fix(unstable-agent-babysitter): respect active sessions
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
This commit is contained in:
@@ -2,6 +2,7 @@ import { afterEach, describe, expect, test } from "bun:test"
|
||||
import { _resetForTesting, setMainSession } from "../../features/claude-code-session-state"
|
||||
import type { BackgroundTask } from "../../features/background-agent"
|
||||
import { OMO_INTERNAL_INITIATOR_MARKER } from "../../shared/internal-initiator-marker"
|
||||
import { releaseAllPromptAsyncReservationsForTesting } from "../shared/prompt-async-gate"
|
||||
import { createUnstableAgentBabysitterHook } from "./index"
|
||||
|
||||
const projectDir = process.cwd()
|
||||
@@ -49,7 +50,7 @@ function createTask(overrides: Partial<BackgroundTask> = {}): BackgroundTask {
|
||||
status: "running",
|
||||
progress: {
|
||||
toolCalls: 1,
|
||||
lastUpdate: new Date(),
|
||||
lastUpdate: new Date(Date.now() - 121000),
|
||||
lastMessage: "still working",
|
||||
lastMessageAt: new Date(Date.now() - 121000),
|
||||
},
|
||||
@@ -61,6 +62,7 @@ function createTask(overrides: Partial<BackgroundTask> = {}): BackgroundTask {
|
||||
describe("unstable-agent-babysitter hook", () => {
|
||||
afterEach(() => {
|
||||
_resetForTesting()
|
||||
releaseAllPromptAsyncReservationsForTesting()
|
||||
})
|
||||
|
||||
test("settles idle before injecting a reminder", async () => {
|
||||
@@ -290,4 +292,281 @@ describe("unstable-agent-babysitter hook", () => {
|
||||
expect(payload.body?.model).toEqual({ providerID: "openai", modelID: "gpt-4" })
|
||||
expect(payload.body?.variant).toBe("max")
|
||||
})
|
||||
|
||||
test("#given the main session has a fresh user message #when it becomes idle #then babysitter does not inject a reminder", async () => {
|
||||
// given
|
||||
const originalNow = Date.now
|
||||
Date.now = () => 10 * 60 * 1000
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: {
|
||||
"main-1": [
|
||||
{ info: { role: "user", time: { created: Date.now() - 1_500 } } },
|
||||
],
|
||||
"bg-1": [
|
||||
{ info: { role: "assistant" }, parts: [{ type: "thinking", thinking: "deep thought" }] },
|
||||
],
|
||||
},
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({
|
||||
progress: {
|
||||
toolCalls: 1,
|
||||
lastUpdate: new Date(0),
|
||||
lastMessage: "still working",
|
||||
lastMessageAt: new Date(0),
|
||||
},
|
||||
})])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
try {
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
expect(promptCalls.length).toBe(0)
|
||||
} finally {
|
||||
Date.now = originalNow
|
||||
}
|
||||
})
|
||||
|
||||
test("#given the latest main-session message is assistant output after a fresh user message #when it becomes idle #then babysitter may inject a reminder", async () => {
|
||||
// given
|
||||
const originalNow = Date.now
|
||||
Date.now = () => 10 * 60 * 1000
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: {
|
||||
"main-1": [
|
||||
{ info: { role: "user", time: { created: Date.now() - 1_500 } } },
|
||||
{ info: { role: "assistant", time: { created: Date.now() - 500 }, agent: "sisyphus" } },
|
||||
],
|
||||
"bg-1": [
|
||||
{ info: { role: "assistant" }, parts: [{ type: "thinking", thinking: "deep thought" }] },
|
||||
],
|
||||
},
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({
|
||||
progress: {
|
||||
toolCalls: 1,
|
||||
lastUpdate: new Date(0),
|
||||
lastMessage: "still working",
|
||||
lastMessageAt: new Date(0),
|
||||
},
|
||||
})])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
try {
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
expect(promptCalls.length).toBe(1)
|
||||
} finally {
|
||||
Date.now = originalNow
|
||||
}
|
||||
})
|
||||
|
||||
test("#given an unstable task has a fresh progress update after its last message #when the main session idles #then babysitter does not treat it as hung", async () => {
|
||||
// given
|
||||
const originalNow = Date.now
|
||||
Date.now = () => 10 * 60 * 1000
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: { "main-1": [], "bg-1": [] },
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({
|
||||
progress: {
|
||||
toolCalls: 2,
|
||||
lastUpdate: new Date(Date.now() - 1_000),
|
||||
lastMessage: "still working",
|
||||
lastMessageAt: new Date(Date.now() - 5 * 60 * 1000),
|
||||
},
|
||||
})])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
try {
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
expect(promptCalls.length).toBe(0)
|
||||
} finally {
|
||||
Date.now = originalNow
|
||||
}
|
||||
})
|
||||
|
||||
test("#given an unstable task has stale progress and stale last message #when the main session idles #then babysitter still reminds", async () => {
|
||||
// given
|
||||
const originalNow = Date.now
