Compare commits

..
Author SHA1 Message Date
paseo-bot[bot]andJacob Dietle 3b5aabe22f test(app): reproduce permission alert loss after a slow reconnect
Preserve the failing reload journey for #7078. The cached directory acknowledges an agent sequence after discarding its pending permission; a changes read leaves the inactive tab showing a running ring. Delay only the reconnect handshake to make the cache-first ordering deterministic. No production fix is included.

Co-authored-by: Jacob Dietle <128077360+jacob-dietle@users.noreply.github.com>
2026-10-17 23:16:05 +02:00
17 changed files with 281 additions and 623 deletions
@@ -5,7 +5,6 @@ import {
seedMockAgentWorkspace,
type MockAgentWorkspace,
} from "../support/helpers/mock-agent";
import { waitForTimelineGeometryToSettle } from "../support/helpers/timeline-pagination";
import { installDaemonWebSocketGate } from "../support/helpers/daemon-websocket-gate";
import {
expectReconnectingToastGone,
@@ -125,49 +124,6 @@ async function closeFind(page: Page) {
.toEqual([]);
}
async function expectChatKeyboardScroll(page: Page, key: "PageUp" | "PageDown") {
// The scrolling surface has no accessible name; use its existing test id.
const viewport = page.getByTestId("agent-chat-scroll");
await waitForTimelineGeometryToSettle(page);
const before = await viewport.evaluate((element) => element.scrollTop);
await page.keyboard.press(key);
const direction = key === "PageUp" ? -1 : 1;
await expect
.poll(
async () => direction * ((await viewport.evaluate((element) => element.scrollTop)) - before),
)
.toBeGreaterThan(100);
}
test("keeps keyboard scrolling the chat after closing Find", async ({ page }, testInfo) => {
const agent = await seedMockAgentWorkspace({
repoPrefix: "chat-find-scroll-focus-",
title: "Chat Find keyboard scrolling",
initialPrompt: "Render the response",
featureValues: { mockAssistantResponse: TALL_RESPONSE },
});
try {
await agent.client.waitForFinish(agent.agentId, 15_000);
await openAgentRoute(page, agent);
await expect(page.getByTestId("assistant-message").first()).toBeVisible();
await searchChat(page, "Closing paragraph");
await expectHighlight(page, "Closing paragraph");
await waitForTimelineGeometryToSettle(page);
await closeFind(page);
await expectChatKeyboardScroll(page, "PageUp");
await page.screenshot({ path: testInfo.outputPath("chat-after-find-page-up.png") });
await expectChatKeyboardScroll(page, "PageDown");
await searchChat(page, "Closing paragraph");
await expectHighlight(page, "Closing paragraph");
await waitForTimelineGeometryToSettle(page);
await page.getByRole("button", { name: "Close Find", exact: true }).click();
await expect(query(page)).toHaveCount(0);
await expectChatKeyboardScroll(page, "PageUp");
} finally {
await agent.cleanup();
}
});
test("loads older history through prompt navigation and finds repeated rendered Markdown", async ({
page,
}, testInfo) => {
@@ -0,0 +1,93 @@
import { expect, test, type Page } from "../support/fixtures";
import { openAgentRoute, seedMockAgentWorkspace } from "../support/helpers/mock-agent";
import { waitForPermissionPrompt } from "../support/helpers/permissions";
import { readReplicaCache } from "../support/helpers/replica-cache-storage";
import { daemonWsRoutePattern } from "../support/helpers/daemon-port";
async function selectConversation(page: Page, title: string): Promise<void> {
await page.getByRole("button", { name: title, exact: true }).click();
}
async function waitForDirectoryCheckpoint(page: Page): Promise<void> {
await expect
.poll(async () => {
const cache = await readReplicaCache(page);
return cache?.hosts?.some((host) => host.directorySync !== undefined);
})
.toBe(true);
}
async function expectWaitingTab(page: Page, title: string): Promise<void> {
const tab = page.getByRole("button", { name: title, exact: true });
await expect(tab).toBeVisible();
await expect(tab.getByRole("progressbar", { name: "Agent running" })).toHaveCount(0);
// The alert SVG has no accessible name; the provider icon is the other SVG.
await expect(tab.locator("svg")).toHaveCount(2);
}
function observeAgentSyncMode(page: Page): () => unknown {
let mode: unknown;
page.on("websocket", (socket) => {
socket.on("framereceived", ({ payload }) => {
const envelope = JSON.parse(String(payload));
if (envelope.type === "session" && envelope.message.type === "fetch_agents_response") {
mode = envelope.message.payload.sync.mode;
}
});
});
return () => mode;
}
async function reloadWithSlowReconnect(page: Page): Promise<void> {
await page.routeWebSocket(daemonWsRoutePattern(), async (socket) => {
// Fault injection: restore the disk baseline before the handshake completes.
