Merge pull request #3939 from code-yeongyu/fix/ralph-loop-message-dispatch

fix(ralph-loop): commit iteration only after continuation is dispatched
This commit is contained in:
YeonGyu-Kim
2026-05-11 13:07:53 +09:00
committed by GitHub
7 changed files with 629 additions and 75 deletions
@@ -0,0 +1,412 @@
/// <reference types="bun-types" />
import { afterEach, beforeEach, describe, expect, test } from "bun:test"
import { existsSync, mkdirSync, rmSync, writeFileSync } from "node:fs"
import { tmpdir } from "node:os"
import { join } from "node:path"
import { createRalphLoopHook } from "./index"
import { ULTRAWORK_VERIFICATION_PROMISE } from "./constants"
import { clearState, writeState } from "./storage"
import { handleFailedVerification } from "./verification-failure-handler"
describe("ralph-loop dispatch failure invariants", () => {
const testDirectory = join(tmpdir(), `ralph-loop-dispatch-failure-${Date.now()}`)
let promptCalls: Array<{ sessionID: string; text: string }>
let toastCalls: Array<{ title: string; message: string; variant: string }>
let messagesCalls: Array<{ sessionID: string }>
let createSessionCalls: Array<{ parentID: string }>
beforeEach(() => {
promptCalls = []
toastCalls = []
messagesCalls = []
createSessionCalls = []
mkdirSync(testDirectory, { recursive: true })
clearState(testDirectory)
})
afterEach(() => {
clearState(testDirectory)
if (existsSync(testDirectory)) {
rmSync(testDirectory, { recursive: true, force: true })
}
})
test("#given idle path #when promptAsync throws #then no state or toast advance", async () => {
// given
const hook = createRalphLoopHook({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async (options: { path: { id: string } }) => {
messagesCalls.push({ sessionID: options.path.id })
return { data: [] }
},
promptAsync: async () => {
throw new Error("simulated dispatch failure")
},
prompt: async () => ({}),
create: async () => ({ data: { id: "new-session-id" } }),
},
tui: {
showToast: async (options: { body: { title: string; message: string; variant: string } }) => {
toastCalls.push(options.body)
return {}
},
},
},
} as never)
hook.startLoop("session-123", "Keep working", {
messageCountAtStart: 0,
maxIterations: 5,
})
expect(hook.getState()?.iteration).toBe(1)
// when
await hook.event({
event: { type: "session.idle", properties: { sessionID: "session-123" } },
})
// then
expect(toastCalls.some((toast) => toast.title === "Ralph Loop" && toast.message.includes("Iteration"))).toBe(false)
expect(hook.getState()).toBeNull()
expect(toastCalls.some((toast) => toast.title === "Ralph Loop Failed" && toast.message.includes("dispatch_rejected"))).toBe(true)
})
test("#given error retry path #when promptAsync throws #then no state or toast advance", async () => {
// given
const hook = createRalphLoopHook({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async (options: { path: { id: string } }) => {
messagesCalls.push({ sessionID: options.path.id })
return { data: [] }
},
promptAsync: async () => {
throw new Error("simulated dispatch failure")
},
prompt: async () => ({}),
create: async () => ({ data: { id: "new-session-id" } }),
},
tui: {
showToast: async (options: { body: { title: string; message: string; variant: string } }) => {
toastCalls.push(options.body)
return {}
},
},
},
} as never)
hook.startLoop("session-123", "Keep working", {
messageCountAtStart: 0,
maxIterations: 5,
})
expect(hook.getState()?.iteration).toBe(1)
// when
await hook.event({
event: {
type: "session.error",
properties: {
sessionID: "session-123",
error: { name: "RuntimeError" },
},
},
})
// then
expect(toastCalls.some((toast) => toast.title === "Ralph Loop" && toast.message.includes("Iteration"))).toBe(false)
expect(hook.getState()).toBeNull()
expect(toastCalls.some((toast) => toast.title === "Ralph Loop Failed" && toast.message.includes("dispatch_rejected"))).toBe(true)