|
||||
Date.now = () => 10 * 60 * 1000
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: { "main-1": [], "bg-1": [] },
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({
|
||||
progress: {
|
||||
toolCalls: 2,
|
||||
lastUpdate: new Date(Date.now() - 5 * 60 * 1000),
|
||||
lastMessage: "still working",
|
||||
lastMessageAt: new Date(Date.now() - 5 * 60 * 1000),
|
||||
},
|
||||
})])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
try {
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
expect(promptCalls.length).toBe(1)
|
||||
} finally {
|
||||
Date.now = originalNow
|
||||
}
|
||||
})
|
||||
|
||||
test("#given a reminder was already sent before an abort #when a user message resumes the session within cooldown #then no duplicate reminder is injected", async () => {
|
||||
// given
|
||||
const originalNow = Date.now
|
||||
let currentNow = 10 * 60 * 1000
|
||||
Date.now = () => currentNow
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: { "main-1": [], "bg-1": [] },
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({
|
||||
progress: {
|
||||
toolCalls: 1,
|
||||
lastUpdate: new Date(0),
|
||||
lastMessage: "still working",
|
||||
lastMessageAt: new Date(0),
|
||||
},
|
||||
})])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
try {
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
await hook.event({ event: { type: "session.error", properties: { sessionID: "main-1", error: { name: "AbortError" } } } })
|
||||
currentNow += 1_000
|
||||
await hook.event({ event: { type: "message.updated", properties: { sessionID: "main-1", info: { role: "user" } } } })
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
expect(promptCalls.length).toBe(1)
|
||||
} finally {
|
||||
Date.now = originalNow
|
||||
}
|
||||
})
|
||||
|
||||
test("#given a reminder was already sent before a stop event #when assistant activity resumes within cooldown #then no duplicate reminder is injected", async () => {
|
||||
// given
|
||||
const originalNow = Date.now
|
||||
let currentNow = 10 * 60 * 1000
|
||||
Date.now = () => currentNow
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: { "main-1": [], "bg-1": [] },
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({
|
||||
progress: {
|
||||
toolCalls: 1,
|
||||
lastUpdate: new Date(0),
|
||||
lastMessage: "still working",
|
||||
lastMessageAt: new Date(0),
|
||||
},
|
||||
})])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
try {
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
await hook.event({ event: { type: "session.stop", properties: { sessionID: "main-1" } } })
|
||||
currentNow += 1_000
|
||||
await hook.event({ event: { type: "message.updated", properties: { sessionID: "main-1", info: { role: "assistant" } } } })
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
expect(promptCalls.length).toBe(1)
|
||||
} finally {
|
||||
Date.now = originalNow
|
||||
}
|
||||
})
|
||||
|
||||
test("#given unstable task agent is a config key #when babysitter builds a reminder #then the reminder uses the canonical display name", async () => {
|
||||
// given
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: { "main-1": [], "bg-1": [] },
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({ agent: "sisyphus" })])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
const payload = promptCalls[0]?.input as { body?: { parts?: Array<{ text?: string }> } } | undefined
|
||||
const text = payload?.body?.parts?.[0]?.text ?? ""
|
||||
expect(text).toContain("Agent: Sisyphus - Ultraworker")
|
||||
expect(text).not.toContain("Agent: sisyphus")
|
||||
})
|
||||
|
||||
test("#given unstable task agent is a legacy display name #when babysitter builds a reminder #then the reminder uses the current display name", async () => {
|
||||
// given
|
||||
setMainSession("main-1")
|
||||
const promptCalls: Array<{ input: unknown }> = []
|
||||
const ctx = createMockPluginInput({
|
||||
messagesBySession: { "main-1": [], "bg-1": [] },
|
||||
promptCalls,
|
||||
})
|
||||
const backgroundManager = createBackgroundManager([createTask({ agent: "Sisyphus (Ultraworker)" })])
|
||||
const hook = createUnstableAgentBabysitterHook(ctx, {
|
||||
backgroundManager,
|
||||
config: { timeout_ms: 120000 },
|
||||
})
|
||||
|
||||
// when
|
||||
await hook.event({ event: { type: "session.idle", properties: { sessionID: "main-1" } } })
|
||||
|
||||
// then
|
||||
const payload = promptCalls[0]?.input as { body?: { parts?: Array<{ text?: string }> } } | undefined
|
||||
const text = payload?.body?.parts?.[0]?.text ?? ""
|
||||
expect(text).toContain("Agent: Sisyphus - Ultraworker")
|
||||
expect(text).not.toContain("Agent: Sisyphus (Ultraworker)")
|
||||
})
|
||||
})
|
||||
|
||||
Reference in New Issue
Block a user