await new Promise((resolve) => setTimeout(resolve, 2_000));
socket.connectToServer();
});
await page.reload();
}
test("keeps an inactive agent's permission alert after reload", async ({ page }, testInfo) => {
test.setTimeout(180_000);
const syncMode = observeAgentSyncMode(page);
const waiting = await seedMockAgentWorkspace({
repoPrefix: "permission-reload-",
title: "Needs approval",
model: "thirty-minute-stream",
initialPrompt: "Emit synthetic plan approval.",
});
try {
await openAgentRoute(page, waiting);
await waitForPermissionPrompt(page);
await waiting.client.createAgent({
provider: "mock",
cwd: waiting.cwd,
workspaceId: waiting.workspaceId,
title: "Other conversation",
modeId: "load-test",
model: "thirty-minute-stream",
initialPrompt: "Long task",
});
await selectConversation(page, "Other conversation");
await expectWaitingTab(page, "Needs approval");
await waitForDirectoryCheckpoint(page);
await testInfo.attach("before-reload", {
body: await page.screenshot(),
contentType: "image/png",
});
await reloadWithSlowReconnect(page);
await expect(
page.getByRole("button", { name: "Other conversation", exact: true }),
).toBeVisible();
await expect.poll(syncMode).toBe("changes");
await testInfo.attach("after-reload", {
body: await page.screenshot(),
contentType: "image/png",
});
await expectWaitingTab(page, "Needs approval");
await selectConversation(page, "Needs approval");
await waitForPermissionPrompt(page);
} finally {
await waiting.cleanup();
}
});
@@ -118,8 +118,8 @@ export function ChatFind({
}, [active, isInteractive, model]);
const close = useCallback(() => {
model.close();
viewportRef.current?.focus?.();
}, [model, viewportRef]);
root.current?.focus({ preventScroll: true });
}, [model]);
let status = "";
if (state.phase === "searching") status = t("paneFind.searching");
else if (state.phase === "loading") status = t("paneFind.loading");
@@ -1162,7 +1162,6 @@ function WebStreamViewport(props: StreamRenderInput & { isMobileBreakpoint: bool
useEffect(() => {
const handle: StreamViewportHandle = {
focus: () => scrollContainerRef.current?.focus({ preventScroll: true }),
scrollToBottom: () => {
setFollowOutput(true);
cancelPendingStickToBottom();
@@ -51,8 +51,6 @@ export interface ScrollToMessageOccurrence {
}
export interface StreamViewportHandle {
/** Give keyboard navigation back to the scrolling surface on web. */
focus?: () => void;
scrollToBottom: (reason?: BottomAnchorLocalRequest["reason"]) => void;
prepareForViewportChange: () => void;
scrollToMessage?: (messageId: string, occurrence?: ScrollToMessageOccurrence) => void;
@@ -41,48 +41,6 @@ function rowWithText(
}
describe("native terminal selection", () => {
it.each([
{ startCol: 1, endCol: 8, expected: "한글 문장" },
{ startCol: 3, endCol: 8, expected: "글 문장" },
{ startCol: 1, endCol: 1, expected: "한" },
{ startCol: 0, endCol: 0, expected: "한" },
{ startCol: 0, endCol: 1, expected: "한" },
{ startCol: 4, endCol: 6, expected: " 문" },
])(
"copies complete wide characters for selected columns $startCol through $endCol",
async ({ startCol, endCol, expected }) => {
const terminal = createNativeHeadlessTerminal({ rows: 3, cols: 30 });
const copied: string[] = [];
try {
await terminal.write("한글 문장");
const bounds = terminal.getBufferBounds();
const selection = createTerminalSelectionModel();
selection.begin({
coordinate: { row: bounds.cursorRow, col: endCol },
bounds,
});
selection.update({
coordinate: { row: bounds.cursorRow, col: startCol },
bounds,
});
await copyTerminalSelection({
terminal,
selection: selection.getSnapshot().range,
clipboard: {
writeText: async (text) => {
copied.push(text);
},
},
});
expect(copied).toEqual([expected]);
} finally {
terminal.dispose();
}
},
);
it("copies Korean output without adding spaces for continuation cells", async () => {
const terminal = createNativeHeadlessTerminal({ rows: 3, cols: 30 });
const copied: string[] = [];
@@ -324,13 +324,9 @@ export function extractTerminalSelectedText(input: TerminalSelectedTextInput): s
const cells = window.rows[offset] ?? [];
const startsOnThisRow = row === selection.start.row;
const endsOnThisRow = row === selection.end.row;
let startCol = Math.max(0, startsOnThisRow ? selection.start.col : 0);
// Selecting either half highlights the whole wide character, so copy its leading cell too.