})
test("#given verification-failure path #when promptAsync throws #then iteration not advanced", async () => {
// given
const parentTranscriptPath = join(testDirectory, "transcript-parent.jsonl")
const oracleTranscriptPath = join(testDirectory, "transcript-oracle.jsonl")
const hook = createRalphLoopHook({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async (options: { path: { id: string } }) => {
messagesCalls.push({ sessionID: options.path.id })
if (options.path.id === "session-123") {
return { data: [{}, {}, {}] }
}
return { data: [] }
},
promptAsync: async (options: { body: { parts: Array<{ type: string; text: string }> } }) => {
if (options.body.parts[0]?.text.includes("Verification failed")) {
throw new Error("simulated dispatch failure")
}
return {}
},
prompt: async () => ({}),
abort: async () => ({}),
create: async () => ({ data: { id: "new-session-id" } }),
},
tui: {
showToast: async (options: { body: { title: string; message: string; variant: string } }) => {
toastCalls.push(options.body)
return {}
},
},
},
} as never, {
getTranscriptPath: (sessionID): string => sessionID === "ses-oracle" ? oracleTranscriptPath : parentTranscriptPath,
})
hook.startLoop("session-123", "Build API", { ultrawork: true })
writeState(testDirectory, {
...hook.getState()!,
iteration: 2,
verification_pending: true,
verification_session_id: "ses-oracle",
completion_promise: ULTRAWORK_VERIFICATION_PROMISE,
initial_completion_promise: "DONE",
})
writeState(testDirectory, {
...hook.getState()!,
verification_session_id: "ses-oracle",
})
writeFileSync(
oracleTranscriptPath,
`${JSON.stringify({ type: "tool_result", timestamp: new Date().toISOString(), tool_output: { output: "verification failed" } })}\n`,
)
const preRestartIteration = hook.getState()?.iteration
// when
await hook.event({ event: { type: "session.idle", properties: { sessionID: "ses-oracle" } } })
// then
expect(preRestartIteration).toBe(2)
expect(hook.getState()).toBeNull()
expect(toastCalls.some((toast) => toast.title === "Ralph Loop Failed" && toast.message.includes("Verification continuation rejected"))).toBe(true)
})
test("#given reset strategy #when createIterationSession returns null #then dispatch failure surfaces", async () => {
// given
const hook = createRalphLoopHook({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async (options: { path: { id: string } }) => {
messagesCalls.push({ sessionID: options.path.id })
return { data: [] }
},
promptAsync: async (options: { path: { id: string }; body: { parts: Array<{ type: string; text: string }> } }) => {
promptCalls.push({
sessionID: options.path.id,
text: options.body.parts[0]?.text ?? "",
})
return {}
},
prompt: async () => ({}),
create: async (options: { body: { parentID: string } }) => {
createSessionCalls.push({ parentID: options.body.parentID })
return { error: "fail", data: undefined }
},
},
tui: {
showToast: async (options: { body: { title: string; message: string; variant: string } }) => {
toastCalls.push(options.body)
return {}
},
},
},
} as never)
hook.startLoop("session-123", "Keep working", {
messageCountAtStart: 0,
maxIterations: 5,
strategy: "reset",
})
expect(hook.getState()?.iteration).toBe(1)
// when
await hook.event({
event: { type: "session.idle", properties: { sessionID: "session-123" } },
})
// then
expect(hook.getState()).toBeNull()
expect(promptCalls).toHaveLength(0)
expect(createSessionCalls).toHaveLength(1)
expect(toastCalls.some((toast) => toast.title === "Ralph Loop Failed" && toast.message.includes("session_creation_rejected"))).toBe(true)
})
test("#given idle path #when state rebound during settle window #then no dispatch against new owner", async () => {
// given
const hook = createRalphLoopHook({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async () => ({ data: [] }),
