secondary", variant: "secondary",
intent: "implement_resume",
}, },
]); ]);
if (!planEvent || planEvent.item.type !== "plan") { if (!requestEvent) {
throw new Error("Expected plan item"); throw new Error("Expected plan permission request");
} }
expect(session.getPendingPermissions()).toEqual([]);
await session.respondToPlan?.(planEvent.item.planId, { actionId: "implement_resume" }); await session.respondToPermission(requestEvent.request.id, {
behavior: "allow",
selectedActionId: "implement_resume",
});
await expect(pendingResolution).resolves.toMatchObject({ await expect(pendingResolution).resolves.toMatchObject({
behavior: "allow", behavior: "allow",
@@ -2,7 +2,7 @@ import * as fs from "node:fs/promises";
import * as os from "node:os"; import * as os from "node:os";
import * as path from "node:path"; import * as path from "node:path";
import { afterEach, describe, expect, test, vi } from "vitest"; 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 { createTestLogger } from "../../../../test-utils/test-logger.js";
import * as executableUtils from "../../../../utils/executable.js"; import * as executableUtils from "../../../../utils/executable.js";
@@ -749,7 +749,10 @@ describe("ClaudeAgentSession context window usage", () => {
return session as unknown as TestClaudeSession; 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> }) => { return vi.fn(({ prompt }: { prompt: AsyncIterable<unknown> }) => {
const queuedMessages: Array<Record<string, unknown>> = []; const queuedMessages: Array<Record<string, unknown>> = [];
const waiters: Array<() => void> = []; const waiters: Array<() => void> = [];
@@ -768,6 +771,7 @@ describe("ClaudeAgentSession context window usage", () => {
void (async () => { void (async () => {
for await (const _prompt of prompt) { for await (const _prompt of prompt) {
onPrompt?.(_prompt);
const turnMessages = turns[turnIndex] ?? []; const turnMessages = turns[turnIndex] ?? [];
turnIndex += 1; turnIndex += 1;
for (const message of turnMessages) { 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 () => { test("convertUsage includes contextWindowMaxTokens and derives used tokens from result usage as initial fallback", async () => {
const session = await createSessionForTest(); const session = await createSessionForTest();
@@ -57,8 +57,6 @@ import {
type AgentMetadata, type AgentMetadata,
type AgentMode, type AgentMode,
type AgentModelDefinition, type AgentModelDefinition,
type AgentPlanAction,
type AgentPlanResponse,
type AgentPermissionRequest, type AgentPermissionRequest,
type AgentPermissionRequestKind, type AgentPermissionRequestKind,
type AgentPermissionResponse, type AgentPermissionResponse,
@@ -218,6 +216,7 @@ const CLAUDE_CAPABILITIES: AgentCapabilityFlags = {
supportsRewindConversation: true, supportsRewindConversation: true,
supportsRewindFiles: true, supportsRewindFiles: true,
supportsRewindBoth: true, supportsRewindBoth: true,
supportsSteering: true,
}; };
const DEFAULT_MODES: AgentMode[] = [ const DEFAULT_MODES: AgentMode[] = [
@@ -899,14 +898,6 @@ function buildClaudePlanPermissionActions(
return actions; return actions;
} }
function buildClaudePlanActions(resumeMode: PermissionMode | null): AgentPlanAction[] {
return buildClaudePlanPermissionActions(resumeMode).map(({ id, label, variant }) => ({
id,
label,
variant,
}));
}
interface TimelineFragment { interface TimelineFragment {
kind: "assistant" | "reasoning"; kind: "assistant" | "reasoning";
text: string; text: string;
@@ -1594,7 +1585,6 @@ class ClaudeAgentSession implements AgentSession {
private toolUseIndexToId = new Map<number, string>(); private toolUseIndexToId = new Map<number, string>();
private toolUseInputBuffers = new Map<string, string>(); private toolUseInputBuffers = new Map<string, string>();
private pendingPermissions = new Map<string, PendingPermission>(); private pendingPermissions = new Map<string, PendingPermission>();
private pendingPlans = new Map<string, string>();
private activeForegroundTurnId: string | null = null; private activeForegroundTurnId: string | null = null;
private autonomousTurn: AutonomousTurnState | null = null; private autonomousTurn: AutonomousTurnState | null = null;
private readonly subscribers = new Set<(event: AgentStreamEvent) => void>(); private readonly subscribers = new Set<(event: AgentStreamEvent) => void>();
@@ -1800,6 +1790,20 @@ class ClaudeAgentSession implements AgentSession {
return { turnId }; 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 { subscribe(callback: (event: AgentStreamEvent) => void): () => void {
this.subscribers.add(callback); this.subscribers.add(callback);
return () => { return () => {
@@ -1930,140 +1934,82 @@ class ClaudeAgentSession implements AgentSession {
} }
getPendingPermissions(): AgentPermissionRequest[] { getPendingPermissions(): AgentPermissionRequest[] {
const hiddenPlanRequestIds = new Set(this.pendingPlans.values()); return Array.from(this.pendingPermissions.values()).map((entry) => entry.request);
return Array.from(this.pendingPermissions.values())
.filter((entry) => !hiddenPlanRequestIds.has(entry.request.id))
.map((entry) => entry.request);
} }
private clearPendingPlanForPermission(requestId: string): void { async respondToPermission(requestId: string, response: AgentPermissionResponse): Promise<void> {
for (const [planId, pendingRequestId] of this.pendingPlans) {
if (pendingRequestId === requestId) {
this.pendingPlans.delete(planId);
}
}
}
async respondToPermission(
requestId: string,
response: AgentPermissionResponse,
emitResolution = true,
): Promise<void> {
const pending = this.pendingPermissions.get(requestId); const pending = this.pendingPermissions.get(requestId);
if (!pending) { if (!pending) {
throw new Error(`No pending permission request with id '${requestId}'`); throw new Error(`No pending permission request with id '${requestId}'`);
} }
this.pendingPermissions.delete(requestId); this.pendingPermissions.delete(requestId);
this.clearPendingPlanForPermission(requestId);
pending.cleanup?.(); pending.cleanup?.();
if (response.behavior === "allow") { if (response.behavior === "allow") {
await this.resolveAllowedPermission(pending, response); if (pending.request.kind === "plan") {
const selectedActionId = response.selectedActionId;
const shouldResumePriorMode =
selectedActionId === "implement_resume" && this.planResumeMode === "bypassPermissions";
const targetMode: PermissionMode = shouldResumePriorMode
? "bypassPermissions"
: "acceptEdits";
await this.setMode(targetMode);
this.pushToolCall(
mapClaudeCompletedToolCall({
name: "plan_approval",
callId: pending.request.id,
input: pending.request.input ?? null,
output: {
approved: true,
actionId: selectedActionId ?? "implement",
},
}),
);
}
const updatedInput =
pending.request.kind === "question"
? normalizeClaudeAskUserQuestionUpdatedInput(
response.updatedInput,
pending.request.input ?? undefined,
)
: (response.updatedInput ?? pending.request.input ?? {});
const result: PermissionResult = {
behavior: "allow",
updatedInput,
updatedPermissions: this.normalizePermissionUpdates(response.updatedPermissions),