if (startCol > 0 && cells[startCol]?.width === 0) {
startCol -= 1;
}
const startCol = startsOnThisRow ? selection.start.col : 0;
const endCol = endsOnThisRow ? selection.end.col : cells.length - 1;
const selectedCells = cells.slice(startCol, Math.max(0, endCol) + 1);
const selectedCells = cells.slice(Math.max(0, startCol), Math.max(0, endCol) + 1);
const line = selectedCells
.filter((cell) => cell.width !== 0)
.map((cell) => cell.char)
-84
View File
@@ -77,90 +77,6 @@ function applySynced(agentId: string, seq: number): void {
afterEach(() => useSessionStore.getState().clearSession(SERVER_ID));
describe("viewed timeline persistence", () => {
it("keeps one complete reply when live chunks arrive before a cached chat catches up", async () => {
useSessionStore.getState().initializeSession(SERVER_ID, null);
const partial = "Paragraph 1. Paragraph 2.";
const complete = `${partial} Paragraph 3. END`;
const owner = createOwner({
readTimeline: async () => ({
agentId: AGENT_ID,
items: [{ ...item("reply-1", partial, 4), messageId: "reply-1" }],
range: { epoch: "epoch-1", startSeq: 1, endSeq: 4 },
hasOlder: false,
}),
commitTimeline: () => undefined,
});
try {
owner.replaceVisibleAgentIds("test", [AGENT_ID]);
await expect
.poll(
() =>
selectAgentTimelineState(useSessionStore.getState().sessions[SERVER_ID], AGENT_ID)
.status,
)
.toBe("painted");
owner.enqueueStreamEvent(AGENT_ID, {
event: {
type: "timeline",
provider: "codex",
item: { type: "assistant_message", text: " Paragraph 3.", messageId: "reply-1" },
} as AgentStreamEventPayload,
seq: 5,
epoch: "epoch-1",
timestamp: new Date("2026-08-26T10:00:01.000Z"),
});
owner.flushStreamAgent(AGENT_ID);
owner.applyTimelineResponse({
requestId: "resume",
agentId: AGENT_ID,
agent: null,
direction: "after",
projection: "projected",
reset: false,
epoch: "epoch-1",
window: { minSeq: 1, maxSeq: 6, nextSeq: 7 },
startCursor: { epoch: "epoch-1", seq: 5 },
endCursor: { epoch: "epoch-1", seq: 6 },
entries: [
{
provider: "codex",
item: { type: "assistant_message", text: complete, messageId: "reply-1" },
timestamp: "2026-08-26T10:00:02.000Z",
seqStart: 2,
seqEnd: 6,
sourceSeqRanges: [{ startSeq: 2, endSeq: 6 }],
collapsed: ["assistant_merge"],
},
],
error: null,
hasNewer: false,
hasOlder: false,
staleCursor: false,
gap: false,
});
const replyTexts = () => {
const session = useSessionStore.getState().sessions[SERVER_ID];
return [
...(session?.agentStreamTail.get(AGENT_ID) ?? []),
...(session?.agentStreamHead.get(AGENT_ID) ?? []),
]
.filter((entry) => entry.kind === "assistant_message")
.map((entry) => entry.text);
};
expect.soft(replyTexts()).toEqual([complete]);
owner.enqueueStreamEvent(AGENT_ID, {
event: { type: "turn_completed", provider: "codex" } as AgentStreamEventPayload,
seq: 7,
epoch: "epoch-1",
timestamp: new Date("2026-08-26T10:00:03.000Z"),
});
owner.flushStreamAgent(AGENT_ID);
expect.soft(replyTexts()).toEqual([complete]);
} finally {
owner.dispose();
}
});
it("shares an in-flight cache preparation with the viewed owner", async () => {
useSessionStore.getState().initializeSession(SERVER_ID, null);
let release!: (value: CachedTimeline) => void;
@@ -811,7 +811,7 @@ function reconcileOverlappingProjectedAssistant(params: {
};
const headMatch = findMatch(params.head);
const tailMatch = findMatch(params.tail);
const tailMatch = headMatch ? null : findMatch(params.tail);
const match = headMatch ?? tailMatch;
if (!match) {
return { tail: params.tail, head: params.head, reconciled: false };
@@ -833,20 +833,8 @@ function reconcileOverlappingProjectedAssistant(params: {
};
if (headMatch) {
const tailCursor = tailMatch?.current.timelineCursor;
const replacesCachedPrefix =
tailMatch !== null &&
tailCursor?.epoch === params.epoch &&
unit.sourceSeqRanges.some(
(range) => range.startSeq <= tailCursor.seq && tailCursor.seq <= range.endSeq,
);
return {
// Live chunks can arrive before catch-up while their cached prefix is still painted.
// The projected reply owns both copies; keeping the prefix would duplicate it and
// cause completion to discard the updated head as an already-committed item.
tail: replacesCachedPrefix
? [...params.tail.slice(0, tailMatch.index), ...params.tail.slice(tailMatch.index + 1)]
: params.tail,
tail: params.tail,
head: replaceMatch(params.head, headMatch.index),
reconciled: true,
};
@@ -122,7 +122,7 @@ async function collectTimelineAcrossTurns(
seq: rows.length + 1,
timestamp: new Date(Date.UTC(2026, 9, 4, 0, 0, rows.length)).toISOString(),
item: event.item,
...((event.turnId ?? currentTurnId) ? { turnId: event.turnId ?? currentTurnId } : {}),
...(currentTurnId ? { turnId: currentTurnId } : {}),
});
}
if (event.type === "turn_completed" && ++completedTurns === turnCount) resolve();
@@ -469,145 +469,122 @@ describe("background Claude subagents", () => {
});
});
test.each(["before", "after"])(
"keeps one logged Task card when child tools start %s the parent finishes",
async (timing) => {
// A child can stream across the main session's turn boundary. Claude opens the next
// main-session turn with system/init; bare child frames do not open a turn.
