fix(ralph-loop): retry runtime errors immediately

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
This commit is contained in:
YeonGyu-Kim
2026-05-06 16:08:13 +09:00
parent 44216a538e
commit 8e1719968b
2 changed files with 197 additions and 19 deletions
@@ -25,7 +25,7 @@ describe("ralph-loop non-abort error continuation", () => {
}
})
test("continues on next idle after non-abort session error", async () => {
test("continues immediately after non-abort session error", async () => {
// given - an active Ralph Loop receives a recoverable command error
const hook = createRalphLoopHook({
directory: testDirectory,
@@ -81,16 +81,75 @@ describe("ralph-loop non-abort error continuation", () => {
},
})
// when - OpenCode emits the idle event caused by that failed command
await hook.event({
event: { type: "session.idle", properties: { sessionID: "session-123" } },
})
// then - the loop should continue instead of skipping idle as recovery
// then - the loop should continue without waiting for a later idle event
expect(promptCalls).toHaveLength(1)
expect(promptCalls[0]?.sessionID).toBe("session-123")
expect(promptCalls[0]?.text).toContain("Keep working")
expect(messagesCalls.length).toBeGreaterThan(0)
expect(hook.getState()?.iteration).toBe(2)
})
test("stops retrying runtime errors after max iterations", async () => {
// given - an active Ralph Loop has one retry remaining
const hook = createRalphLoopHook({
directory: testDirectory,
project: testDirectory,
worktree: testDirectory,
serverUrl: "http://localhost:4096",
$: async () => ({}),
client: {
session: {
messages: async (options: { path: { id: string } }) => {
messagesCalls.push({ sessionID: options.path.id })
return { data: [] }
},
promptAsync: async (options: {
path: { id: string }
body: { parts: Array<{ type: string; text: string }> }
}) => {
promptCalls.push({
sessionID: options.path.id,
text: options.body.parts[0]?.text ?? "",
})
return {}
},
prompt: async () => ({}),
},
tui: {
showToast: async () => ({}),
},
},
} as never)
hook.startLoop("session-123", "Keep working", {
messageCountAtStart: 0,
maxIterations: 2,
})
// when - the first runtime error consumes the final allowed attempt
await hook.event({
event: {
type: "session.error",
properties: {
sessionID: "session-123",
error: { name: "RuntimeError" },
},
},
})
// when - another runtime error arrives after the retry budget is exhausted
await hook.event({
event: {
type: "session.error",
properties: {
sessionID: "session-123",
error: { name: "RuntimeError" },
},
},
})
// then - the loop does not exceed the configured retry count
expect(promptCalls).toHaveLength(1)
expect(hook.getState()).toBeNull()
})
})
+131 -12
View File
@@ -22,11 +22,42 @@ type LoopStateController = {
}
type RalphLoopEventHandlerOptions = { directory: string; apiTimeoutMs: number; getTranscriptPath: (sessionID: string) => string | undefined; checkSessionExists?: RalphLoopOptions["checkSessionExists"]; backgroundManager?: RalphLoopOptions["backgroundManager"]; loopState: LoopStateController }
function isAbortError(error: unknown): boolean {
return typeof error === "object"
&& error !== null
&& "name" in error
&& (error as { name?: unknown }).name === "MessageAbortedError"
}
async function showMaxIterationsToast(
ctx: PluginInput,
state: RalphLoopState,
): Promise<void> {
await ctx.client.tui?.showToast?.({
body: { title: "Ralph Loop Stopped", message: `Max iterations (${state.max_iterations}) reached without completion`, variant: "warning", duration: 5000 },
}).catch(() => {})
}
async function showIterationToast(
ctx: PluginInput,
state: RalphLoopState,
): Promise<void> {
await ctx.client.tui?.showToast?.({
body: {
title: "Ralph Loop",
message: `Iteration ${state.iteration}/${typeof state.max_iterations === "number" ? state.max_iterations : "unbounded"}`,
variant: "info",
duration: 2000,
},
}).catch(() => {})
}