};
pending.resolve(result);
} else { } else {
this.resolveDeniedPermission(pending, response); if (pending.request.kind === "tool") {
this.pushToolCall(
mapClaudeFailedToolCall({
name: pending.request.name,
callId:
(typeof pending.request.metadata?.toolUseId === "string"
? pending.request.metadata.toolUseId
: null) ?? pending.request.id,
input: pending.request.input ?? null,
output: null,
error: { message: response.message ?? "Permission denied" },
}),
);
}
const result: PermissionResult = {
behavior: "deny",
message: response.message ?? "Permission request denied",
interrupt: response.interrupt,
};
pending.resolve(result);
} }
if (emitResolution) { this.pushEvent({
this.pushEvent({ type: "permission_resolved",
type: "permission_resolved", provider: "claude",
provider: "claude", requestId,
requestId, resolution: response,
resolution: response,
});
}
}
private async resolveAllowedPermission(
pending: PendingPermission,
response: Extract<AgentPermissionResponse, { behavior: "allow" }>,
): Promise<void> {
if (pending.request.kind === "plan") {
const selectedActionId = response.selectedActionId;
const shouldResumePriorMode =
selectedActionId === "implement_resume" && this.planResumeMode === "bypassPermissions";
const targetMode: PermissionMode = shouldResumePriorMode
? "bypassPermissions"
: "acceptEdits";
await this.setMode(targetMode);
this.pushToolCall(
mapClaudeCompletedToolCall({
name: "plan_approval",
callId: pending.request.id,
input: pending.request.input ?? null,
output: {
approved: true,
actionId: selectedActionId ?? "implement",
},
}),
);
}
const updatedInput =
pending.request.kind === "question"
? normalizeClaudeAskUserQuestionUpdatedInput(
response.updatedInput,
pending.request.input ?? undefined,
)
: (response.updatedInput ?? pending.request.input ?? {});
pending.resolve({
behavior: "allow",
updatedInput,
updatedPermissions: this.normalizePermissionUpdates(response.updatedPermissions),
}); });
} }
private resolveDeniedPermission(
pending: PendingPermission,
response: Extract<AgentPermissionResponse, { behavior: "deny" }>,
): void {
if (pending.request.kind === "tool") {
this.pushToolCall(
mapClaudeFailedToolCall({
name: pending.request.name,
callId:
(typeof pending.request.metadata?.toolUseId === "string"
? pending.request.metadata.toolUseId
: null) ?? pending.request.id,
input: pending.request.input ?? null,
output: null,
error: { message: response.message ?? "Permission denied" },
}),
);
}
pending.resolve({
behavior: "deny",
message: response.message ?? "Permission request denied",
interrupt: response.interrupt,
});
}
async respondToPlan(planId: string, response: AgentPlanResponse): Promise<void> {
const requestId = this.pendingPlans.get(planId);
if (!requestId) {
throw new Error(`No pending Claude plan with id '${planId}'`);
}
this.pendingPlans.delete(planId);
if (response.actionId === "implement" || response.actionId === "implement_resume") {
await this.respondToPermission(
requestId,
{ behavior: "allow", selectedActionId: response.actionId },
false,
);
return;
}
if (response.actionId === "reject") {
await this.respondToPermission(
requestId,
{ behavior: "deny", selectedActionId: response.actionId, message: response.feedback },
false,
);
return;
}
throw new Error(`Unknown Claude plan action '${response.actionId}'`);
}
describePersistence(): AgentPersistenceHandle | null { describePersistence(): AgentPersistenceHandle | null {
if (this.persistence) { if (this.persistence) {
return this.persistence; return this.persistence;
@@ -3843,26 +3789,11 @@ class ClaudeAgentSession implements AgentSession {
metadata: Object.keys(metadata).length ? metadata : undefined, metadata: Object.keys(metadata).length ? metadata : undefined,
}; };
if (kind === "plan" && typeof input.plan === "string") { this.pushEvent({
const planId = `plan-${randomUUID()}`; type: "permission_requested",
this.pendingPlans.set(planId, requestId); provider: "claude",
this.pushEvent({ request,
type: "timeline", });
provider: "claude",
item: {
type: "plan",
planId,
text: input.plan,
actions: buildClaudePlanActions(this.planResumeMode),
},
});
} else {
this.pushEvent({
type: "permission_requested",
provider: "claude",
request,
});
}
return await new Promise<PermissionResult>((resolve, reject) => { return await new Promise<PermissionResult>((resolve, reject) => {
const cleanupFns: Array<() => void> = []; const cleanupFns: Array<() => void> = [];
@@ -3879,11 +3810,6 @@ class ClaudeAgentSession implements AgentSession {
const abortHandler = () => { const abortHandler = () => {
this.pendingPermissions.delete(requestId); this.pendingPermissions.delete(requestId);
for (const [planId, pendingRequestId] of this.pendingPlans) {
if (pendingRequestId === requestId) {
this.pendingPlans.delete(planId);
}
}
cleanup(); cleanup();
reject(new Error("Permission request aborted")); reject(new Error("Permission request aborted"));
}; };
@@ -63,6 +63,7 @@ type CodexTestSession = AgentSession & {
connected: boolean; connected: boolean;
currentThreadId: string | null; currentThreadId: string | null;
activeForegroundTurnId: string | null; activeForegroundTurnId: string | null;
activeAppServerTurnId: string | null;
client: CodexClientLike | null; client: CodexClientLike | null;
}; };
@@ -99,6 +100,7 @@ function createSession(
session.connected = true; session.connected = true;
session.currentThreadId = "test-thread"; session.currentThreadId = "test-thread";
session.activeForegroundTurnId = "test-turn"; session.activeForegroundTurnId = "test-turn";
session.activeAppServerTurnId = "test-turn";
return session; 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 () => { test("passes ephemeral: true to thread/start when constructed as ephemeral", async () => {
const requests: Array<{ method: string; params: unknown }> = []; const requests: Array<{ method: string; params: unknown }> = [];
const fakeClient: CodexClientLike = { 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 () => { test("resolves Codex skill slash commands into app-server skill input", async () => {
const session = createSession(); const session = createSession();
const request = vi.fn(async (method: string) => { const request = vi.fn(async (method: string) => {
@@ -1989,7 +2152,7 @@ describe("Codex app-server provider", () => {
}); });
}); });
test("emits an actionable plan item after a successful Codex plan turn", () => { test("emits a synthetic plan approval permission after a successful Codex plan turn", () => {
const session = createSession({ const session = createSession({
featureValues: { plan_mode: true, fast_mode: true }, featureValues: { plan_mode: true, fast_mode: true },
}); });
@@ -2018,20 +2181,27 @@ describe("Codex app-server provider", () => {
), ),
).toBe(false); ).toBe(false);
expect(events.at(-2)).toEqual({ expect(events.at(-2)).toEqual({
type: "timeline", type: "permission_requested",
provider: "codex", provider: "codex",
turnId: "test-turn", turnId: "test-turn",
item: expect.objectContaining({ request: expect.objectContaining({
type: "plan", provider: "codex",