queryFactory.mockImplementation(() =>
buildQueryMock([
{ type: "system", subtype: "init", session_id: "bg-session", permissionMode: "default" },
{
type: "assistant",
message: {
model: "claude-haiku-4-5",
content: [
{
type: "tool_use",
id: "toolu_background",
name: "Agent",
input: {
subagent_type: "general-purpose",
description: "Count files here",
prompt: "Run ls and reply with the number of entries.",
run_in_background: true,
},
test("keeps one logged Task card when the child streams on after the parent's turn ends", async () => {
// A child can stream across the main session's turn boundary. Claude opens the next
// main-session turn with system/init; bare child frames do not open a turn.
queryFactory.mockImplementation(() =>
buildQueryMock([
{ type: "system", subtype: "init", session_id: "bg-session", permissionMode: "default" },
{
type: "assistant",
message: {
model: "claude-haiku-4-5",
content: [
{
type: "tool_use",
id: "toolu_background",
name: "Agent",
input: {
subagent_type: "general-purpose",
description: "Count files here",
prompt: "Run ls and reply with the number of entries.",
run_in_background: true,
},
],
},
},
],
},
{
type: "system",
subtype: "task_started",
task_id: "bg-task",
tool_use_id: "toolu_background",
task_type: "local_agent",
subagent_type: "general-purpose",
description: "Count files here",
prompt: "Run ls and reply with the number of entries.",
},
{
type: "user",
message: {
role: "user",
content: [
{
type: "tool_result",
tool_use_id: "toolu_background",
content: "Async agent launched successfully.",
},
],
},
},
...(timing === "after"
? [
{ type: "result", subtype: "success", usage: {}, total_cost_usd: 0 },
{
type: "system",
subtype: "init",
session_id: "bg-session",
permissionMode: "default",
},
]
: []),
{
type: "assistant",
parent_tool_use_id: "toolu_background",
message: {
model: "claude-haiku-4-5",
content: [
{
type: "tool_use",
id: "toolu_ls",
name: "Bash",
input: { command: "ls -1", description: "List files" },
},
],
},
},
...(timing === "before"
? [
{ type: "result", subtype: "success", usage: {}, total_cost_usd: 0 },
{
type: "system",
subtype: "init",
session_id: "bg-session",
permissionMode: "default",
},
]
: []),
{
type: "user",
parent_tool_use_id: "toolu_background",
message: {
role: "user",
content: [{ type: "tool_result", tool_use_id: "toolu_ls", content: "a\nb\nc" }],
},
},
{
type: "assistant",
parent_tool_use_id: "toolu_background",
message: { model: "claude-haiku-4-5", content: [{ type: "text", text: "3" }] },
},
{
type: "system",
subtype: "task_updated",
task_id: "bg-task",
patch: { status: "completed" },
},
{
type: "system",
subtype: "task_notification",
task_id: "bg-task",
tool_use_id: "toolu_background",
status: "completed",
},
{ type: "result", subtype: "success", usage: {}, total_cost_usd: 0 },
]),
);
const session = await new ClaudeAgentClient({
logger: createTestLogger(),
queryFactory,
resolveBinary: async () => "/test/claude/bin",
}).createSession({ provider: "claude", cwd: process.cwd() });
const rows = await collectTimelineAcrossTurns(session, "delegate work", 2);
const cards = projectTimelineRows({ rows, mode: "projected" })
.map((entry) => entry.item)
.filter((item) => item.type === "tool_call" && item.callId === "toolu_background");
expect(cards).toHaveLength(1);
expect(cards[0]).toMatchObject({
status: "completed",
detail: {
type: "sub_agent",
subAgentType: "general-purpose",
description: "Count files here",
log: "[Bash] ls -1",
},
});
},
);
{
type: "system",
subtype: "task_started",
task_id: "bg-task",
tool_use_id: "toolu_background",
task_type: "local_agent",
subagent_type: "general-purpose",
description: "Count files here",
prompt: "Run ls and reply with the number of entries.",
},
{
type: "user",
message: {
role: "user",
content: [
{
type: "tool_result",
tool_use_id: "toolu_background",
content: "Async agent launched successfully.",
},
],
},
},
{
type: "assistant",
parent_tool_use_id: "toolu_background",
message: {
model: "claude-haiku-4-5",
content: [
{
type: "tool_use",
id: "toolu_ls",
name: "Bash",
input: { command: "ls -1", description: "List files" },
},
],
},
},
{ type: "result", subtype: "success", usage: {}, total_cost_usd: 0 },
{ type: "system", subtype: "init", session_id: "bg-session", permissionMode: "default" },
{
type: "user",
parent_tool_use_id: "toolu_background",
message: {
role: "user",
content: [{ type: "tool_result", tool_use_id: "toolu_ls", content: "a\nb\nc" }],
},
},
{
type: "assistant",
parent_tool_use_id: "toolu_background",
message: { model: "claude-haiku-4-5", content: [{ type: "text", text: "3" }] },
},
{
type: "system",
subtype: "task_updated",
task_id: "bg-task",
patch: { status: "completed" },
},
{
type: "system",
subtype: "task_notification",
task_id: "bg-task",
tool_use_id: "toolu_background",
status: "completed",
},
{ type: "result", subtype: "success", usage: {}, total_cost_usd: 0 },
]),
);
const session = await new ClaudeAgentClient({
logger: createTestLogger(),
queryFactory,
resolveBinary: async () => "/test/claude/bin",
}).createSession({ provider: "claude", cwd: process.cwd() });
const rows = await collectTimelineAcrossTurns(session, "delegate work", 2);
const cards = projectTimelineRows({ rows, mode: "projected" })
.map((entry) => entry.item)
.filter((item) => item.type === "tool_call" && item.callId === "toolu_background");
expect(cards).toHaveLength(1);
expect(cards[0]).toMatchObject({
status: "completed",
detail: {
type: "sub_agent",
subAgentType: "general-purpose",
description: "Count files here",
log: "[Bash] ls -1",
},
});
});
test.each([
{
@@ -2250,7 +2250,7 @@ class ClaudeAgentSession implements AgentSession {
private readonly sidechainTracker = new ClaudeSidechainTracker({
getToolInput: (toolUseId) =>
this.toolUseCache.get(toolUseId)?.input ??
