fix(todo-continuation): remove activity-based stagnation bypass
This commit is contained in:
@@ -18,7 +18,7 @@ describe("createSessionStateStore regressions", () => {
|
|||||||
|
|
||||||
describe("#given external activity happens after a successful continuation", () => {
|
describe("#given external activity happens after a successful continuation", () => {
|
||||||
describe("#when todos stay unchanged", () => {
|
describe("#when todos stay unchanged", () => {
|
||||||
test("#then it treats the activity as progress instead of stagnation", () => {
|
test("#then it keeps counting stagnation", () => {
|
||||||
const sessionID = "ses-activity-progress"
|
const sessionID = "ses-activity-progress"
|
||||||
const todos = [
|
const todos = [
|
||||||
{ id: "1", content: "Task 1", status: "pending", priority: "high" },
|
{ id: "1", content: "Task 1", status: "pending", priority: "high" },
|
||||||
@@ -37,9 +37,9 @@ describe("createSessionStateStore regressions", () => {
|
|||||||
trackedState.abortDetectedAt = undefined
|
trackedState.abortDetectedAt = undefined
|
||||||
const progressUpdate = sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
const progressUpdate = sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
||||||
|
|
||||||
expect(progressUpdate.hasProgressed).toBe(true)
|
expect(progressUpdate.hasProgressed).toBe(false)
|
||||||
expect(progressUpdate.progressSource).toBe("activity")
|
expect(progressUpdate.progressSource).toBe("none")
|
||||||
expect(progressUpdate.stagnationCount).toBe(0)
|
expect(progressUpdate.stagnationCount).toBe(1)
|
||||||
})
|
})
|
||||||
})
|
})
|
||||||
})
|
})
|
||||||
@@ -72,7 +72,7 @@ describe("createSessionStateStore regressions", () => {
|
|||||||
|
|
||||||
describe("#given stagnation already halted a session", () => {
|
describe("#given stagnation already halted a session", () => {
|
||||||
describe("#when new activity appears before the next idle check", () => {
|
describe("#when new activity appears before the next idle check", () => {
|
||||||
test("#then it resets the stop condition on the next progress check", () => {
|
test("#then it does not reset the stop condition", () => {
|
||||||
const sessionID = "ses-stagnation-recovery"
|
const sessionID = "ses-stagnation-recovery"
|
||||||
const todos = [
|
const todos = [
|
||||||
{ id: "1", content: "Task 1", status: "pending", priority: "high" },
|
{ id: "1", content: "Task 1", status: "pending", priority: "high" },
|
||||||
@@ -96,9 +96,9 @@ describe("createSessionStateStore regressions", () => {
|
|||||||
const progressUpdate = sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
const progressUpdate = sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
||||||
|
|
||||||
expect(progressUpdate.previousStagnationCount).toBe(MAX_STAGNATION_COUNT)
|
expect(progressUpdate.previousStagnationCount).toBe(MAX_STAGNATION_COUNT)
|
||||||
expect(progressUpdate.hasProgressed).toBe(true)
|
expect(progressUpdate.hasProgressed).toBe(false)
|
||||||
expect(progressUpdate.progressSource).toBe("activity")
|
expect(progressUpdate.progressSource).toBe("none")
|
||||||
expect(progressUpdate.stagnationCount).toBe(0)
|
expect(progressUpdate.stagnationCount).toBe(MAX_STAGNATION_COUNT)
|
||||||
})
|
})
|
||||||
})
|
})
|
||||||
})
|
})
|
||||||
|
|||||||
@@ -16,8 +16,6 @@ interface TrackedSessionState {
|
|||||||
lastAccessedAt: number
|
lastAccessedAt: number
|
||||||
lastCompletedCount?: number
|
lastCompletedCount?: number
|
||||||
lastTodoSnapshot?: string
|
lastTodoSnapshot?: string
|
||||||
activitySignalCount: number
|
|
||||||
lastObservedActivitySignalCount?: number
|
|
||||||
}
|
}
|
||||||
|
|
||||||
export interface ContinuationProgressUpdate {
|
export interface ContinuationProgressUpdate {
|
||||||
@@ -25,7 +23,7 @@ export interface ContinuationProgressUpdate {
|
|||||||
previousStagnationCount: number
|
previousStagnationCount: number
|
||||||
stagnationCount: number
|
stagnationCount: number
|
||||||
hasProgressed: boolean
|
hasProgressed: boolean
|
||||||
progressSource: "none" | "todo" | "activity"
|
progressSource: "none" | "todo"
|
||||||
}
|
}
|
||||||
|
|
||||||
export interface SessionStateStore {
|
export interface SessionStateStore {
|
||||||
@@ -98,17 +96,7 @@ export function createSessionStateStore(): SessionStateStore {