promptAsync: async (options: { path: { id: string }; body: { parts: Array<{ type: string; text: string }> } }) => {
promptCalls.push({ sessionID: options.path.id, text: options.body.parts[0]?.text ?? "" })
return {}
},
prompt: async () => ({}),
create: async () => ({ data: { id: "new-session-id" } }),
},
tui: {
showToast: async (options: { body: { title: string; message: string; variant: string } }) => {
toastCalls.push(options.body)
return {}
},
},
},
} as never, {
idleSettleMs: 50,
})
hook.startLoop("session-A", "Keep working", { messageCountAtStart: 0, maxIterations: 5 })
expect(hook.getState()?.session_id).toBe("session-A")
// when
const eventPromise = hook.event({
event: { type: "session.idle", properties: { sessionID: "session-A" } },
})
await new Promise((resolve) => setTimeout(resolve, 10))
writeState(testDirectory, { ...hook.getState()!, session_id: "session-B" })
await eventPromise
// then
expect(promptCalls).toHaveLength(0)
expect(hook.getState()?.session_id).toBe("session-B")
expect(hook.getState()?.iteration).toBe(1)
})
test("#given verification-failure path #when incrementIteration fails #then loud failure not success", async () => {
// given
const loopState = {
clearVerificationState: () => ({
active: true,
iteration: 2,
prompt: "Build API",
started_at: new Date().toISOString(),
session_id: "session-123",
completion_promise: ULTRAWORK_VERIFICATION_PROMISE,
message_count_at_start: 3,
}),
incrementIteration: () => null,
clear: () => true,
}
const result = await handleFailedVerification({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async () => ({ data: [{}, {}, {}] }),
promptAsync: async (options: { path: { id: string }; body: { parts: Array<{ type: string; text: string }> } }) => {
promptCalls.push({ sessionID: options.path.id, text: options.body.parts[0]?.text ?? "" })
return {}
},
abort: async () => ({}),
},
tui: {
showToast: async (options: { body: { title: string; message: string; variant: string } }) => {
toastCalls.push(options.body)
return {}
},
},
},
} as never, {
state: {
active: true,
iteration: 2,
prompt: "Build API",
started_at: new Date().toISOString(),
session_id: "session-123",
completion_promise: ULTRAWORK_VERIFICATION_PROMISE,
verification_pending: true,
verification_session_id: "ses-oracle",
},
directory: testDirectory,
apiTimeoutMs: 5000,
loopState,
})
// then
expect(result).toBe(false)
expect(promptCalls).toHaveLength(1)
expect(toastCalls.some((toast) => toast.title === "ULTRAWORK LOOP")).toBe(false)
expect(
toastCalls.some(
(toast) => toast.title === "Ralph Loop Failed" && toast.message.includes("iteration commit failed"),
),
).toBe(true)
})
test("#given reset strategy #when session.create throws #then dispatch failure surfaces", async () => {
// given
const hook = createRalphLoopHook({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async () => ({ data: [] }),
promptAsync: async (options: { path: { id: string }; body: { parts: Array<{ type: string; text: string }> } }) => {
promptCalls.push({ sessionID: options.path.id, text: options.body.parts[0]?.text ?? "" })
return {}
},
prompt: async () => ({}),
create: async () => {
throw new Error("simulated network error during session.create")
},
},
tui: {
showToast: async (options: { body: { title: string; message: string; variant: string } }) => {
toastCalls.push(options.body)
return {}
},
},
},
} as never)
hook.startLoop("session-123", "Keep working", {
messageCountAtStart: 0,
maxIterations: 5,
strategy: "reset",
})
// when
await hook.event({
event: { type: "session.idle", properties: { sessionID: "session-123" } },
})
// then
expect(hook.getState()).toBeNull()
expect(promptCalls).toHaveLength(0)
expect(toastCalls.some((toast) => toast.title === "Ralph Loop Failed" && toast.message.includes("session_creation_rejected"))).toBe(true)
})
})
+32 -17
View File
@@ -15,11 +15,16 @@ type ContinuationOptions = {
}
}
export type ContinuationResult =