text: "- Inspect the existing auth flow\n- Implement the button behavior", name: "CodexPlanApproval",
kind: "plan",
title: "Plan",
input: {
plan: "- Inspect the existing auth flow\n- Implement the button behavior",
},
actions: [ actions: [
expect.objectContaining({ expect.objectContaining({
id: "reject", id: "reject",
label: "Reject", label: "Reject",
behavior: "deny",
}), }),
expect.objectContaining({ expect.objectContaining({
id: "implement", id: "implement",
label: "Implement", label: "Implement",
behavior: "allow",
}), }),
], ],
}), }),
@@ -2075,12 +2245,16 @@ describe("Codex app-server provider", () => {
}), }),
); );
expect(events.at(-2)).toEqual({ expect(events.at(-2)).toEqual({
type: "timeline", type: "permission_requested",
provider: "codex", provider: "codex",
turnId: "test-turn", turnId: "test-turn",
item: expect.objectContaining({ request: expect.objectContaining({
type: "plan", provider: "codex",
text: "- Inspect README\n- Add a short note", name: "CodexPlanApproval",
kind: "plan",
input: {
plan: "- Inspect README\n- Add a short note",
},
}), }),
}); });
}); });
@@ -2458,7 +2632,7 @@ describe("Codex app-server provider", () => {
]); ]);
}); });
test("responding to a Codex plan item disables plan mode, preserves fast mode, and returns follow-up prompt", async () => { test("approving a synthetic Codex plan permission disables plan mode, preserves fast mode, and returns follow-up prompt", async () => {
const session = createSession({ const session = createSession({
featureValues: { plan_mode: true, fast_mode: true }, featureValues: { plan_mode: true, fast_mode: true },
}); });
@@ -2475,16 +2649,19 @@ describe("Codex app-server provider", () => {
turn: { status: "completed", error: null }, turn: { status: "completed", error: null },
}); });
const plan = events.find( const request = events.find(
(event): event is Extract<AgentStreamEvent, { type: "timeline" }> => (event): event is Extract<AgentStreamEvent, { type: "permission_requested" }> =>
event.type === "timeline" && event.item.type === "plan", event.type === "permission_requested" && event.request.kind === "plan",
); );
expect(plan).toBeDefined(); expect(request).toBeDefined();
if (!plan || plan.item.type !== "plan") { if (!request) {
throw new Error("Expected plan item"); throw new Error("Expected synthetic plan approval permission");
} }
const result = await session.respondToPlan?.(plan.item.planId, { actionId: "implement" }); const result = await session.respondToPermission(request.request.id, {
behavior: "allow",
selectedActionId: "implement",
});
expect(asInternals(session).serviceTier).toBe("fast"); expect(asInternals(session).serviceTier).toBe("fast");
expect(asInternals(session).planModeEnabled).toBe(false); expect(asInternals(session).planModeEnabled).toBe(false);
@@ -2498,10 +2675,18 @@ describe("Codex app-server provider", () => {
expect(result!.followUpPrompt).toEqual( expect(result!.followUpPrompt).toEqual(
expect.stringContaining("The user approved the plan. Implement it now."), expect.stringContaining("The user approved the plan. Implement it now."),
); );
expect(events).not.toContainEqual(expect.objectContaining({ type: "permission_resolved" })); expect(events.at(-1)).toEqual({
type: "permission_resolved",
provider: "codex",
requestId: request.request.id,
resolution: {
behavior: "allow",
selectedActionId: "implement",
},
});
}); });
test("responding to a Codex plan item keeps fast mode disabled when it started disabled", async () => { test("approving a synthetic Codex plan permission keeps fast mode disabled when it started disabled", async () => {
const session = createSession({ const session = createSession({
featureValues: { plan_mode: true, fast_mode: false }, featureValues: { plan_mode: true, fast_mode: false },
}); });
@@ -2518,16 +2703,19 @@ describe("Codex app-server provider", () => {
turn: { status: "completed", error: null }, turn: { status: "completed", error: null },
}); });
const plan = events.find( const request = events.find(
(event): event is Extract<AgentStreamEvent, { type: "timeline" }> => (event): event is Extract<AgentStreamEvent, { type: "permission_requested" }> =>
event.type === "timeline" && event.item.type === "plan", event.type === "permission_requested" && event.request.kind === "plan",
); );
expect(plan).toBeDefined(); expect(request).toBeDefined();
if (!plan || plan.item.type !== "plan") { if (!request) {
throw new Error("Expected plan item"); throw new Error("Expected synthetic plan approval permission");
} }
const result = await session.respondToPlan?.(plan.item.planId, { actionId: "implement" }); const result = await session.respondToPermission(request.request.id, {
behavior: "allow",
selectedActionId: "implement",
});
expect(asInternals(session).serviceTier).toBeNull(); expect(asInternals(session).serviceTier).toBeNull();
expect(asInternals(session).planModeEnabled).toBe(false); expect(asInternals(session).planModeEnabled).toBe(false);
@@ -2583,16 +2771,19 @@ describe("Codex app-server provider", () => {
turn: { status: "completed", error: null }, turn: { status: "completed", error: null },
}); });
const plan = events.find( const permissionRequest = events.find(
(event): event is Extract<AgentStreamEvent, { type: "timeline" }> => (event): event is Extract<AgentStreamEvent, { type: "permission_requested" }> =>
event.type === "timeline" && event.item.type === "plan", event.type === "permission_requested" && event.request.kind === "plan",
); );
expect(plan).toBeDefined(); expect(permissionRequest).toBeDefined();
if (!plan || plan.item.type !== "plan") { if (!permissionRequest) {
throw new Error("Expected plan item"); throw new Error("Expected synthetic plan approval permission");
} }
const result = await session.respondToPlan?.(plan.item.planId, { actionId: "implement" }); const result = await session.respondToPermission(permissionRequest.request.id, {
behavior: "allow",
selectedActionId: "implement",
});
expect(result?.followUpPrompt).toEqual(expect.any(String)); expect(result?.followUpPrompt).toEqual(expect.any(String));
await session.startTurn(result!.followUpPrompt!); await session.startTurn(result!.followUpPrompt!);
@@ -8,9 +8,6 @@ import {
type AgentLaunchContext, type AgentLaunchContext,
type AgentMode, type AgentMode,
type AgentModelDefinition, type AgentModelDefinition,
type AgentPlanAction,
type AgentPlanResponse,
type AgentPlanResult,
type McpServerConfig, type McpServerConfig,
type AgentPersistenceHandle, type AgentPersistenceHandle,
type AgentPermissionRequest, type AgentPermissionRequest,
@@ -44,7 +41,6 @@ import os from "node:os";
import path from "node:path"; import path from "node:path";
import { z } from "zod"; import { z } from "zod";
import { renderPromptAttachmentAsText } from "../prompt-attachments.js"; import { renderPromptAttachmentAsText } from "../prompt-attachments.js";
import { planItemFromToolCall } from "../plan-files.js";
import { composeSystemPromptParts } from "../system-prompt.js"; import { composeSystemPromptParts } from "../system-prompt.js";