export function createRalphLoopEventHandler(
ctx: PluginInput,
options: RalphLoopEventHandlerOptions,
) {
const inFlightSessions = new Set<string>()
const runtimeErrorRetriedSessions = new Map<string, number>()
return async ({ event }: { event: { type: string; properties?: unknown } }): Promise<void> => {
const props = event.properties as Record<string, unknown> | undefined
@@ -121,6 +152,7 @@ export function createRalphLoopEventHandler(
})
if (completionViaTranscript || completionViaApi) {
runtimeErrorRetriedSessions.delete(sessionID)
log(`[${HOOK_NAME}] Completion detected!`, {
sessionID,
iteration: state.iteration,
@@ -160,6 +192,15 @@ export function createRalphLoopEventHandler(
return
}
if (runtimeErrorRetriedSessions.get(sessionID) === state.iteration) {
runtimeErrorRetriedSessions.delete(sessionID)
log(`[${HOOK_NAME}] Skipped stale idle after runtime error retry`, {
sessionID,
iteration: state.iteration,
})
return
}
if (
typeof state.max_iterations === "number"
&& state.iteration >= state.max_iterations
@@ -171,9 +212,7 @@ export function createRalphLoopEventHandler(
})
options.loopState.clear()
await ctx.client.tui?.showToast?.({
body: { title: "Ralph Loop Stopped", message: `Max iterations (${state.max_iterations}) reached without completion`, variant: "warning", duration: 5000 },
}).catch(() => {})
await showMaxIterationsToast(ctx, state)
return
}
@@ -189,14 +228,7 @@ export function createRalphLoopEventHandler(
max: newState.max_iterations,
})
await ctx.client.tui?.showToast?.({
body: {
title: "Ralph Loop",
message: `Iteration ${newState.iteration}/${typeof newState.max_iterations === "number" ? newState.max_iterations : "unbounded"}`,
variant: "info",
duration: 2000,
},
}).catch(() => {})
await showIterationToast(ctx, newState)
try {
await continueIteration(ctx, newState, {
@@ -223,7 +255,94 @@ export function createRalphLoopEventHandler(
}
if (event.type === "session.error") {
handleErroredLoopSession(props, options.loopState)
const sessionID = props?.sessionID as string | undefined
const error = props?.error
if (!sessionID || isAbortError(error)) {
handleErroredLoopSession(props, options.loopState)
return
}
if (inFlightSessions.has(sessionID)) {
log(`[${HOOK_NAME}] Skipped runtime error retry: handler in flight`, { sessionID })
return
}
inFlightSessions.add(sessionID)
try {
const state = options.loopState.getState()
if (!state || !state.active) {
handleErroredLoopSession(props, options.loopState)
return
}
const verificationSessionID = state.verification_pending
? state.verification_session_id
: undefined
const matchesParentSession = state.session_id === undefined || state.session_id === sessionID
const matchesVerificationSession = verificationSessionID === sessionID
if (!matchesParentSession && !matchesVerificationSession) {
handleErroredLoopSession(props, options.loopState)
return
}
log(`[${HOOK_NAME}] Retrying after runtime session error`, {
sessionID,
iteration: state.iteration,
error: String(error),
})
if (state.verification_pending) {
await handlePendingVerification(ctx, {
sessionID,
state,
verificationSessionID,
matchesParentSession,
matchesVerificationSession,
loopState: options.loopState,
directory: options.directory,
apiTimeoutMs: options.apiTimeoutMs,
})
return
}
if (
typeof state.max_iterations === "number"
&& state.iteration >= state.max_iterations
) {
log(`[${HOOK_NAME}] Runtime error retry budget exhausted`, {
sessionID,
iteration: state.iteration,
max: state.max_iterations,
})
options.loopState.clear()
await showMaxIterationsToast(ctx, state)
return
}
const newState = options.loopState.incrementIteration()
if (!newState) {
log(`[${HOOK_NAME}] Failed to increment iteration after runtime error`, { sessionID })
return
}
await showIterationToast(ctx, newState)
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),
})
}
} finally {
inFlightSessions.delete(sessionID)
}
}
}
}