| { status: "dispatched" }
| { status: "session_creation_rejected" }
| { status: "dispatch_rejected"; error: unknown }
export async function continueIteration(
ctx: PluginInput,
state: RalphLoopState,
options: ContinuationOptions,
): Promise<void> {
): Promise<ContinuationResult> {
const strategy = state.strategy ?? "continue"
const continuationPrompt = buildContinuationPrompt(state)
@@ -30,16 +35,20 @@ export async function continueIteration(
options.directory,
)
if (!newSessionID) {
return
return { status: "session_creation_rejected" }
}
await injectContinuationPrompt(ctx, {
sessionID: newSessionID,
inheritFromSessionID: options.previousSessionID,
prompt: continuationPrompt,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
})
try {
await injectContinuationPrompt(ctx, {
sessionID: newSessionID,
inheritFromSessionID: options.previousSessionID,
prompt: continuationPrompt,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
})
} catch (error: unknown) {
return { status: "dispatch_rejected", error }
}
await selectSessionInTui(ctx.client, newSessionID)
@@ -49,16 +58,22 @@ export async function continueIteration(
previousSessionID: options.previousSessionID,
newSessionID,
})
return
return { status: "dispatched" }
}
return
return { status: "dispatched" }
}
await injectContinuationPrompt(ctx, {
sessionID: options.previousSessionID,
prompt: continuationPrompt,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
})
try {
await injectContinuationPrompt(ctx, {
sessionID: options.previousSessionID,
prompt: continuationPrompt,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
})
} catch (error: unknown) {
return { status: "dispatch_rejected", error }
}
return { status: "dispatched" }
}
@@ -174,5 +174,27 @@ export function createLoopStateController(options: {
return state
},
clearVerificationState(sessionID: string, messageCountAtStart?: number): RalphLoopState | null {
const state = readState(directory, stateDir)
if (!state || state.session_id !== sessionID || !state.ultrawork || !state.verification_pending) {
return null
}
state.started_at = new Date().toISOString()
state.completion_promise = state.initial_completion_promise ?? DEFAULT_COMPLETION_PROMISE
state.verification_pending = undefined
state.verification_attempt_id = undefined
state.verification_session_id = undefined
if (typeof messageCountAtStart === "number") {
state.message_count_at_start = messageCountAtStart
}
if (!writeState(directory, state, stateDir)) {
return null
}
return state
},
}
}
@@ -82,6 +82,9 @@ async function detectOracleVerificationFromParentSession(
type LoopStateController = {
restartAfterFailedVerification: (sessionID: string, messageCountAtStart?: number) => RalphLoopState | null
clearVerificationState: (sessionID: string, messageCountAtStart?: number) => RalphLoopState | null
incrementIteration: () => RalphLoopState | null
clear: () => boolean
setVerificationSessionID: (sessionID: string, verificationSessionID: string) => RalphLoopState | null
}
@@ -19,6 +19,7 @@ type LoopStateController = {
markVerificationPending: (sessionID: string) => RalphLoopState | null
setVerificationSessionID: (sessionID: string, verificationSessionID: string) => RalphLoopState | null
restartAfterFailedVerification: (sessionID: string, messageCountAtStart?: number) => RalphLoopState | null
clearVerificationState: (sessionID: string, messageCountAtStart?: number) => RalphLoopState | null
}
type RalphLoopEventHandlerOptions = { directory: string; apiTimeoutMs: number; idleSettleMs: number; getTranscriptPath: (sessionID: string) => string | undefined; checkSessionExists?: RalphLoopOptions["checkSessionExists"]; backgroundManager?: RalphLoopOptions["backgroundManager"]; loopState: LoopStateController }
@@ -272,34 +273,59 @@ export function createRalphLoopEventHandler(
return
}
const newState = options.loopState.incrementIteration()
