File size: 12,262 Bytes
4440aec | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 | import { sanitizeForLog } from "../../../packages/terminal-core/src/ansi.js";
import {
classifyOAuthRefreshFailure,
classifyOAuthRefreshFailureError,
} from "../../agents/auth-profiles/oauth-refresh-failure.js";
import {
isCompactionFailureError,
isLikelyContextOverflowError,
} from "../../agents/embedded-agent-helpers.js";
import { findCliTimeoutError, isFailoverError } from "../../agents/failover-error.js";
import {
GENERIC_EXTERNAL_RUN_FAILURE_TEXT,
HEARTBEAT_EXTERNAL_RUN_FAILURE_TEXT,
renderControlUiAgentFailureCopy,
renderFailoverCodeUserCopy,
} from "../../agents/failover/user-copy.js";
import { isAgentHarnessPreflightError } from "../../agents/harness/errors.js";
import { LiveSessionModelSwitchError } from "../../agents/live-model-switch-error.js";
import { formatErrorMessage } from "../../infra/errors.js";
import { CommandLaneClearedError, GatewayDrainingError } from "../../process/command-queue.js";
import { defaultRuntime } from "../../runtime.js";
import type { ReplyPayload } from "../types.js";
import { createAgentLifecycleTerminalBackstop } from "./agent-lifecycle-terminal.js";
import { buildContextOverflowRecoveryText } from "./agent-runner-context-recovery.js";
import type { AgentTurnInternalResult, AgentTurnParams } from "./agent-runner-execution.types.js";
import {
buildAuthProfileFailoverFailureText,
buildExternalRunFailureReply,
isNonDirectConversationContext,
isVerboseFailureDetailEnabled,
markAgentRunFailureReplyPayload,
resolveExternalRunFailureTextForConversation,
resolveReplyFailureSummary,
resolveReplyFailoverFacts,
} from "./agent-runner-failure-reply.js";
import type { AgentFallbackCycleState } from "./agent-runner-fallback-cycle.js";
import type { AgentTurnTimingTracker } from "./agent-runner-turn-timing.js";
import {
buildRestartLifecycleReplyText,
resolveReplyOperationAbortReason,
resolveReplyOperationTerminationFields,
resolveRestartLifecycleError,
} from "./reply-operation-abort.js";
const MAX_LIVE_SWITCH_RETRIES = 2;
type ErrorAction =
| { kind: "retry"; liveModelSwitchError?: LiveSessionModelSwitchError }
| Extract<AgentTurnInternalResult, { kind: "final" | "aborted" }>;
export async function handleAgentExecutionError(params: {
turn: AgentTurnParams;
error: unknown;
runtimeConfig: AgentTurnParams["followupRun"]["run"]["config"];
runId: string;
state: AgentFallbackCycleState;
liveModelSwitchRetries: number;
shouldSurfaceToControlUi: boolean;
timing: AgentTurnTimingTracker;
modelPatch: { fail: (error: unknown) => Promise<void> };
}): Promise<ErrorAction> {
const turn = params.turn;
const err = params.error;
// A failed candidate leaves its backstop pending; settlement takes it before later work.
// This keeps session-override failures from being mislabeled as model failures.
const postCompactionModelFailure =
params.state.postCompactionModelAttempted && params.state.pendingLifecycleTerminal
? true
: undefined;
const takePendingLifecycleTerminal = () => {
const terminal =
params.state.pendingLifecycleTerminal?.backstop ??
createAgentLifecycleTerminalBackstop({
runId: params.runId,
sessionKey: turn.sessionKey,
startedAt: params.state.turnStartedAtMs,
getLifecycleGeneration: () => params.state.lifecycleGeneration,
resolveTerminationFields: (error) =>
resolveReplyOperationTerminationFields(
error,
turn.replyOperation?.abortSignal ?? turn.opts?.abortSignal,
turn.replyOperation,
),
});
params.state.pendingLifecycleTerminal = undefined;
return terminal;
};
const settleFailure = async (
payload: ReplyPayload,
): Promise<Extract<AgentTurnInternalResult, { kind: "final" }>> => {
takePendingLifecycleTerminal().emit("error", err);
turn.replyOperation?.fail("run_failed", err);
await params.modelPatch.fail(err);
return {
kind: "final",
payload: markAgentRunFailureReplyPayload(payload),
postCompactionModelFailure,
};
};
const resolveReplyOperationAbortAction = (abortError: unknown): ErrorAction | undefined => {
const reason = resolveReplyOperationAbortReason(turn.replyOperation, abortError);
if (!reason) {
return undefined;
}
// Preserve signal-owned timeout attribution; only normalized restart/supersession need metadata.