import { curateAgentActivity } from "../activity-curator.js"; import { curateAgentActivity } from "../activity-curator.js";
import { import {
@@ -175,6 +171,7 @@ const CODEX_APP_SERVER_CAPABILITIES: AgentCapabilityFlags = {
supportsRewindConversation: true, supportsRewindConversation: true,
supportsRewindFiles: false, supportsRewindFiles: false,
supportsRewindBoth: false, supportsRewindBoth: false,
supportsSteering: true,
}; };
const CODEX_MODES: AgentMode[] = [ const CODEX_MODES: AgentMode[] = [
@@ -945,10 +942,6 @@ function buildPlanPermissionActions(options?: {
return actions; return actions;
} }
function buildPlanActions(): AgentPlanAction[] {
return buildPlanPermissionActions().map(({ id, label, variant }) => ({ id, label, variant }));
}
function buildCodexPlanImplementationPrompt(planText: string): string { function buildCodexPlanImplementationPrompt(planText: string): string {
const normalizedPlan = normalizePlanMarkdown(planText); const normalizedPlan = normalizePlanMarkdown(planText);
if (!normalizedPlan) { if (!normalizedPlan) {
@@ -2896,6 +2889,7 @@ export class CodexAppServerAgentSession implements AgentSession {
private readonly subscribers = new Set<(event: AgentStreamEvent) => void>(); private readonly subscribers = new Set<(event: AgentStreamEvent) => void>();
private nextTurnOrdinal = 0; private nextTurnOrdinal = 0;
private activeForegroundTurnId: string | null = null; private activeForegroundTurnId: string | null = null;
private activeAppServerTurnId: string | null = null;
private cachedRuntimeInfo: AgentRuntimeInfo | null = null; private cachedRuntimeInfo: AgentRuntimeInfo | null = null;
private serviceTier: "fast" | null = null; private serviceTier: "fast" | null = null;
private planModeEnabled = false; private planModeEnabled = false;
@@ -2933,7 +2927,6 @@ export class CodexAppServerAgentSession implements AgentSession {
private latestPlanResult: { callId: string; text: string; turnId: string | null } | null = null; private latestPlanResult: { callId: string; text: string; turnId: string | null } | null = null;
private readonly userMessageTurnIndexes = new Map<string, number>(); private readonly userMessageTurnIndexes = new Map<string, number>();
private readonly userMessageTurnIds: string[] = []; private readonly userMessageTurnIds: string[] = [];
private pendingPlans = new Map<string, { text: string }>();
private pendingManualCompactionStarts = 0; private pendingManualCompactionStarts = 0;
private compactionTriggerByItemId = new Map<string, "auto" | "manual">(); private compactionTriggerByItemId = new Map<string, "auto" | "manual">();
// Codex can report one completed compaction through both channels: // Codex can report one completed compaction through both channels:
@@ -3196,36 +3189,30 @@ export class CodexAppServerAgentSession implements AgentSession {
}; };
} }
private emitPlanFileItemFromToolCall(item: ToolCallTimelineItem): void { private emitSyntheticPlanApprovalRequest(planText: string): void {
void planItemFromToolCall({ item, cwd: this.config.cwd, homeDir: homedir() }) const requestId = `permission-${randomUUID()}`;
.then((planItem) => { const request: AgentPermissionRequest = {
if (planItem) { id: requestId,
this.emitEvent({ type: "timeline", provider: CODEX_PROVIDER, item: planItem });
}
return undefined;
})
.catch((error) => {
this.logger.debug({ error, callId: item.callId }, "Failed to emit plan file item");
});
}
private emitPlanApprovalItem(planText: string): void {
const text = normalizePlanMarkdown(planText);
if (!text) {
return;
}
const planId = `plan-${randomUUID()}`;
this.pendingPlans.set(planId, { text });
this.emitEvent({
type: "timeline",
provider: CODEX_PROVIDER, provider: CODEX_PROVIDER,
item: { name: "CodexPlanApproval",
type: "plan", kind: "plan",
planId, title: "Plan",
text, description: "Review the proposed plan before implementation starts.",
actions: buildPlanActions(), input: { plan: planText },
actions: buildPlanPermissionActions(),
metadata: {
planText,
source: "codex_plan_approval",
}, },
};
this.pendingPermissions.set(requestId, request);
this.pendingPermissionHandlers.set(requestId, {
resolve: () => undefined,
kind: "plan",
planText,
}); });
this.emitEvent({ type: "permission_requested", provider: CODEX_PROVIDER, request });
} }
/** /**
@@ -3496,6 +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> { async run(prompt: AgentPromptInput, options?: AgentRunOptions): Promise<AgentRunResult> {
return runProviderTurn({ return runProviderTurn({
prompt, prompt,
@@ -3528,10 +3522,7 @@ export class CodexAppServerAgentSession implements AgentSession {
throw new Error("Codex client not initialized"); throw new Error("Codex client not initialized");
} }
const slashCommand = await this.resolveSlashCommandInvocation(prompt); const effectivePrompt = await this.buildEffectivePromptInput(prompt);
const effectivePrompt = slashCommand
? await this.buildCommandPromptInput(slashCommand.commandName, slashCommand.args)
: prompt;
if (this.currentThreadId) { if (this.currentThreadId) {
await this.ensureThreadLoaded(); await this.ensureThreadLoaded();
@@ -3543,6 +3534,7 @@ export class CodexAppServerAgentSession implements AgentSession {
const turnId = this.createTurnId(); const turnId = this.createTurnId();
this.activeForegroundTurnId = turnId; this.activeForegroundTurnId = turnId;
this.activeAppServerTurnId = null;
try { try {
this.logTurnStartSummary({ this.logTurnStartSummary({
@@ -3554,9 +3546,15 @@ export class CodexAppServerAgentSession implements AgentSession {
hasDeveloperInstructions: turnStart.hasDeveloperInstructions, hasDeveloperInstructions: turnStart.hasDeveloperInstructions,
hasCodexConfig: turnStart.hasCodexConfig, 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) { } catch (error) {
this.activeForegroundTurnId = null; this.activeForegroundTurnId = null;
this.activeAppServerTurnId = null;
throw error; throw error;
} }
@@ -3598,6 +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 { subscribe(callback: (event: AgentStreamEvent) => void): () => void {
this.subscribers.add(callback); this.subscribers.add(callback);
return () => { return () => {
@@ -3785,29 +3813,6 @@ export class CodexAppServerAgentSession implements AgentSession {
pending.resolve({ answers: {} }); pending.resolve({ answers: {} });
} }
async respondToPlan(
planId: string,
response: AgentPlanResponse,
): Promise<AgentPlanResult | void> {
const pending = this.pendingPlans.get(planId);
if (!pending) {
throw new Error(`No pending Codex app-server plan with id '${planId}'`);
}
this.pendingPlans.delete(planId);
if (response.actionId === "implement" || response.actionId === "implement_resume") {
return {
followUpPrompt: this.preparePlanImplementation({ planText: pending.text }),
};
}
if (response.actionId === "reject") {
return;
}
throw new Error(`Unknown Codex plan action '${response.actionId}'`);
}
private handlePlanPermissionResponse(params: { private handlePlanPermissionResponse(params: {
requestId: string; requestId: string;
response: AgentPermissionResponse; response: AgentPermissionResponse;
@@ -3958,6 +3963,7 @@ export class CodexAppServerAgentSession implements AgentSession {
