fix(agents): prevent totalTokens crash when assistant usage is missing (#34977)

Merged via squash.

Prepared head SHA: 1c14094f3f
Co-authored-by: sp-hk2ldn <8068616+sp-hk2ldn@users.noreply.github.com>
Co-authored-by: jalehman <550978+jalehman@users.noreply.github.com>
Reviewed-by: @jalehman
This commit is contained in:
SP
2026-03-06 23:59:16 +00:00
committed by GitHub
parent 48b3c4a043
commit 942c53e7f0
3 changed files with 240 additions and 3 deletions

View File

@@ -213,6 +213,7 @@ Docs: https://docs.openclaw.ai
- Gateway/probes: keep `/health`, `/healthz`, `/ready`, and `/readyz` reachable when the Control UI is mounted at `/`, preserve plugin-owned route precedence on those paths, and make `/ready` and `/readyz` report channel-backed readiness with startup grace plus `503` on disconnected managed channels, while `/health` and `/healthz` stay shallow liveness probes. (#18446) Thanks @vibecodooor, @mahsumaktas, and @vincentkoc.
- Feishu/media downloads: drop invalid timeout fields from SDK method calls now that client-level `httpTimeoutMs` applies to requests. (#38267) Thanks @ant1eicher and @thewilloftheshadow.
- PI embedded runner/Feishu docs: propagate sender identity into embedded attempts so Feishu doc auto-grant restores requester access for embedded-runner executions. (#32915) thanks @cszhouwei.
- Agents/usage normalization: normalize missing or partial assistant usage snapshots before compaction accounting so `openclaw agent --json` no longer crashes when provider payloads omit `totalTokens` or related usage fields. (#34977) thanks @sp-hk2ldn.
## 2026.3.2

View File

@@ -330,6 +330,131 @@ describe("sanitizeSessionHistory", () => {
expect(assistants[1]?.usage).toBeDefined();
});
it("adds a zeroed assistant usage snapshot when usage is missing", async () => {
vi.mocked(helpers.isGoogleModelApi).mockReturnValue(false);
const messages = castAgentMessages([
{ role: "user", content: "question" },
{
role: "assistant",
content: [{ type: "text", text: "answer without usage" }],
},
]);
const result = await sanitizeOpenAIHistory(messages);
const assistant = result.find((message) => message.role === "assistant") as
| (AgentMessage & { usage?: unknown })
| undefined;
expect(assistant?.usage).toEqual(makeZeroUsageSnapshot());
});
it("normalizes mixed partial assistant usage fields to numeric totals", async () => {
vi.mocked(helpers.isGoogleModelApi).mockReturnValue(false);
const messages = castAgentMessages([
{ role: "user", content: "question" },
{
role: "assistant",
content: [{ type: "text", text: "answer with partial usage" }],
usage: {
output: 3,
cache_read_input_tokens: 9,
},
},
]);
const result = await sanitizeOpenAIHistory(messages);
const assistant = result.find((message) => message.role === "assistant") as
| (AgentMessage & { usage?: unknown })
| undefined;
expect(assistant?.usage).toEqual({
input: 0,
output: 3,
cacheRead: 9,
cacheWrite: 0,
totalTokens: 12,
});
});
it("preserves existing usage cost while normalizing token fields", async () => {
vi.mocked(helpers.isGoogleModelApi).mockReturnValue(false);
const messages = castAgentMessages([
{ role: "user", content: "question" },
{
role: "assistant",
content: [{ type: "text", text: "answer with partial usage and cost" }],
usage: {
output: 3,
cache_read_input_tokens: 9,
cost: {
input: 1.25,
output: 2.5,
cacheRead: 0.25,
cacheWrite: 0,
total: 4,
},
},
},
]);
const result = await sanitizeOpenAIHistory(messages);
const assistant = result.find((message) => message.role === "assistant") as
| (AgentMessage & { usage?: unknown })
| undefined;
expect(assistant?.usage).toEqual({
...makeZeroUsageSnapshot(),
input: 0,
output: 3,
cacheRead: 9,
cacheWrite: 0,
totalTokens: 12,
cost: {
input: 1.25,
output: 2.5,
cacheRead: 0.25,
cacheWrite: 0,
total: 4,
},
});
});
it("preserves unknown cost when token fields already match", async () => {
vi.mocked(helpers.isGoogleModelApi).mockReturnValue(false);
const messages = castAgentMessages([
{ role: "user", content: "question" },
{
role: "assistant",
content: [{ type: "text", text: "answer with complete numeric usage but no cost" }],
usage: {
input: 1,
output: 2,
cacheRead: 3,
cacheWrite: 4,
totalTokens: 10,
},
},
]);
const result = await sanitizeOpenAIHistory(messages);
const assistant = result.find((message) => message.role === "assistant") as
| (AgentMessage & { usage?: unknown })
| undefined;
expect(assistant?.usage).toEqual({
input: 1,
output: 2,
cacheRead: 3,
cacheWrite: 4,
totalTokens: 10,
});
expect((assistant?.usage as { cost?: unknown } | undefined)?.cost).toBeUndefined();
});
it("drops stale usage when compaction summary appears before kept assistant messages", async () => {
vi.mocked(helpers.isGoogleModelApi).mockReturnValue(false);