const terminalMetadata = reason === "user" ? undefined : { aborted: true, stopReason: reason };
takePendingLifecycleTerminal().emit(
reason === "restart" ? "end" : "error",
abortError,
terminalMetadata,
);
return { kind: "aborted", reason };
};
const replyOperationAbortAction = resolveReplyOperationAbortAction(err);
if (replyOperationAbortAction) {
return replyOperationAbortAction;
}
if (err instanceof LiveSessionModelSwitchError) {
if (params.liveModelSwitchRetries <= MAX_LIVE_SWITCH_RETRIES) {
params.state.pendingLifecycleTerminal = undefined;
return { kind: "retry", liveModelSwitchError: err };
}
const visibleReplyDelivered = await turn.resolveVisibleReplyDelivery?.();
defaultRuntime.error(
`Live model switch failed after ${MAX_LIVE_SWITCH_RETRIES} retries ` +
`(${sanitizeForLog(err.provider)}/${sanitizeForLog(err.model)}). The requested model may be unavailable.`,
);
takePendingLifecycleTerminal().emit("error", err);
const switchErrorText = params.shouldSurfaceToControlUi
? renderControlUiAgentFailureCopy(
"model switch could not be completed. The requested model may be temporarily unavailable.",
)
: isVerboseFailureDetailEnabled(turn.resolvedVerboseLevel)
? "⚠️ Agent failed before reply: model switch could not be completed. " +
"The requested model may be temporarily unavailable. Please try again shortly."
: "⚠️ Model switch could not be completed. The requested model may be temporarily unavailable. Please try again shortly.";
turn.replyOperation?.fail("run_failed", err);
await params.modelPatch.fail(err);
return {
kind: "final",
payload: markAgentRunFailureReplyPayload({
text: resolveExternalRunFailureTextForConversation({
text: switchErrorText,
visibleReplyDelivered,
sessionCtx: turn.sessionCtx,
isGenericRunnerFailure: !params.shouldSurfaceToControlUi,
cfg: turn.followupRun.run.config,
}),
}),
};
}
const message = formatErrorMessage(err);
params.timing.logIfSlow({
runId: params.runId,
sessionId: turn.followupRun.run.sessionId,
sessionKey: turn.sessionKey,
outcome: "error",
error: message,
});
// The exhausted preflight is deliberate, even if its diagnostic cause looks
// like HTTP/overload. Settle delivery and normal diagnostic policy without replay.