this.resolvedPermissionRequests.clear(); this.resolvedPermissionRequests.clear();
this.subscribers.clear(); this.subscribers.clear();
this.activeForegroundTurnId = null; this.activeForegroundTurnId = null;
this.activeAppServerTurnId = null;
if (this.client) { if (this.client) {
await this.client.dispose(); await this.client.dispose();
} }
@@ -4579,6 +4585,7 @@ export class CodexAppServerAgentSession implements AgentSession {
return; return;
} }
this.currentTurnId = parsed.turnId; this.currentTurnId = parsed.turnId;
this.activeAppServerTurnId = parsed.turnId;
this.resetTurnTrackingState(); this.resetTurnTrackingState();
this.emitEvent({ type: "turn_started", provider: CODEX_PROVIDER }); this.emitEvent({ type: "turn_started", provider: CODEX_PROVIDER });
} }
@@ -4607,7 +4614,7 @@ export class CodexAppServerAgentSession implements AgentSession {
this.emitEvent({ type: "turn_canceled", provider: CODEX_PROVIDER, reason: "interrupted" }); this.emitEvent({ type: "turn_canceled", provider: CODEX_PROVIDER, reason: "interrupted" });
} else { } else {
if (this.planModeEnabled && this.latestPlanResult?.text) { if (this.planModeEnabled && this.latestPlanResult?.text) {
this.emitPlanApprovalItem(this.latestPlanResult.text); this.emitSyntheticPlanApprovalRequest(this.latestPlanResult.text);
} }
this.emitEvent({ this.emitEvent({
type: "turn_completed", type: "turn_completed",
@@ -4616,6 +4623,7 @@ export class CodexAppServerAgentSession implements AgentSession {
}); });
} }
this.activeForegroundTurnId = null; this.activeForegroundTurnId = null;
this.activeAppServerTurnId = null;
this.resetTurnTrackingState(); this.resetTurnTrackingState();
} }
@@ -4921,9 +4929,6 @@ export class CodexAppServerAgentSession implements AgentSession {
this.warnOnIncompleteEditToolCall(timelineItem, "item_completed", parsed.item); this.warnOnIncompleteEditToolCall(timelineItem, "item_completed", parsed.item);
} }
this.emitEvent({ type: "timeline", provider: CODEX_PROVIDER, item: timelineItem }); this.emitEvent({ type: "timeline", provider: CODEX_PROVIDER, item: timelineItem });
if (timelineItem.type === "tool_call") {
this.emitPlanFileItemFromToolCall(timelineItem);
}
if (timelineItem.type === "assistant_message") { if (timelineItem.type === "assistant_message") {
this.pendingAssistantMessageBoundary = true; this.pendingAssistantMessageBoundary = true;
} }
@@ -24,7 +24,7 @@ describe("Codex app-server provider (real) plan mode", () => {
} }
}); });
test("maps gpt-5.4 markdown plans to a normalized plan item instead of todo items", async () => { test("maps gpt-5.4 markdown plans to a plan tool call instead of todo items", async () => {
const cwd = tmpCwd(); const cwd = tmpCwd();
const client = new CodexAppServerAgentClient(createTestLogger()); const client = new CodexAppServerAgentClient(createTestLogger());
@@ -50,16 +50,18 @@ describe("Codex app-server provider (real) plan mode", () => {
}), }),
); );
const planItem = result.timeline.find((item) => item.type === "plan"); const planCall = result.timeline.find(
(item) => item.type === "tool_call" && item.detail.type === "plan",
);
expect(planItem).toBeDefined(); expect(planCall).toBeDefined();
if (!planItem || planItem.type !== "plan") { if (!planCall || planCall.type !== "tool_call" || planCall.detail.type !== "plan") {
throw new Error("Expected a normalized plan item"); throw new Error("Expected a plan tool call");
} }
expect(planItem.text).toContain("Login"); expect(planCall.detail.text).toContain("Login");
expect(planItem.text).toContain("- "); expect(planCall.detail.text).toContain("- ");
expect(planItem.actions?.some((action) => action.id === "implement")).toBe(true); expect(result.finalText).toBe(planCall.detail.text);
} finally { } finally {
await session.close(); await session.close();
} }
@@ -57,7 +57,6 @@ import {
resolveProviderLaunch, resolveProviderLaunch,
type ProviderRuntimeSettings, type ProviderRuntimeSettings,
} from "../provider-launch-config.js"; } from "../provider-launch-config.js";
import { isPlanFilePath, planItemFromToolCall } from "../plan-files.js";
import { withTimeout } from "../../../utils/promise-timeout.js"; import { withTimeout } from "../../../utils/promise-timeout.js";
import { execCommand } from "../../../utils/spawn.js"; import { execCommand } from "../../../utils/spawn.js";
import { buildToolCallDisplayModel } from "@getpaseo/protocol/tool-call-display"; import { buildToolCallDisplayModel } from "@getpaseo/protocol/tool-call-display";
@@ -2155,22 +2154,6 @@ function appendOpenCodeToolCallTimelineItem(
provider: "opencode", provider: "opencode",
item: timelineItem, item: timelineItem,
}); });
if (
timelineItem.status === "completed" &&
timelineItem.detail.type === "write" &&
timelineItem.detail.content?.trim() &&
isPlanFilePath(timelineItem.detail.filePath)
) {
events.push({
type: "timeline",
provider: "opencode",
item: {
type: "plan",
planId: `plan-file:${timelineItem.detail.filePath}`,
text: timelineItem.detail.content.trim(),
},
});
}
if (timelineItem.detail.type === "sub_agent" && timelineItem.detail.childSessionId) { if (timelineItem.detail.type === "sub_agent" && timelineItem.detail.childSessionId) {
flushOpenCodeSubAgentChildToolParts(timelineItem.detail.childSessionId, state, events); flushOpenCodeSubAgentChildToolParts(timelineItem.detail.childSessionId, state, events);
} }
@@ -3230,28 +3213,9 @@ class OpenCodeAgentSession implements AgentSession {
return; return;
} }
this.notifySubscribers(e, turnId); this.notifySubscribers(e, turnId);
if (e.type === "timeline" && e.item.type === "tool_call") {
this.emitPlanFileItemFromToolCall(e.item, turnId);
}
} }
} }
private emitPlanFileItemFromToolCall(item: ToolCallTimelineItem, turnId: string): void {
void planItemFromToolCall({ item, cwd: this.config.cwd, homeDir: homedir() })
.then((planItem) => {
if (planItem) {
this.notifySubscribers(
{ type: "timeline", provider: "opencode", item: planItem },
turnId,
);
}
return undefined;
})
.catch((error) => {
this.logger.debug({ error, callId: item.callId }, "Failed to emit plan file item");
});
}
private finishForegroundTurn( private finishForegroundTurn(
event: Extract<AgentStreamEvent, { type: "turn_completed" | "turn_failed" | "turn_canceled" }>, event: Extract<AgentStreamEvent, { type: "turn_completed" | "turn_failed" | "turn_canceled" }>,
turnId: string, turnId: string,
@@ -1,273 +0,0 @@
import { afterEach, beforeEach, describe, expect, test } from "vitest";
import { mkdtempSync, rmSync } from "node:fs";
import { tmpdir } from "node:os";
import path from "node:path";
import pino from "pino";
import { createDaemonTestContext, type DaemonTestContext } from "../test-utils/index.js";
import { createMessageCollector, type MessageCollector } from "../test-utils/message-collector.js";
import { CodexAppServerAgentClient } from "../agent/providers/codex-app-server-agent.js";
import { ClaudeAgentClient } from "../agent/providers/claude/agent.js";
import { createTestPaseoDaemon } from "../test-utils/paseo-daemon.js";
import { DaemonClient } from "../test-utils/daemon-client.js";