if (!newState) {
log(`[${HOOK_NAME}] Failed to increment iteration`, { sessionID })
await sleep(options.idleSettleMs)
const stateAfterSettle = options.loopState.getState()
if (!stateAfterSettle || !stateAfterSettle.active) {
return
}
if (stateAfterSettle.session_id !== undefined && stateAfterSettle.session_id !== sessionID) {
log(`[${HOOK_NAME}] Skipped: state rebound during settle window`, {
sessionID,
currentOwner: stateAfterSettle.session_id,
})
return
}
if (stateAfterSettle.verification_pending) {
log(`[${HOOK_NAME}] Skipped: state entered verification_pending during settle window`, { sessionID })
return
}
const nextIteration = stateAfterSettle.iteration + 1
const previewState: RalphLoopState = { ...stateAfterSettle, iteration: nextIteration }
log(`[${HOOK_NAME}] Continuing loop`, {
sessionID,
iteration: newState.iteration,
max: newState.max_iterations,
iteration: nextIteration,
max: previewState.max_iterations,
})
showIterationToast(ctx, newState)
await sleep(options.idleSettleMs)
const result = await continueIteration(ctx, previewState, {
previousSessionID: sessionID,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
loopState: options.loopState,
})
try {
await continueIteration(ctx, newState, {
previousSessionID: sessionID,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
loopState: options.loopState,
})
} catch (err) {
log(`[${HOOK_NAME}] Failed to inject continuation`, {
sessionID,
error: String(err),
})
if (result.status === "dispatched") {
const committed = options.loopState.incrementIteration()
if (committed) {
showIterationToast(ctx, committed)
} else {
log(`[${HOOK_NAME}] Dispatch succeeded but iteration commit failed`, { sessionID })
}
return
}
log(`[${HOOK_NAME}] Dispatch failed`, { sessionID, status: result.status })
options.loopState.clear()
showToastBestEffort(ctx, {
title: "Ralph Loop Failed",
message: result.status === "dispatch_rejected"
? `Dispatch ${result.status}: ${String(result.error)}`
: `Dispatch ${result.status}`,
variant: "warning",
duration: 5000,
})
return
} finally {
inFlightSessions.delete(sessionID)
@@ -381,28 +407,54 @@ export function createRalphLoopEventHandler(
return
}
const newState = options.loopState.incrementIteration()
if (!newState) {
log(`[${HOOK_NAME}] Failed to increment iteration after runtime error`, { sessionID })
await sleep(options.idleSettleMs)
const stateAfterSettle = options.loopState.getState()
if (!stateAfterSettle || !stateAfterSettle.active) {
return
}
if (stateAfterSettle.session_id !== undefined && stateAfterSettle.session_id !== sessionID) {
log(`[${HOOK_NAME}] Skipped: state rebound during settle window`, {
sessionID,
currentOwner: stateAfterSettle.session_id,
})
return
}
if (stateAfterSettle.verification_pending) {
log(`[${HOOK_NAME}] Skipped: state entered verification_pending during settle window`, { sessionID })
return
}
showIterationToast(ctx, newState)
await sleep(options.idleSettleMs)
try {
await continueIteration(ctx, newState, {
previousSessionID: sessionID,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
loopState: options.loopState,
})
runtimeErrorRetriedSessions.set(sessionID, newState.iteration)
} catch (err) {
log(`[${HOOK_NAME}] Failed to retry after runtime error`, {
sessionID,
error: String(err),
})
const nextIteration = stateAfterSettle.iteration + 1
const previewState: RalphLoopState = { ...stateAfterSettle, iteration: nextIteration }
const result = await continueIteration(ctx, previewState, {
previousSessionID: sessionID,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
loopState: options.loopState,
})
if (result.status === "dispatched") {
const committed = options.loopState.incrementIteration()
if (committed) {
showIterationToast(ctx, committed)
runtimeErrorRetriedSessions.set(sessionID, committed.iteration)
} else {
log(`[${HOOK_NAME}] Dispatch succeeded but iteration commit failed after runtime error`, { sessionID })
