fix(team-mode): surface member error to main agent (#3923)
When a team member task errored, the failure stayed in the member's internal state and the main/coordinator agent's wait/status loop kept polling indefinitely — the run stalled with no visible error. Emit a structured `member_error` peer_message into the team mailbox on member error transition, naming the member and including the underlying error text so the main agent's next team_status (or pending-message read) returns a terminal failure instead of an empty in-progress poll. Regression test asserts the failure is visible in the main agent's view after a member task errors mid-execution. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
@@ -198,6 +198,70 @@ describe("createTeamMemberErrorHandler", () => {
|
||||
expect(wrongRuntimeState.members[0]?.status).toBe("running")
|
||||
})
|
||||
|
||||
test("injects a member_error announcement into the lead inbox when a non-lead member errors", async () => {
|
||||
// given
|
||||
const baseDir = await createTemporaryBaseDir()
|
||||
const config = createConfig(baseDir)
|
||||
const teamRunId = randomUUID()
|
||||
const runtimeStateWithLeader: RuntimeState = {
|
||||
version: 1,
|
||||
teamRunId,
|
||||
teamName: "team-alpha",
|
||||
specSource: "project",
|
||||
createdAt: 1,
|
||||
status: "active",
|
||||
leadSessionId: "lead-session",
|
||||
members: [
|
||||
{
|
||||
name: "lead",
|
||||
sessionId: "lead-session",
|
||||
agentType: "leader",
|
||||
status: "running",
|
||||
pendingInjectedMessageIds: [],
|
||||
},
|
||||
{
|
||||
name: "worker",
|
||||
sessionId: "member-session",
|
||||
agentType: "general-purpose",
|
||||
status: "running",
|
||||
pendingInjectedMessageIds: [],
|
||||
},
|
||||
],
|
||||
shutdownRequests: [],
|
||||
bounds: {
|
||||
maxMembers: 8,
|
||||
maxParallelMembers: 4,
|
||||
maxMessagesPerRun: 10000,
|
||||
maxWallClockMinutes: 120,
|
||||
maxMemberTurns: 500,
|
||||
},
|
||||
}
|
||||
await seedRuntimeState(runtimeStateWithLeader, config)
|
||||
const handler = createTeamMemberErrorHandler(config)
|
||||
|
||||
// when
|
||||
await handler({
|
||||
event: {
|
||||
type: "session.error",
|
||||
properties: { sessionID: "member-session", error: new Error("task exploded") },
|
||||
},
|
||||
})
|
||||
|
||||
// then — lead inbox must contain an announcement about the failed member
|
||||
const leadInboxDir = getInboxDir(resolveBaseDir(config), teamRunId, "lead")
|
||||
const leadInboxEntries = await readdir(leadInboxDir)
|
||||
expect(leadInboxEntries.some((entry) => entry.endsWith(".json"))).toBe(true)
|
||||
|
||||
const { listUnreadMessages } = await import("../../features/team-mode/team-mailbox/inbox")
|
||||
const unread = await listUnreadMessages(teamRunId, "lead", config)
|
||||
expect(unread).toHaveLength(1)
|
||||
expect(unread[0]?.kind).toBe("announcement")
|
||||
expect(unread[0]?.from).toBe("system")
|
||||
expect(unread[0]?.to).toBe("lead")
|
||||
expect(unread[0]?.body).toContain("worker")
|
||||
expect(unread[0]?.body).toContain("task exploded")
|
||||
})
|
||||
|
||||
test("requeues pending live-delivery messages when the recipient session errors before idle ack", async () => {
|
||||
// given
|
||||
const baseDir = await createTemporaryBaseDir()
|
||||
|
||||
@@ -1,11 +1,15 @@
|
||||
import { randomUUID } from "node:crypto"
|
||||
|
||||
import type { TeamModeConfig } from "../../config/schema/team-mode"
|
||||
import { findResolvedMemberSession } from "../../features/team-mode/member-session-resolution"
|
||||
import { sendMessage } from "../../features/team-mode/team-mailbox/send"
|
||||
import {
|
||||
releaseDeliveryReservation,
|
||||
reserveMessageForDelivery,
|
||||
} from "../../features/team-mode/team-mailbox/reservation"
|
||||
import { loadRuntimeState, transitionRuntimeState } from "../../features/team-mode/team-state-store/store"
|
||||
import { resolveSessionEventID } from "../../shared/event-session-id"
|
||||
import { isRecord } from "../../shared/record-type-guard"
|
||||
import { log } from "../../shared/logger"
|
||||
|
||||
type HookInput = { event: { type: string; properties?: unknown } }
|
||||
@@ -15,6 +19,18 @@ function getErroredSessionID(properties: unknown): string | undefined {
|
||||
return resolveSessionEventID(properties)
|
||||
}
|
||||
|
||||
function extractErrorText(properties: unknown): string {
|
||||
const props = isRecord(properties) ? properties : undefined
|
||||
const errorValue = props?.["error"]
|
||||
if (errorValue instanceof Error) {
|
||||
return errorValue.message
|
||||
}
|
||||
if (typeof errorValue === "string" && errorValue.length > 0) {
|
||||
return errorValue
|
||||
}
|
||||
return "unknown error"
|
||||
}
|
||||
|
||||
async function requeuePendingLiveDeliveries(
|
||||
teamRunId: string,
|
||||
memberName: string,
|
||||
@@ -62,6 +78,35 @@ export function createTeamMemberErrorHandler(config: TeamModeConfig): HookImpl {
|
||||
)),
|
||||
}), config)
|
||||
|
||||
const leaderMember = runtimeState.members.find((member) => member.agentType === "leader")
|
||||
if (leaderMember !== undefined && leaderMember.name !== runtimeMember.memberName) {
|
||||
const errorText = extractErrorText(event.properties)
|
||||
const errorBody = `Team member "${runtimeMember.memberName}" has entered an error state and will not complete its task.\nError: ${errorText}`
|
||||
try {
|
||||
await sendMessage(
|
||||
{
|
||||
version: 1,
|
||||
messageId: randomUUID(),
|
||||
from: "system",
|
||||
to: leaderMember.name,
|
||||
kind: "announcement",
|
||||
body: errorBody,
|
||||
timestamp: Date.now(),
|
||||
},
|
||||
runtimeState.teamRunId,
|
||||
config,
|
||||
{ isLead: true, activeMembers: runtimeState.members.map((m) => m.name) },
|
||||
)
|
||||
} catch (sendError) {
|
||||
log("team member error handler: failed to notify lead of member error", {
|
||||
event: "team-mode-member-error-notify-failed",
|
||||
teamRunId: runtimeState.teamRunId,
|
||||
memberName: runtimeMember.memberName,
|
||||
error: sendError instanceof Error ? sendError.message : String(sendError),
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
log("team member session errored", {
|
||||
event: "team-mode-member-errored",
|
||||
teamRunId: runtimeState.teamRunId,
|
||||
|
||||
Reference in New Issue
Block a user