import { isProviderAvailable } from "./agent-configs.js";
import type { PlanTimelineItem } from "../agent/agent-sdk-types.js";
function tmpCwd(): string {
return mkdtempSync(path.join(tmpdir(), "daemon-plans-"));
}
function waitForPlanMessage(
collector: MessageCollector,
agentId: string,
timeoutMs: number,
): Promise<PlanTimelineItem> {
return new Promise((resolve, reject) => {
const deadline = Date.now() + timeoutMs;
const timer = setInterval(() => {
const message = collector.messages.find((candidate) => {
if (candidate.type !== "agent_stream") return false;
if (candidate.payload.agentId !== agentId) return false;
return (
candidate.payload.event.type === "timeline" &&
candidate.payload.event.item.type === "plan"
);
});
if (message?.type === "agent_stream") {
const event = message.payload.event;
if (event.type === "timeline" && event.item.type === "plan") {
clearInterval(timer);
resolve(event.item);
}
}
if (Date.now() > deadline) {
clearInterval(timer);
reject(new Error(`Timed out waiting for plan item after ${timeoutMs}ms`));
}
}, 100);
});
}
describe("daemon E2E - first-class plans", () => {
let ctx: DaemonTestContext;
let collector: MessageCollector;
beforeEach(async () => {
ctx = await createDaemonTestContext();
collector = createMessageCollector(ctx.client);
});
afterEach(async () => {
collector.unsubscribe();
await ctx.cleanup();
}, 60_000);
test("surfaces an actionable plan and routes the response through the daemon", async () => {
const cwd = tmpCwd();
try {
const agent = await ctx.client.createAgent({
provider: "codex",
cwd,
title: "Plan E2E",
modeId: "full-access",
});
collector.clear();
await ctx.client.sendMessage(agent.id, "Emit an actionable plan.");
const finalState = await ctx.client.waitForFinish(agent.id, 5_000);
expect(finalState.status).toBe("idle");
const planMessage = collector.messages.find((message) => {
if (message.type !== "agent_stream") return false;
if (message.payload.agentId !== agent.id) return false;
return (
message.payload.event.type === "timeline" && message.payload.event.item.type === "plan"
);
});
expect(planMessage?.type).toBe("agent_stream");
if (planMessage?.type !== "agent_stream") {
throw new Error("Expected plan stream message");
}
const event = planMessage.payload.event;
if (event.type !== "timeline" || event.item.type !== "plan") {
throw new Error("Expected normalized plan item");
}
expect(event.item.actions).toEqual([
{ id: "implement", label: "Implement", variant: "primary" },
]);
const timeline = await ctx.client.fetchAgentTimeline(agent.id, {
direction: "tail",
limit: 0,
projection: "canonical",
});
expect(
timeline.entries.some(
(entry) => entry.item.type === "plan" && entry.item.planId === event.item.planId,
),
).toBe(true);
const response = await ctx.client.respondToPlan(agent.id, event.item.planId, {
actionId: "implement",
});
expect(response).toMatchObject({
agentId: agent.id,
planId: event.item.planId,
ok: true,
error: null,
});
} finally {
rmSync(cwd, { recursive: true, force: true });
}
}, 30_000);
test("surfaces a plan file as a non-actionable plan", async () => {
const cwd = tmpCwd();
try {
const agent = await ctx.client.createAgent({
provider: "opencode",
cwd,
title: "Plan File E2E",
modeId: "full-access",
});
collector.clear();
await ctx.client.sendMessage(agent.id, "Emit a plan file.");
await ctx.client.waitForFinish(agent.id, 5_000);
const timeline = await ctx.client.fetchAgentTimeline(agent.id, {
direction: "tail",
limit: 0,
projection: "canonical",
});
const plan = timeline.entries.find(
(entry) =>
entry.item.type === "plan" && entry.item.planId === "plan-file:.paseo/plans/fake.md",
);
expect(plan?.item).toEqual({
type: "plan",
planId: "plan-file:.paseo/plans/fake.md",
text: "# File plan\n\n- From disk",
});
const response = await ctx.client.respondToPlan(agent.id, "plan-file:.paseo/plans/fake.md", {
actionId: "implement",
});
expect(response.ok).toBe(false);
expect(response.error).toContain("No pending fake plan");
} finally {
rmSync(cwd, { recursive: true, force: true });
}
}, 30_000);
});
describe("daemon E2E - first-class plans with real providers", () => {
test("real Codex plan mode surfaces a normalized actionable plan", async (context) => {
if (!(await isProviderAvailable("codex"))) {
context.skip();
}
const cwd = tmpCwd();
const logger = pino({ level: "silent" });
const daemon = await createTestPaseoDaemon({
agentClients: { codex: new CodexAppServerAgentClient(logger) },
logger,
});
const client = new DaemonClient({ url: `ws://127.0.0.1:${daemon.port}/ws` });
try {
await client.connect();
await client.fetchAgents({ subscribe: { subscriptionId: "real-codex-plan" } });
const agent = await client.createAgent({
provider: "codex",
cwd,
title: "Real Codex Plan E2E",
modeId: "auto",
model: "gpt-5.4",
thinkingOptionId: "medium",
featureValues: { plan_mode: true },
});
await client.sendMessage(
agent.id,
"You are in plan mode. Produce a markdown plan with a short heading and exactly 3 bullets for implementing a login screen. Do not ask questions.",
);
await client.waitForFinish(agent.id, 240_000);
const timeline = await client.fetchAgentTimeline(agent.id, {
direction: "tail",
limit: 0,
projection: "canonical",
});
const plan = timeline.entries.find((entry) => entry.item.type === "plan");
expect(plan?.item.type).toBe("plan");
if (!plan || plan.item.type !== "plan") {
throw new Error("Expected normalized plan item");
}
expect(plan.item.text).toContain("Login");
expect(plan.item.actions?.some((action) => action.id === "implement")).toBe(true);
} finally {
await client.close().catch(() => undefined);
await daemon.close();
rmSync(cwd, { recursive: true, force: true });
}
}, 300_000);
test("real Claude plan mode surfaces a normalized actionable plan", async (context) => {
if (!(await isProviderAvailable("claude"))) {
context.skip();
}
const cwd = tmpCwd();
const logger = pino({ level: "silent" });
const daemon = await createTestPaseoDaemon({
agentClients: { claude: new ClaudeAgentClient({ logger }) },
logger,
});
const client = new DaemonClient({ url: `ws://127.0.0.1:${daemon.port}/ws` });
const collector = createMessageCollector(client);
try {
await client.connect();
await client.fetchAgents({ subscribe: { subscriptionId: "real-claude-plan" } });
const agent = await client.createAgent({
provider: "claude",
cwd,
title: "Real Claude Plan E2E",
modeId: "plan",
model: "haiku",
});
collector.clear();
await client.sendMessage(
agent.id,
[
"Create a short implementation plan for a login screen.",
"Use plan mode and call ExitPlanMode with a markdown plan.",
"Do not edit files.",
].join(" "),
);
const plan = await waitForPlanMessage(collector, agent.id, 120_000);
expect(plan.text).toContain("login");
expect(plan.actions?.some((action) => action.id === "implement")).toBe(true);
const snapshot = await client.fetchAgent(agent.id);
expect(snapshot.agent?.pendingPermissions ?? []).toEqual([]);
const response = await client.respondToPlan(agent.id, plan.planId, { actionId: "reject" });
expect(response).toMatchObject({
agentId: agent.id,
planId: plan.planId,