if (isAgentHarnessPreflightError(err)) {
const externalReply = buildExternalRunFailureReply(
{ message, error: err },
{
includeDetails: isVerboseFailureDetailEnabled(turn.resolvedVerboseLevel),
isHeartbeat: turn.isHeartbeat,
},
);
const text = resolveExternalRunFailureTextForConversation({
text: params.shouldSurfaceToControlUi
? renderControlUiAgentFailureCopy(message)
: externalReply.text,
visibleReplyDelivered: await turn.resolveVisibleReplyDelivery?.(),
sessionCtx: turn.sessionCtx,
isGenericRunnerFailure: externalReply.isGenericRunnerFailure,
cfg: turn.followupRun.run.config,
});
return await settleFailure({ text });
}
const failoverFacts = resolveReplyFailoverFacts(err, message);
const failureSummary = resolveReplyFailureSummary({
error: err,
message,
reason: failoverFacts.reason,
attempts: isFailoverError(err) ? err.attempts : undefined,
});
const failoverReason = failoverFacts.reason;
const isBilling = failureSummary?.kind === "billing";
const isContextOverflow =
!isBilling && (failoverReason === "context_overflow" || isLikelyContextOverflowError(message));
const isCompactionFailure = !isBilling && isCompactionFailureError(message);
const oauthRefreshFailure =
classifyOAuthRefreshFailureError(err) ?? classifyOAuthRefreshFailure(message);
const hasAuthProfileFailoverFailure = buildAuthProfileFailoverFailureText(err) !== null;
const providerRequestError =
!isBilling &&
!oauthRefreshFailure &&
!hasAuthProfileFailoverFailure &&
!params.shouldSurfaceToControlUi
? failoverFacts.providerRequestError
: undefined;
const restartLifecycleError = resolveRestartLifecycleError(err);
if (
restartLifecycleError instanceof GatewayDrainingError ||
restartLifecycleError instanceof CommandLaneClearedError
) {
takePendingLifecycleTerminal().emit("error", restartLifecycleError);
turn.replyOperation?.fail(
restartLifecycleError instanceof GatewayDrainingError
? "gateway_draining"
: "command_lane_cleared",
restartLifecycleError,
);
return {
kind: "final",
payload: markAgentRunFailureReplyPayload({ text: buildRestartLifecycleReplyText() }),
};
}
if (isCompactionFailure) {
takePendingLifecycleTerminal().emit("error", err);
defaultRuntime.error(
`Auto-compaction failed (${message}). Preserving existing session mapping for ${turn.sessionKey ?? turn.followupRun.run.sessionId}.`,
);
turn.replyOperation?.fail("run_failed", err);
return {
kind: "final",
payload: markAgentRunFailureReplyPayload({
text: buildContextOverflowRecoveryText({
duringCompaction: true,
preserveSessionMapping: true,
cfg: params.runtimeConfig,
agentId: turn.followupRun.run.agentId,
primaryProvider: turn.followupRun.run.provider,
primaryModel: turn.followupRun.run.model,
runtimeProvider: params.state.attemptedRuntimeProvider,
runtimeModel: params.state.attemptedRuntimeModel,
activeSessionEntry: turn.getActiveSessionEntry(),
}),
}),
};
}
const replayPrevented = findCliTimeoutError(err)?.cliTimeout.observedActivity === true;
if (providerRequestError) {
return await settleFailure({
// Curated facet copy beats the generic classified summary; see
// buildExternalRunFailureReply for the same priority.
text: providerRequestError.userMessage,
});
}
defaultRuntime.error(`Embedded agent failed before reply: ${message}`);
const externalRunFailureCandidate =
!failureSummary && !isContextOverflow
? buildExternalRunFailureReply(
{ message, error: err },
{
includeAuthProfileId: !isNonDirectConversationContext(turn.sessionCtx),
includeDetails: isVerboseFailureDetailEnabled(turn.resolvedVerboseLevel),
isHeartbeat: turn.isHeartbeat,
replayPrevented,
failoverFacts,
},
)
: undefined;
const externalRunFailureReply =
!params.shouldSurfaceToControlUi ||
externalRunFailureCandidate?.presentation ||
renderFailoverCodeUserCopy(failoverFacts.code)
? externalRunFailureCandidate
: undefined;
const fallbackText =
failureSummary?.text ??
(isContextOverflow
? "⚠️ Context overflow — prompt too large for this model. Try a shorter message or a larger-context model."
: (externalRunFailureReply?.text ??
(params.shouldSurfaceToControlUi
? renderControlUiAgentFailureCopy(message)
: turn.isHeartbeat
? HEARTBEAT_EXTERNAL_RUN_FAILURE_TEXT
: GENERIC_EXTERNAL_RUN_FAILURE_TEXT)));
const userVisibleFallbackText = resolveExternalRunFailureTextForConversation({
text: fallbackText,
visibleReplyDelivered: await turn.resolveVisibleReplyDelivery?.(),
sessionCtx: turn.sessionCtx,
isGenericRunnerFailure: externalRunFailureReply?.isGenericRunnerFailure ?? false,
cfg: turn.followupRun.run.config,
});
return await settleFailure({
text: userVisibleFallbackText,
...(externalRunFailureReply?.presentation
? { presentation: externalRunFailureReply.presentation }
: {}),
});
}
|