this.settledSubagentToolCalls.get(toolUseId)?.input ??
this.settledSubagentToolCallInCurrentTurn(toolUseId)?.input ??
null,
// Releases that predate the task protocol announce nothing, so the tracker keeps deriving
// identity and status from frames for them. Detecting the capability beats comparing version
@@ -2258,11 +2258,13 @@ class ClaudeAgentSession implements AgentSession {
isDescriptorOwnedElsewhere: () => this.taskProtocolSource.isActive,
// A background child streams its frames after Claude has already answered the call, so its
// card is rebuilt from the settled call: a running one would replace the settled, labeled card.
// Late card updates retain the turn that answered the call.
// A card belongs to the turn that answered the call, so frames from a later turn leave it be.
needsSyntheticParentToolCard: (toolUseId) =>
(this.toolUseCache.has(toolUseId) || this.settledSubagentToolCalls.has(toolUseId)) &&
(this.toolUseCache.has(toolUseId) ||
this.settledSubagentToolCallInCurrentTurn(toolUseId) !== undefined) &&
this.taskProtocolSource.needsSyntheticParentToolCard(toolUseId),
getSettledParentToolCall: (toolUseId) => this.settledSubagentToolCalls.get(toolUseId)?.item,
getSettledParentToolCall: (toolUseId) =>
this.settledSubagentToolCallInCurrentTurn(toolUseId)?.item,
});
private persistedHistory: PersistedTimelineEntry[] = [];
private persistedProviderSubagentEvents: Extract<
@@ -4488,12 +4490,7 @@ class ClaudeAgentSession implements AgentSession {
),
).map((event): AgentStreamEvent => ({ type: "provider_subagent", provider: "claude", event }));
const routedId = canonicalSubagentId ?? parentToolUseId;
const settled = this.settledSubagentToolCalls.get(routedId);
const sidechainEvents = this.sidechainTracker.handleMessage(message, routedId).map((event) => {
if (event.type === "timeline" && settled?.turnId) event.turnId = settled.turnId;
return event;
});
return [...runtimeEvents, ...sidechainEvents];
return [...runtimeEvents, ...this.sidechainTracker.handleMessage(message, routedId)];
}
/**
@@ -5148,8 +5145,7 @@ class ClaudeAgentSession implements AgentSession {
}
private notifySubscribers(event: AgentStreamEvent): void {
const turnId =
getAgentStreamEventTurnId(event) ?? this.activeForegroundTurnId ?? this.autonomousTurn?.id;
const turnId = this.activeForegroundTurnId ?? this.autonomousTurn?.id;
const tagged = turnId ? { ...event, turnId } : event;
this.logger.trace(
{
@@ -5625,12 +5621,17 @@ class ClaudeAgentSession implements AgentSession {
});
}
private settledSubagentToolCallInCurrentTurn(subagentId: string) {
const settled = this.settledSubagentToolCalls.get(subagentId);
const currentTurnId = this.activeForegroundTurnId ?? this.autonomousTurn?.id ?? null;
return settled && settled.turnId === currentTurnId ? settled : undefined;
}
/** A finished child streams no more frames, so its settled call has nothing left to update. */
private releaseFinishedSubagentToolCalls(observations: SubagentObservation[]): void {
for (const observation of observations) {
if (observation.kind === "status" && observation.status !== "running") {
this.settledSubagentToolCalls.delete(observation.id);
this.sidechainTracker.delete(observation.id);
}
}
}
@@ -4,40 +4,6 @@ import type { SDKMessage } from "@anthropic-ai/claude-agent-sdk";
import { ClaudeSidechainTracker } from "./sidechain-tracker.js";
describe("ClaudeSidechainTracker", () => {
it("retains a background child's actions across the parent finishing", () => {
const tracker = new ClaudeSidechainTracker({
getToolInput: () => ({ subagent_type: "general-purpose", description: "List files" }),
isDescriptorOwnedElsewhere: () => true,
});
const action = (id: string, command: string) =>
({
type: "assistant",
parent_tool_use_id: "task-1",
message: { content: [{ type: "tool_use", id, name: "Bash", input: { command } }] },
}) as unknown as SDKMessage;
tracker.handleMessage(action("ls", "ls -1"), "task-1");
expect(tracker.finishAll("completed")).toEqual([]);
const events = tracker.handleMessage(action("pwd", "pwd"), "task-1");
expect(events).toContainEqual(
expect.objectContaining({
type: "timeline",
item: expect.objectContaining({
detail: expect.objectContaining({ log: "[Bash] ls -1\n[Bash] pwd" }),
}),
}),
);
tracker.delete("task-1");
const next = tracker.handleMessage(action("next", "ls"), "task-1");
expect(next).toContainEqual(
expect.objectContaining({
type: "timeline",
item: expect.objectContaining({
detail: expect.objectContaining({ log: "[Bash] ls" }),
}),
}),
);
});
it("uses Claude's native agent name for the provider subagent title", () => {
const tracker = new ClaudeSidechainTracker({
getToolInput: () => ({
@@ -199,7 +199,7 @@ export class ClaudeSidechainTracker {