ok: true,
error: null,
});
} finally {
collector.unsubscribe();
await client.close().catch(() => undefined);
await daemon.close();
rmSync(cwd, { recursive: true, force: true });
}
}, 180_000);
});
@@ -63,49 +63,6 @@ describe("serializeAgentStreamEvent", () => {
expect(serialized.item.messageId).toBe("m1"); expect(serialized.item.messageId).toBe("m1");
}); });
test("accepts normalized plan timeline items", () => {
const event: AgentStreamEvent = {
type: "timeline",
provider: "codex",
item: {
type: "plan",
planId: "plan-1",
text: "# Plan\n\n- Ship it",
actions: [{ id: "implement", label: "Implement", variant: "primary" }],
},
};
const serialized = serializeAgentStreamEvent(event);
expect(serialized).toMatchObject({
type: "timeline",
item: {
type: "plan",
planId: "plan-1",
text: "# Plan\n\n- Ship it",
actions: [{ id: "implement", label: "Implement", variant: "primary" }],
},
});
});
test("accepts plan response requests", () => {
const parsed = SessionInboundMessageSchema.parse({
type: "agent.plan.respond.request",
agentId: "agent-1",
planId: "plan-1",
actionId: "implement",
requestId: "req-plan-1",
});
expect(parsed).toMatchObject({
type: "agent.plan.respond.request",
agentId: "agent-1",
planId: "plan-1",
actionId: "implement",
requestId: "req-plan-1",
});
});
test("passes canonical tool_call payloads through unchanged", () => { test("passes canonical tool_call payloads through unchanged", () => {
const event: AgentStreamEvent = { const event: AgentStreamEvent = {
type: "timeline", type: "timeline",
+2 -97
View File
@@ -68,7 +68,6 @@ import { ensureAgentLoaded } from "./agent/agent-loading.js";
import { import {
formatSystemNotificationPrompt, formatSystemNotificationPrompt,
sendPromptToAgent, sendPromptToAgent,
startAgentRun,
waitForAgentRunStartWithTimeout, waitForAgentRunStartWithTimeout,
unarchiveAgentState, unarchiveAgentState,
} from "./agent/agent-prompt.js"; } from "./agent/agent-prompt.js";
@@ -143,8 +142,6 @@ import {
type AgentPromptInput, type AgentPromptInput,
type AgentRunOptions, type AgentRunOptions,
type AgentSessionConfig, type AgentSessionConfig,
type AgentStreamEvent,
type AgentTimelineItem,
type ProviderSnapshotEntry, type ProviderSnapshotEntry,
} from "./agent/agent-sdk-types.js"; } from "./agent/agent-sdk-types.js";
import type { StoredAgentRecord } from "./agent/agent-storage.js"; import type { StoredAgentRecord } from "./agent/agent-storage.js";
@@ -719,45 +716,6 @@ function parseClientCapabilities(
return new Set(result); return new Set(result);
} }
function projectTimelineItemForClient(
item: AgentTimelineItem,
capabilities: ReadonlySet<ClientCapability>,
): AgentTimelineItem {
if (item.type !== "plan" || capabilities.has(CLIENT_CAPS.firstClassPlans)) {
return item;
}
// COMPAT(firstClassPlans): added in v0.1.82, remove shim after 2026-11-28.
return {
type: "tool_call",
callId: item.planId,
name: "Plan",
status: "completed",
error: null,
detail: {
type: "plan",
text: item.text,
},
};
}
function projectAgentStreamEventForClient(
event: AgentStreamEvent,
capabilities: ReadonlySet<ClientCapability>,
): AgentStreamEvent {
if (event.type !== "timeline") {
return event;
}
const item = projectTimelineItemForClient(event.item, capabilities);
if (item === event.item) {
return event;
}
return {
...event,
item,
};
}
/** /**
* Session represents a single connected client session. * Session represents a single connected client session.
* It owns all state management, orchestration logic, and message processing. * It owns all state management, orchestration logic, and message processing.
@@ -1377,11 +1335,7 @@ export class Session {
}); });
} }
const projectedEvent = projectAgentStreamEventForClient( const serializedEvent = serializeAgentStreamEvent(event.event);
event.event,
this.clientCapabilities,
);
const serializedEvent = serializeAgentStreamEvent(projectedEvent);
if (!serializedEvent) { if (!serializedEvent) {
return; return;
} }
@@ -1786,7 +1740,6 @@ export class Session {
const promise = const promise =
this.dispatchVoiceAndControlMessage(msg) ?? this.dispatchVoiceAndControlMessage(msg) ??
this.dispatchAgentRewindMessage(msg) ?? this.dispatchAgentRewindMessage(msg) ??
this.dispatchAgentPlanMessage(msg) ??
this.dispatchAgentLifecycleMessage(msg) ?? this.dispatchAgentLifecycleMessage(msg) ??
this.dispatchAgentConfigMessage(msg) ?? this.dispatchAgentConfigMessage(msg) ??
this.dispatchCheckoutMessage(msg) ?? this.dispatchCheckoutMessage(msg) ??
@@ -1798,13 +1751,6 @@ export class Session {
if (promise) await promise; if (promise) await promise;
} }
private dispatchAgentPlanMessage(msg: SessionInboundMessage): Promise<void> | undefined {
if (msg.type === "agent.plan.respond.request") {
return this.handleAgentPlanRespondRequest(msg);
}
return undefined;
}
private dispatchVoiceAndControlMessage(msg: SessionInboundMessage): Promise<void> | undefined { private dispatchVoiceAndControlMessage(msg: SessionInboundMessage): Promise<void> | undefined {
switch (msg.type) { switch (msg.type) {
case "voice_audio_chunk": case "voice_audio_chunk":
@@ -4615,47 +4561,6 @@ export class Session {
} }
} }
private async handleAgentPlanRespondRequest(
msg: Extract<SessionInboundMessage, { type: "agent.plan.respond.request" }>,
): Promise<void> {
const { agentId, planId, actionId, feedback, requestId } = msg;
try {
const result = await this.agentManager.respondToPlan(agentId, planId, { actionId, feedback });
this.emit({
type: "agent.plan.respond.response",
payload: {
requestId,
agentId,
planId,
ok: true,
error: null,
},
});
if (result?.followUpPrompt) {
startAgentRun(this.agentManager, agentId, result.followUpPrompt, this.sessionLogger, {
replaceRunning: true,
});
}
} catch (error) {
const message = getErrorMessage(error);
this.sessionLogger.error(
{ err: error, agentId, planId, actionId },
"Failed to respond to plan",
);
this.emit({
type: "agent.plan.respond.response",
payload: {
requestId,
agentId,
planId,
ok: false,
error: message,
},
});
}
}
private async handleCheckoutStatusRequest( private async handleCheckoutStatusRequest(
msg: Extract<SessionInboundMessage, { type: "checkout_status_request" }>, msg: Extract<SessionInboundMessage, { type: "checkout_status_request" }>,
): Promise<void> { ): Promise<void> {
@@ -7660,7 +7565,7 @@ export class Session {
hasNewer, hasNewer,
entries: entries.map((entry) => ({ entries: entries.map((entry) => ({
provider: snapshot.provider, provider: snapshot.provider,
item: projectTimelineItemForClient(entry.item, this.clientCapabilities), item: entry.item,
timestamp: entry.timestamp, timestamp: entry.timestamp,
seqStart: entry.seqStart, seqStart: entry.seqStart,
seqEnd: entry.seqEnd, seqEnd: entry.seqEnd,
@@ -10,8 +10,6 @@ import type {
AgentLaunchContext, AgentLaunchContext,
AgentMode, AgentMode,
AgentModelDefinition, AgentModelDefinition,
AgentPlanResponse,
AgentPlanResult,
AgentPersistenceHandle, AgentPersistenceHandle,
AgentPromptInput, AgentPromptInput,
AgentRunOptions, AgentRunOptions,
@@ -306,7 +304,6 @@ class FakeAgentSession implements AgentSession {