|
|||||||
const trackedSession: TrackedSessionState = {
|
const trackedSession: TrackedSessionState = {
|
||||||
state: rawState,
|
state: rawState,
|
||||||
lastAccessedAt: Date.now(),
|
lastAccessedAt: Date.now(),
|
||||||
activitySignalCount: 0,
|
|
||||||
}
|
}
|
||||||
trackedSession.state = new Proxy(rawState, {
|
|
||||||
set(target, property, value, receiver) {
|
|
||||||
if (property === "abortDetectedAt" && value === undefined) {
|
|
||||||
trackedSession.activitySignalCount += 1
|
|
||||||
}
|
|
||||||
|
|
||||||
return Reflect.set(target, property, value, receiver)
|
|
||||||
},
|
|
||||||
})
|
|
||||||
sessions.set(sessionID, trackedSession)
|
sessions.set(sessionID, trackedSession)
|
||||||
return trackedSession
|
return trackedSession
|
||||||
}
|
}
|
||||||
@@ -137,7 +125,6 @@ export function createSessionStateStore(): SessionStateStore {
|
|||||||
const previousStagnationCount = state.stagnationCount
|
const previousStagnationCount = state.stagnationCount
|
||||||
const currentCompletedCount = todos?.filter((todo) => todo.status === "completed").length
|
const currentCompletedCount = todos?.filter((todo) => todo.status === "completed").length
|
||||||
const currentTodoSnapshot = todos ? getTodoSnapshot(todos) : undefined
|
const currentTodoSnapshot = todos ? getTodoSnapshot(todos) : undefined
|
||||||
const currentActivitySignalCount = trackedSession.activitySignalCount
|
|
||||||
const hasCompletedMoreTodos =
|
const hasCompletedMoreTodos =
|
||||||
currentCompletedCount !== undefined
|
currentCompletedCount !== undefined
|
||||||
&& trackedSession.lastCompletedCount !== undefined
|
&& trackedSession.lastCompletedCount !== undefined
|
||||||
@@ -146,9 +133,6 @@ export function createSessionStateStore(): SessionStateStore {
|
|||||||
currentTodoSnapshot !== undefined
|
currentTodoSnapshot !== undefined
|
||||||
&& trackedSession.lastTodoSnapshot !== undefined
|
&& trackedSession.lastTodoSnapshot !== undefined
|
||||||
&& currentTodoSnapshot !== trackedSession.lastTodoSnapshot
|
&& currentTodoSnapshot !== trackedSession.lastTodoSnapshot
|
||||||
const hasObservedExternalActivity =
|
|
||||||
trackedSession.lastObservedActivitySignalCount !== undefined
|
|
||||||
&& currentActivitySignalCount > trackedSession.lastObservedActivitySignalCount
|
|
||||||
const hadSuccessfulInjectionAwaitingProgressCheck = state.awaitingPostInjectionProgressCheck === true
|
const hadSuccessfulInjectionAwaitingProgressCheck = state.awaitingPostInjectionProgressCheck === true
|
||||||
|
|
||||||
state.lastIncompleteCount = incompleteCount
|
state.lastIncompleteCount = incompleteCount
|
||||||
@@ -158,7 +142,6 @@ export function createSessionStateStore(): SessionStateStore {
|
|||||||
if (currentTodoSnapshot !== undefined) {
|
if (currentTodoSnapshot !== undefined) {
|
||||||
trackedSession.lastTodoSnapshot = currentTodoSnapshot
|
trackedSession.lastTodoSnapshot = currentTodoSnapshot
|
||||||
}
|
}
|
||||||
trackedSession.lastObservedActivitySignalCount = currentActivitySignalCount
|
|
||||||
|
|
||||||
if (previousIncompleteCount === undefined) {
|
if (previousIncompleteCount === undefined) {
|
||||||
state.stagnationCount = 0
|
state.stagnationCount = 0
|
||||||
@@ -173,9 +156,7 @@ export function createSessionStateStore(): SessionStateStore {
|
|||||||
|
|
||||||
const progressSource = incompleteCount < previousIncompleteCount || hasCompletedMoreTodos || hasTodoSnapshotChanged
|
const progressSource = incompleteCount < previousIncompleteCount || hasCompletedMoreTodos || hasTodoSnapshotChanged
|
||||||
? "todo"
|
? "todo"
|
||||||
: hasObservedExternalActivity
|
: "none"
|
||||||
? "activity"
|
|
||||||
: "none"
|
|
||||||
|
|
||||||
if (progressSource !== "none") {
|
if (progressSource !== "none") {
|
||||||
state.stagnationCount = 0
|
state.stagnationCount = 0
|
||||||
@@ -223,8 +204,6 @@ export function createSessionStateStore(): SessionStateStore {
|
|||||||
state.awaitingPostInjectionProgressCheck = false
|
state.awaitingPostInjectionProgressCheck = false
|
||||||
trackedSession.lastCompletedCount = undefined
|
trackedSession.lastCompletedCount = undefined
|
||||||
trackedSession.lastTodoSnapshot = undefined
|
trackedSession.lastTodoSnapshot = undefined
|
||||||
trackedSession.activitySignalCount = 0