finishAll(status: "completed" | "failed" | "canceled"): AgentStreamEvent[] {
const events: AgentStreamEvent[] = [];
if (this.isDescriptorOwnedElsewhere()) {
// The task protocol owns child completion; the parent finishing does not end its children.
this.activeSidechains.clear();
return events;
}
for (const [id, state] of this.activeSidechains) {
@@ -1,105 +0,0 @@
import { execFile } from "node:child_process";
import { promisify } from "node:util";
import { fileURLToPath } from "node:url";
import { mkdtempSync, rmSync } from "node:fs";
import { tmpdir } from "node:os";
import path from "node:path";
import pino from "pino";
import { expect, test } from "vitest";
import { ClaudeAgentClient } from "../agent/providers/claude/agent.js";
import { createTestPaseoDaemon } from "../test-utils/paseo-daemon.js";
import { DaemonClient } from "../test-utils/daemon-client.js";
test("background cards retain child actions after the parent finishes (#6253)", async () => {
const cwd = mkdtempSync(path.join(tmpdir(), "paseo-background-card-"));
const logger = pino({ level: "silent" });
const daemon = await createTestPaseoDaemon({
logger,
agentClients: { claude: new ClaudeAgentClient({ logger }) },
});
const client = new DaemonClient({ url: `ws://127.0.0.1:${daemon.port}/ws` });
try {
await client.connect();
await client.fetchAgents({ subscribe: {} });
const agent = await client.createAgent({
provider: "claude",
model: "haiku",
modeId: "bypassPermissions",
cwd,
title: "Background action log reproduction",
});
await client.sendMessage(
agent.id,
'In one message, call Agent twice with run_in_background true: first with subagent_type "general-purpose", description "List files", prompt "Run ls -1, then run sleep 20, then reply with the number of entries."; second with subagent_type "general-purpose", description "Print cwd", prompt "Run pwd, then run sleep 20, then reply with the path." After launching both, reply "launched" and end your turn immediately without waiting.',
);
await client.waitForFinish(agent.id, 180_000);
const atFinish = await client.fetchAgentTimeline(agent.id, {
direction: "tail",
limit: 0,
projection: "projected",
});
await expect
.poll(
async () => {
const { subagents } = await client.listProviderSubagents(agent.id);
return subagents.length === 2 && subagents.every((child) => child.status === "completed");
},
{ timeout: 180_000, interval: 1000 },
)
.toBe(true);
const { subagents } = await client.listProviderSubagents(agent.id);
const children = await Promise.all(
subagents.map(async (child) => ({
description: child.description,
timeline: await client.fetchProviderSubagentTimeline(agent.id, child.id, {
direction: "tail",
limit: 0,
}),
})),
);
const final = await client.fetchAgentTimeline(agent.id, {
direction: "tail",
limit: 0,
projection: "projected",
});
const { stdout } = await promisify(execFile)(
process.execPath,
[
"--conditions=source",
"--import",
"tsx",
fileURLToPath(new URL("../../../../cli/src/index.ts", import.meta.url)),
"--host",
`127.0.0.1:${daemon.port}`,
"logs",
agent.id,
],
{ env: { ...process.env, PASEO_HOME: daemon.paseoHome }, timeout: 30_000 },
);
expect(stdout).toContain("[Bash] ls -1");
expect(stdout).toContain("[Bash] pwd");
expect(
atFinish.entries.filter(
(row) => row.item.type === "tool_call" && row.item.detail?.type === "sub_agent",
),
).toHaveLength(2);
const cards = final.entries
.map((row) => row.item)
.filter((item) => item.type === "tool_call" && item.detail?.type === "sub_agent");
expect(cards).toHaveLength(2);
for (const child of children) {
expect(
child.timeline.rows.some(
(row) => row.item.type === "tool_call" && row.item.name === "Bash",
),
).toBe(true);
}
for (const card of cards) {
expect(card.detail?.type === "sub_agent" ? card.detail.log : "").toContain("[Bash]");
}
} finally {
await client.close();
await daemon.close();
rmSync(cwd, { recursive: true, force: true });
}
}, 420_000);
@@ -1,5 +1,5 @@
import { createHook } from "node:async_hooks";
import { execFileSync, spawn, type ChildProcess } from "node:child_process";
import { execFileSync } from "node:child_process";
import {
mkdtempSync,
mkdirSync,
@@ -1066,16 +1066,6 @@ describe("relative typed-entry configuration", () => {
});
});
interface SearchProcessExit {
code: number | null;
signal: string | null;
}
interface SearchProcess {
child: ChildProcess;
exited: Promise<SearchProcessExit>;
}
// Reading a tree is cheap; re-deriving each entry's path state is not. A scan that asks whether
// every ancestor of every entry is inside the root, or is Git-ignored, costs multiples of the
// read and is what made large home directories exceed the client request timeout.