}
return
}
log(`[${HOOK_NAME}] Dispatch failed after runtime error`, { sessionID, status: result.status })
options.loopState.clear()
showToastBestEffort(ctx, {
title: "Ralph Loop Failed",
message: result.status === "dispatch_rejected"
? `Dispatch ${result.status}: ${String(result.error)}`
: `Dispatch ${result.status}`,
variant: "warning",
duration: 5000,
})
} finally {
inFlightSessions.delete(sessionID)
}
+20 -12
View File
@@ -7,23 +7,31 @@ export async function createIterationSession(
parentSessionID: string,
directory: string,
): Promise<string | null> {
const createResult = await ctx.client.session.create({
body: {
parentID: parentSessionID,
title: "Ralph Loop Iteration",
},
query: { directory },
})
try {
const createResult = await ctx.client.session.create({
body: {
parentID: parentSessionID,
title: "Ralph Loop Iteration",
},
query: { directory },
})
if (createResult.error || !createResult.data?.id) {
log("[ralph-loop] Failed to create iteration session", {
if (createResult.error || !createResult.data?.id) {
log("[ralph-loop] Failed to create iteration session", {
parentSessionID,
error: String(createResult.error ?? "No session ID returned"),
})
return null
}
return createResult.data.id
} catch (error: unknown) {
log("[ralph-loop] session.create threw during iteration session creation", {
parentSessionID,
error: String(createResult.error ?? "No session ID returned"),
error: String(error),
})
return null
}
return createResult.data.id
}
export async function selectSessionInTui(
@@ -6,10 +6,22 @@ import { injectContinuationPrompt } from "./continuation-prompt-injector"
import type { RalphLoopState } from "./types"
type LoopStateController = {
restartAfterFailedVerification: (
clearVerificationState: (
sessionID: string,
messageCountAtStart?: number,
) => RalphLoopState | null
incrementIteration: () => RalphLoopState | null
clear: () => boolean
}
function showToastBestEffort(
ctx: PluginInput,
body: { title: string; message: string; variant: "warning" | "info"; duration: number },
): void {
try {
void Promise.resolve(ctx.client.tui?.showToast?.({ body })).catch(() => {})
} catch {
}
}
function getMessageCountFromResponse(messagesResponse: unknown): number {
@@ -72,23 +84,53 @@ export async function handleFailedVerification(
ctx.client.session.abort({ path: { id: state.verification_session_id } }).catch(() => {})
}
const resumedState = loopState.restartAfterFailedVerification(
const clearedState = loopState.clearVerificationState(
parentSessionID,
messageCountAtStart,
)
if (!resumedState) {
if (!clearedState) {
log(`[${HOOK_NAME}] Failed to restart loop after verification failure`, {
parentSessionID,
})
return false
}
await injectContinuationPrompt(ctx, {
sessionID: parentSessionID,
prompt: buildVerificationFailurePrompt(resumedState),
directory,
apiTimeoutMs,
})
const previewState: RalphLoopState = { ...clearedState, iteration: clearedState.iteration + 1 }
try {
await injectContinuationPrompt(ctx, {
sessionID: parentSessionID,
prompt: buildVerificationFailurePrompt(previewState),
directory,
apiTimeoutMs,
})
} catch (error) {
log(`[${HOOK_NAME}] Failed to inject verification failure prompt`, {
parentSessionID,
error: String(error),
})
loopState.clear()
showToastBestEffort(ctx, {
title: "Ralph Loop Failed",
message: `Verification continuation rejected: ${String(error)}`,
variant: "warning",
duration: 5000,
})
return false
}
const committed = loopState.incrementIteration()
if (!committed) {
log(`[${HOOK_NAME}] Failed to commit iteration after verification restart`, { parentSessionID })
loopState.clear()
showToastBestEffort(ctx, {
title: "Ralph Loop Failed",
message: "Verification continuation dispatched but iteration commit failed",
variant: "warning",
duration: 5000,
})
return false
}
await ctx.client.tui?.showToast?.({
body: {