|
|
||||||
trackedSession.lastObservedActivitySignalCount = undefined
|
|
||||||
}
|
}
|
||||||
|
|
||||||
function cancelCountdown(sessionID: string): void {
|
function cancelCountdown(sessionID: string): void {
|
||||||
|
|||||||
@@ -3,6 +3,8 @@
|
|||||||
import { describe, expect, it as test } from "bun:test"
|
import { describe, expect, it as test } from "bun:test"
|
||||||
|
|
||||||
import { MAX_STAGNATION_COUNT } from "./constants"
|
import { MAX_STAGNATION_COUNT } from "./constants"
|
||||||
|
import { handleNonIdleEvent } from "./non-idle-events"
|
||||||
|
import { createSessionStateStore } from "./session-state"
|
||||||
import { shouldStopForStagnation } from "./stagnation-detection"
|
import { shouldStopForStagnation } from "./stagnation-detection"
|
||||||
|
|
||||||
describe("shouldStopForStagnation", () => {
|
describe("shouldStopForStagnation", () => {
|
||||||
@@ -25,7 +27,7 @@ describe("shouldStopForStagnation", () => {
|
|||||||
})
|
})
|
||||||
})
|
})
|
||||||
|
|
||||||
describe("#when activity progress is detected after the halt", () => {
|
describe("#when todo progress is detected after the halt", () => {
|
||||||
test("#then it clears the stop condition", () => {
|
test("#then it clears the stop condition", () => {
|
||||||
const shouldStop = shouldStopForStagnation({
|
const shouldStop = shouldStopForStagnation({
|
||||||
sessionID: "ses-recovered",
|
sessionID: "ses-recovered",
|
||||||
@@ -35,7 +37,7 @@ describe("shouldStopForStagnation", () => {
|
|||||||
previousStagnationCount: MAX_STAGNATION_COUNT,
|
previousStagnationCount: MAX_STAGNATION_COUNT,
|
||||||
stagnationCount: 0,
|
stagnationCount: 0,
|
||||||
hasProgressed: true,
|
hasProgressed: true,
|
||||||
progressSource: "activity",
|
progressSource: "todo",
|
||||||
},
|
},
|
||||||
})
|
})
|
||||||
|
|
||||||
@@ -43,4 +45,60 @@ describe("shouldStopForStagnation", () => {
|
|||||||
})
|
})
|
||||||
})
|
})
|
||||||
})
|
})
|
||||||
|
|
||||||
|
describe("#given only non-idle tool and message events happen between idle checks", () => {
|
||||||
|
describe("#when todo state does not change across three idle cycles", () => {
|
||||||
|
test("#then stagnation count reaches three", () => {
|
||||||
|
// given
|
||||||
|
const sessionStateStore = createSessionStateStore()
|
||||||
|
const sessionID = "ses-non-idle-activity-without-progress"
|
||||||
|
const state = sessionStateStore.getState(sessionID)
|
||||||
|
const todos = [
|
||||||
|
{ id: "1", content: "Task 1", status: "pending", priority: "high" },
|
||||||
|
{ id: "2", content: "Task 2", status: "pending", priority: "medium" },
|
||||||
|
]
|
||||||
|
|
||||||
|
sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
||||||
|
|
||||||
|
// when
|
||||||
|
state.awaitingPostInjectionProgressCheck = true
|
||||||
|
const firstCycle = sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
||||||
|
|
||||||
|
handleNonIdleEvent({
|
||||||
|
eventType: "tool.execute.before",
|
||||||
|
properties: { sessionID },
|
||||||
|
sessionStateStore,
|
||||||
|
})
|
||||||
|
handleNonIdleEvent({
|
||||||
|
eventType: "message.updated",
|
||||||
|
properties: { info: { sessionID, role: "assistant" } },
|
||||||
|
sessionStateStore,
|
||||||
|
})
|
||||||
|
|
||||||
|
state.awaitingPostInjectionProgressCheck = true
|
||||||
|
const secondCycle = sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
||||||
|
|
||||||
|
handleNonIdleEvent({
|
||||||
|
eventType: "tool.execute.after",
|
||||||
|
properties: { sessionID },
|
||||||
|
sessionStateStore,
|
||||||
|
})
|
||||||
|
handleNonIdleEvent({
|
||||||
|
eventType: "message.part.updated",
|
||||||
|
properties: { info: { sessionID, role: "assistant" } },
|
||||||
|
sessionStateStore,
|
||||||
|
})
|
||||||
|
|
||||||
|
state.awaitingPostInjectionProgressCheck = true
|
||||||
|
const thirdCycle = sessionStateStore.trackContinuationProgress(sessionID, 2, todos)
|
||||||
|
|
||||||
|
// then
|
||||||
|
expect(firstCycle.stagnationCount).toBe(1)
|
||||||
|
expect(secondCycle.stagnationCount).toBe(2)
|
||||||
|
expect(thirdCycle.stagnationCount).toBe(3)
|
||||||
|
|
||||||
|
sessionStateStore.shutdown()
|
||||||
|
})
|
||||||
|
})
|
||||||
|
})
|
||||||
})
|
})
|
||||||
|
|||||||
Reference in New Issue
Block a user