@@ -1083,7 +1073,6 @@ describe("home-tree scan cost", () => {
const DEPTH = 8;
const CHAINS = 40;
let scanRoot: string;
let searchProcess: SearchProcess | null = null;
beforeEach(() => {
scanRoot = realpathSync.native(mkdtempSync(path.join(tmpdir(), "directory-scan-cost-")));
@@ -1094,16 +1083,8 @@ describe("home-tree scan cost", () => {
}
});
afterEach(async () => {
try {
if (searchProcess) {
searchProcess.child.kill();
await searchProcess.exited;
}
} finally {
searchProcess = null;
rmSync(scanRoot, { recursive: true, force: true });
}
afterEach(() => {
rmSync(scanRoot, { recursive: true, force: true });
});
// Nothing matches, so the scan walks the whole tree instead of stopping at a confident result.
@@ -1121,78 +1102,6 @@ describe("home-tree scan cost", () => {
return stopPathContainmentMetrics();
}
it("finishes overlapping broad-tree searches within a bounded heap", async () => {
const broadRoot = path.join(scanRoot, "broad");
mkdirSync(broadRoot);
for (let branch = 0; branch < 30; branch += 1) {
const branchRoot = path.join(broadRoot, `branch-${branch}`);
mkdirSync(branchRoot);
for (let group = 0; group < 30; group += 1) {
const groupRoot = path.join(branchRoot, `group-${group}`);
mkdirSync(groupRoot);
for (let leaf = 0; leaf < 30; leaf += 1) {
mkdirSync(path.join(groupRoot, `leaf-${leaf}`));
}
}
}
// A smaller real tree and explicit heap budget keep the million-directory repro cheap in CI.
const moduleUrl = new URL("./directory-suggestions.ts", import.meta.url).href;
const source = `
const { searchDirectoryEntries } = await import(${JSON.stringify(moduleUrl)});
const results = await Promise.all(Array.from({ length: 14 }, (_, index) =>
searchDirectoryEntries({
root: ${JSON.stringify(broadRoot)},
query: "z".repeat(index + 1),
pathFormat: "absolute",
includeFiles: false,
includeDirectories: true,
pathQueryPolicy: "rooted",
rootAliases: ["~"],
blankQueryBehavior: "none",
confidentResultScanThreshold: 5000,
maxEntriesScanned: 2000,
limit: 30,
})
));
process.stdout.write(JSON.stringify(results));
`;
const child = spawn(
process.execPath,
[
"--max-old-space-size=128",
"--conditions=source",
"--import",
"tsx",
"--input-type=module",
"--eval",
source,
],
{ stdio: ["ignore", "pipe", "pipe"], timeout: 20_000 },
);
let stdout = "";
let stderr = "";
child.stdout.setEncoding("utf8");
child.stderr.setEncoding("utf8");
child.stdout.on("data", (chunk: string) => {
stdout += chunk;
});
child.stderr.on("data", (chunk: string) => {
stderr += chunk;
});
const exited = new Promise<SearchProcessExit>((resolve, reject) => {
child.on("error", reject);
child.on("close", (code, signal) => resolve({ code, signal }));
});
searchProcess = { child, exited };
const exit = await exited;
expect({ ...exit, outOfMemory: stderr.includes("heap out of memory") }).toEqual({
code: 0,
signal: null,
outOfMemory: false,
});
expect(JSON.parse(stdout)).toEqual(Array.from({ length: 14 }, () => []));
});
it("derives no containment for a tree that cannot leave the root", async () => {
const checks = await countContainmentChecks(scanWholeTree);
@@ -63,13 +63,6 @@ interface WalkStart {
resolvedPath: string;
}
interface WalkChildrenOptions {
parent: WalkStart;
depth: number;
input: SearchInput;
visited: Set<string>;
}
interface TraversedEntry extends ChildEntry {
visiblePath: string;
depth: number;
@@ -325,18 +318,17 @@ async function searchChildren(input: SearchInput): Promise<RankedEntry[]> {
const discoveryInput = isInsideExcludedPath(parent, input)
? { ...input, excludedDiscoveryPaths: new Set<string>() }
: input;
const ranked: RankedEntry[] = [];
for await (const entry of readChildren(parent)) {
const entries = await readChildren(parent);
return entries.flatMap((entry) => {
if (!isPathInsideRoot(input.root, entry.resolvedPath) || !shouldDiscover(entry, discoveryInput))
continue;
return [];
const candidate: TraversedEntry = {
...entry,
visiblePath: path.join(visibleParent, entry.name),
depth: 1,
};
if (shouldSuggest(candidate, input)) ranked.push(rank(candidate, input));
}
return ranked;