private memoryMarker: string | null = null; private memoryMarker: string | null = null;
private pendingPermissions: AgentPermissionRequest[] = []; private pendingPermissions: AgentPermissionRequest[] = [];
private permissionGate: Deferred<AgentPermissionResponse> | null = null; private permissionGate: Deferred<AgentPermissionResponse> | null = null;
private pendingPlans = new Map<string, { text: string }>();
private readonly historyPath: string; private readonly historyPath: string;
private readonly subscribers = new Set<(event: AgentStreamEvent) => void>(); private readonly subscribers = new Set<(event: AgentStreamEvent) => void>();
private nextTurnOrdinal = 0; private nextTurnOrdinal = 0;
@@ -525,55 +522,6 @@ class FakeAgentSession implements AgentSession {
this.notifySubscribers(completed); this.notifySubscribers(completed);
} }
private async emitActionablePlanTurn(text: string): Promise<void> {
const planId = `fake-plan-${randomUUID()}`;
const planText = text.includes("custom plan body") ? "custom plan body" : "# Plan\n\n- Test it";
this.pendingPlans.set(planId, { text: planText });
const planEvent: AgentStreamEvent = {
type: "timeline",
provider: this.providerName,
item: {
type: "plan",
planId,
text: planText,
actions: [{ id: "implement", label: "Implement", variant: "primary" }],
},
};
await this.appendHistoryEvent(planEvent);
this.notifySubscribers(planEvent);
const completed: AgentStreamEvent = {
type: "turn_completed",
provider: this.providerName,
usage: { inputTokens: 1, outputTokens: 1 },
};
await this.appendHistoryEvent(completed);
this.notifySubscribers(completed);
}
private async emitPlanFileTurn(): Promise<void> {
const planEvent: AgentStreamEvent = {
type: "timeline",
provider: this.providerName,
item: {
type: "plan",
planId: "plan-file:.paseo/plans/fake.md",
text: "# File plan\n\n- From disk",
},
};
await this.appendHistoryEvent(planEvent);
this.notifySubscribers(planEvent);
const completed: AgentStreamEvent = {
type: "turn_completed",
provider: this.providerName,
usage: { inputTokens: 1, outputTokens: 1 },
};
await this.appendHistoryEvent(completed);
this.notifySubscribers(completed);
}
private async resolveToolPermission(tool: { private async resolveToolPermission(tool: {
name: string; name: string;
input?: Record<string, unknown>; input?: Record<string, unknown>;
@@ -781,16 +729,6 @@ class FakeAgentSession implements AgentSession {
return; return;
} }
if (textPrompt.toLowerCase().includes("emit an actionable plan")) {
await this.emitActionablePlanTurn(textPrompt);
return;
}
if (textPrompt.toLowerCase().includes("emit a plan file")) {
await this.emitPlanFileTurn();
return;
}
const tool = buildToolCallForPrompt(this.providerName, textPrompt); const tool = buildToolCallForPrompt(this.providerName, textPrompt);
if (tool) { if (tool) {
const returnedEarly = await this.emitToolCallTurn(tool, textPrompt); const returnedEarly = await this.emitToolCallTurn(tool, textPrompt);
@@ -896,20 +834,6 @@ class FakeAgentSession implements AgentSession {
this.permissionGate = null; this.permissionGate = null;
} }
async respondToPlan(
planId: string,
response: AgentPlanResponse,
): Promise<AgentPlanResult | void> {
const pending = this.pendingPlans.get(planId);
if (!pending) {
throw new Error(`No pending fake plan with id '${planId}'`);
}
this.pendingPlans.delete(planId);
if (response.actionId === "implement") {
return { followUpPrompt: `Implement fake plan:\n${pending.text}` };
}
}
describePersistence(): AgentPersistenceHandle | null { describePersistence(): AgentPersistenceHandle | null {
return buildPersistence( return buildPersistence(
this.providerName, this.providerName,
@@ -1043,8 +1043,6 @@ export class VoiceAssistantWebSocketServer {
providersSnapshot: true, providersSnapshot: true,
// COMPAT(checkoutGithubSetAutoMerge): added in v0.1.75, remove gate after 2026-11-13. // COMPAT(checkoutGithubSetAutoMerge): added in v0.1.75, remove gate after 2026-11-13.
checkoutGithubSetAutoMerge: true, checkoutGithubSetAutoMerge: true,
// COMPAT(firstClassPlans): added in v0.1.82, remove gate after 2026-11-28.
firstClassPlans: true,
// COMPAT(daemonStatusRpc): added in v0.1.76, remove gate after 2026-11-18. // COMPAT(daemonStatusRpc): added in v0.1.76, remove gate after 2026-11-18.
daemonStatusRpc: true, daemonStatusRpc: true,
// COMPAT(terminalRestoreModes): added in v0.1.81, remove gate after 2026-11-23. // COMPAT(terminalRestoreModes): added in v0.1.81, remove gate after 2026-11-23.
@@ -73,18 +73,6 @@ const LegacyAgentSnapshotPayloadSchema = AgentSnapshotPayloadSchema.extend({
capabilities: LegacyAgentCapabilityFlagsSchema, capabilities: LegacyAgentCapabilityFlagsSchema,
}); });
const LegacyPlanToolCallSchema = z.object({
type: z.literal("tool_call"),
callId: z.string(),
name: z.string(),
status: z.enum(["running", "completed", "failed", "canceled"]),
error: z.unknown().nullable(),
detail: z.object({
type: z.literal("plan"),
text: z.string(),
}),
});
interface SessionInternals { interface SessionInternals {
handleFetchAgentTimelineRequest: ( handleFetchAgentTimelineRequest: (
message: Extract< message: Extract<
@@ -244,16 +232,6 @@ function createSessionForWireCompatTest(options?: {
timestamp: "2026-05-02T00:00:00.200Z", timestamp: "2026-05-02T00:00:00.200Z",
item: { type: "assistant_message", text: "done" }, item: { type: "assistant_message", text: "done" },
}, },
{
seq: 4,
timestamp: "2026-05-02T00:00:00.300Z",
item: {
type: "plan",
planId: "plan-1",
text: "# Plan\n\n- Do the thing",
actions: [{ id: "implement", label: "Implement", variant: "primary" }],
},
},
]; ];
const session = new Session({ const session = new Session({
@@ -389,38 +367,6 @@ describe("wire compatibility", () => {
expect(currentParsed.payload.entries[0]?.collapsed).toContain("reasoning_merge"); expect(currentParsed.payload.entries[0]?.collapsed).toContain("reasoning_merge");
}); });
test("downgrades plan timeline items for clients that do not declare the capability", async () => {
const response = await emitTimelineResponse();
const entry = response.payload.entries.find((item) => item.seqStart === 4);
expect(entry?.item).toEqual({
type: "tool_call",
callId: "plan-1",
name: "Plan",
status: "completed",
error: null,
detail: {
type: "plan",
text: "# Plan\n\n- Do the thing",
},
});
expect(() => LegacyPlanToolCallSchema.parse(entry?.item)).not.toThrow();
});
test("preserves plan timeline items for clients that declare the capability", async () => {
const response = await emitTimelineResponse({
[CLIENT_CAPS.firstClassPlans]: true,
});
const entry = response.payload.entries.find((item) => item.seqStart === 4);
expect(entry?.item).toEqual({
type: "plan",
planId: "plan-1",
text: "# Plan\n\n- Do the thing",
actions: [{ id: "implement", label: "Implement", variant: "primary" }],
});
});
test("sub_agent tool-call payload still parses against the v0.1.65-beta.3 schema", () => { test("sub_agent tool-call payload still parses against the v0.1.65-beta.3 schema", () => {
const parsed = LegacySubAgentToolCallSchema.parse({ const parsed = LegacySubAgentToolCallSchema.parse({
type: "tool_call", type: "tool_call",