View File

@@ -25,7 +25,12 @@ import {
} from "../session-transcript-repair.js";
import type { TranscriptPolicy } from "../transcript-policy.js";
import { resolveTranscriptPolicy } from "../transcript-policy.js";
import { makeZeroUsageSnapshot } from "../usage.js";
import {
makeZeroUsageSnapshot,
normalizeUsage,
type AssistantUsageSnapshot,
type UsageLike,
} from "../usage.js";
import { log } from "./logger.js";
import { dropThinkingBlocks } from "./thinking.js";
import { describeUnknownError } from "./utils.js";
@@ -200,6 +205,111 @@ function stripStaleAssistantUsageBeforeLatestCompaction(messages: AgentMessage[]
return touched ? out : messages;
}
function normalizeAssistantUsageSnapshot(usage: unknown) {
const normalized = normalizeUsage((usage ?? undefined) as UsageLike | undefined);
if (!normalized) {
return makeZeroUsageSnapshot();
}
const input = normalized.input ?? 0;
const output = normalized.output ?? 0;
const cacheRead = normalized.cacheRead ?? 0;
const cacheWrite = normalized.cacheWrite ?? 0;
const totalTokens = normalized.total ?? input + output + cacheRead + cacheWrite;
const cost = normalizeAssistantUsageCost(usage);
return {
input,
output,
cacheRead,
cacheWrite,
totalTokens,
...(cost ? { cost } : {}),
};
}
function normalizeAssistantUsageCost(usage: unknown): AssistantUsageSnapshot["cost"] | undefined {
const base = makeZeroUsageSnapshot().cost;
if (!usage || typeof usage !== "object") {
return undefined;
}
const rawCost = (usage as { cost?: unknown }).cost;
if (!rawCost || typeof rawCost !== "object") {
return undefined;
}
const cost = rawCost as Record<string, unknown>;
const inputRaw = toFiniteCostNumber(cost.input);
const outputRaw = toFiniteCostNumber(cost.output);
const cacheReadRaw = toFiniteCostNumber(cost.cacheRead);
const cacheWriteRaw = toFiniteCostNumber(cost.cacheWrite);
const totalRaw = toFiniteCostNumber(cost.total);
if (
inputRaw === undefined &&
outputRaw === undefined &&
cacheReadRaw === undefined &&
cacheWriteRaw === undefined &&
totalRaw === undefined
) {
return undefined;
}
const input = inputRaw ?? base.input;
const output = outputRaw ?? base.output;
const cacheRead = cacheReadRaw ?? base.cacheRead;
const cacheWrite = cacheWriteRaw ?? base.cacheWrite;
const total = totalRaw ?? input + output + cacheRead + cacheWrite;
return { input, output, cacheRead, cacheWrite, total };
}
function toFiniteCostNumber(value: unknown): number | undefined {
return typeof value === "number" && Number.isFinite(value) ? value : undefined;
}
function ensureAssistantUsageSnapshots(messages: AgentMessage[]): AgentMessage[] {
if (messages.length === 0) {
return messages;
}
let touched = false;
const out = [...messages];
for (let i = 0; i < out.length; i += 1) {
const message = out[i] as (AgentMessage & { role?: unknown; usage?: unknown }) | undefined;
if (!message || message.role !== "assistant") {
continue;
}
const normalizedUsage = normalizeAssistantUsageSnapshot(message.usage);
const usageCost =
message.usage && typeof message.usage === "object"
? (message.usage as { cost?: unknown }).cost
: undefined;
const normalizedCost = normalizedUsage.cost;
if (
message.usage &&
typeof message.usage === "object" &&
(message.usage as { input?: unknown }).input === normalizedUsage.input &&
(message.usage as { output?: unknown }).output === normalizedUsage.output &&
(message.usage as { cacheRead?: unknown }).cacheRead === normalizedUsage.cacheRead &&
(message.usage as { cacheWrite?: unknown }).cacheWrite === normalizedUsage.cacheWrite &&
(message.usage as { totalTokens?: unknown }).totalTokens === normalizedUsage.totalTokens &&
((normalizedCost &&
usageCost &&
typeof usageCost === "object" &&
(usageCost as { input?: unknown }).input === normalizedCost.input &&
(usageCost as { output?: unknown }).output === normalizedCost.output &&
(usageCost as { cacheRead?: unknown }).cacheRead === normalizedCost.cacheRead &&
(usageCost as { cacheWrite?: unknown }).cacheWrite === normalizedCost.cacheWrite &&
(usageCost as { total?: unknown }).total === normalizedCost.total) ||
(!normalizedCost && usageCost === undefined))
) {
continue;
}
out[i] = {
...(message as unknown as Record<string, unknown>),
usage: normalizedUsage,
} as AgentMessage;
touched = true;
}
return touched ? out : messages;
}
export function findUnsupportedSchemaKeywords(schema: unknown, path: string): string[] {
if (!schema || typeof schema !== "object") {
return [];
@@ -449,8 +559,9 @@ export async function sanitizeSessionHistory(params: {
? sanitizeToolUseResultPairing(sanitizedToolCalls)
: sanitizedToolCalls;
const sanitizedToolResults = stripToolResultDetails(repairedTools);
const sanitizedCompactionUsage =
stripStaleAssistantUsageBeforeLatestCompaction(sanitizedToolResults);
const sanitizedCompactionUsage = ensureAssistantUsageSnapshots(
stripStaleAssistantUsageBeforeLatestCompaction(sanitizedToolResults),
);
const isOpenAIResponsesApi =
params.modelApi === "openai-responses" || params.modelApi === "openai-codex-responses";