return shouldSuggest(candidate, input) ? [rank(candidate, input)] : [];
});
}
async function searchTree(
@@ -344,8 +336,21 @@ async function searchTree(
start: WalkStart = { visiblePath: input.root, resolvedPath: input.root },
): Promise<RankedEntry[]> {
if (!(input.maxEntriesScanned > 0)) return [];
const roots = (await readChildren(start.resolvedPath)).filter((entry) =>
staysInsideRoot(entry, input.root),
);
const visited = new Set<string>([start.resolvedPath]);
const branches = walkChildren({ parent: start, depth: 1, input, visited });
const branches = roots.flatMap((entry) =>
shouldDiscover(entry, input)
? [
walkBranch(
{ ...entry, visiblePath: path.join(start.visiblePath, entry.name), depth: 1 },
input,
visited,
),
]
: [],
);
const ranked: RankedEntry[] = [];
let scanned = 0;
const threshold = input.confidentResultScanThreshold;
@@ -376,39 +381,38 @@ async function* walkBranch(
)
return;
visited.add(entry.resolvedPath);
yield* roundRobin(walkChildren({ parent: entry, depth: entry.depth + 1, input, visited }));
const children = (await readChildren(entry.resolvedPath)).filter((child) =>
staysInsideRoot(child, input.root),
);
const branches = children.flatMap((child) =>
shouldDiscover(child, input)
? [
walkBranch(
{
...child,
visiblePath: path.join(entry.visiblePath, child.name),
depth: entry.depth + 1,
},
input,
visited,
),
]
: [],
);
yield* roundRobin(branches);
}
// Create a child's path and iterator only when its branch gets a turn. Eagerly constructing every
// sibling's subtree retained hundreds of thousands of unvisited children beyond the scan budget.
async function* walkChildren({
parent,
depth,
input,
visited,
}: WalkChildrenOptions): AsyncGenerator<AsyncGenerator<TraversedEntry>> {
for await (const child of readChildren(parent.resolvedPath)) {
if (!staysInsideRoot(child, input.root) || !shouldDiscover(child, input)) continue;
yield walkBranch(
{ ...child, visiblePath: path.join(parent.visiblePath, child.name), depth },
input,
visited,
);
}
}
async function* roundRobin<T>(branches: AsyncIterable<AsyncGenerator<T>>): AsyncGenerator<T> {
let active: AsyncIterable<AsyncGenerator<T>> | Array<AsyncGenerator<T>> = branches;
while (true) {
async function* roundRobin<T>(branches: Array<AsyncGenerator<T>>): AsyncGenerator<T> {
let active = branches;
while (active.length) {
const nextRound: Array<AsyncGenerator<T>> = [];
for await (const branch of active) {
for (const branch of active) {
const next = await branch.next();
if (!next.done) {
nextRound.push(branch);
yield next.value;
}
}
if (!nextRound.length) return;
active = nextRound;
}
}
@@ -738,9 +742,9 @@ async function resolveDirectory(inputPath: string): Promise<string | null> {
return info?.isDirectory() ? resolved : null;
}
async function* readChildren(directory: string): AsyncGenerator<ChildEntry> {
async function readChildren(directory: string): Promise<ChildEntry[]> {
const directoryInfo = await stat(directory).catch(() => null);
if (!directoryInfo?.isDirectory()) return;
if (!directoryInfo?.isDirectory()) return [];
const cached = CAN_VALIDATE_DIRECTORY_CACHE_FROM_METADATA
? directoryListCache.get(directory)
@@ -770,10 +774,9 @@ async function* readChildren(directory: string): AsyncGenerator<ChildEntry> {
}
}
for (const rawEntry of rawEntries) {
const entry = await resolveChild(directory, rawEntry);
if (entry) yield entry;
}
return (await Promise.all(rawEntries.map((entry) => resolveChild(directory, entry))))
.filter((entry): entry is ChildEntry => entry !== null)
.sort((left, right) => left.name.localeCompare(right.name));
}
async function loadGitIgnoredPaths(root: string): Promise<Set<string>> {
+3
View File
@@ -19,6 +19,9 @@ export default defineConfig({
setupFiles: [path.resolve(__dirname, "./src/test-utils/vitest-setup.ts")],
pool: "forks",
fileParallelism: false,
// Even two concurrent Git-heavy suites can exhaust Windows test deadlines.
// Run files serially so their subprocesses do not compete across suites.
maxWorkers: process.platform === "win32" ? 1 : undefined,
exclude: ["**/node_modules/**", "**/dist/**", "**/.claude/**", "**/.dev/**"],
},
});