Skip to content

Commit fa5c834

Browse files
steipetesteipete
committed
test: isolate Codex terminal diagnostic fallback
1 parent f603fa5 commit fa5c834

2 files changed

Lines changed: 75 additions & 91 deletions

File tree

‎extensions/codex/src/app-server/run-attempt.test.ts‎

Lines changed: 74 additions & 91 deletions
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,6 @@ import {
99
onAgentEvent,
1010
queueAgentHarnessMessage,
1111
resetAgentEventsForTest,
12-
wrapToolWithBeforeToolCallHook,
1312
type AgentEventPayload,
1413
type EmbeddedRunAttemptParams,
1514
} from "openclaw/plugin-sdk/agent-harness-runtime";
@@ -3366,112 +3365,96 @@ describe("runCodexAppServerAttempt", () => {
33663365
});
33673366

33683367
it("emits request-boundary terminal diagnostics when a wrapped dynamic tool does not", async () => {
3369-
const harness = createStartedThreadHarness();
33703368
const diagnosticEvents: DiagnosticEventPayload[] = [];
33713369
const unsubscribeDiagnostics = onInternalDiagnosticEvent((event) =>
33723370
diagnosticEvents.push(event),
33733371
);
3374-
const rawTool = {
3375-
name: "echo",
3376-
description: "echo test tool",
3377-
parameters: {
3378-
type: "object",
3379-
properties: {},
3380-
additionalProperties: false,
3381-
},
3382-
execute: vi.fn(async () => ({
3383-
content: [{ type: "text" as const, text: "echo done" }],
3384-
details: {},
3385-
})),
3386-
};
3387-
rawTool.execute.mockImplementationOnce(async () => {
3388-
emitTrustedDiagnosticEvent({
3389-
type: "tool.execution.completed",
3390-
runId: "other-run",
3391-
sessionId: "session-1",
3392-
sessionKey: "agent:main:session-1",
3393-
toolName: "echo",
3394-
toolCallId: "call-echo-unobserved-terminal",
3395-
durationMs: 1,
3396-
});
3397-
return {
3398-
content: [{ type: "text" as const, text: "echo done" }],
3399-
details: {},
3400-
};
3401-
});
3402-
const markedWrappedTool = {
3403-
...wrapToolWithBeforeToolCallHook(rawTool as never),
3404-
execute: rawTool.execute,
3405-
};
3406-
testing.setOpenClawCodingToolsFactoryForTests(() => [markedWrappedTool as never]);
3407-
3408-
const params = createParams(
3409-
path.join(tempDir, "session.jsonl"),
3410-
path.join(tempDir, "workspace"),
3411-
);
3412-
params.disableTools = false;
3413-
params.runtimePlan = createCodexRuntimePlanFixture();
3414-
3415-
const run = runCodexAppServerAttempt(params);
3416-
await harness.waitForMethod("thread/start");
3417-
3418-
const toolResult = (await harness.handleServerRequest({
3419-
id: "request-echo-unobserved-terminal-tool",
3420-
method: "item/tool/call",
3421-
params: {
3372+
try {
3373+
const call = {
34223374
threadId: "thread-1",
34233375
turnId: "turn-1",
34243376
callId: "call-echo-unobserved-terminal",
34253377
namespace: null,
34263378
tool: "echo",
34273379
arguments: {},
3428-
},
3429-
})) as {
3430-
contentItems?: Array<{ text?: string; type?: string }>;
3431-
success?: boolean;
3432-
};
3433-
expect(toolResult.success).toBe(true);
3434-
3435-
await harness.completeTurn({ threadId: "thread-1", turnId: "turn-1" });
3436-
await run;
3437-
await flushDiagnosticEvents();
3438-
unsubscribeDiagnostics();
3380+
} satisfies CodexDynamicToolCallParams;
34393381

3440-
const toolDiagnosticEvents = diagnosticEvents.filter(
3441-
(
3442-
event,
3443-
): event is Extract<
3444-
DiagnosticEventPayload,
3445-
{ type: "tool.execution.started" | "tool.execution.completed" | "tool.execution.error" }
3446-
> => event.type.startsWith("tool.execution."),
3447-
);
3448-
expect(
3449-
toolDiagnosticEvents.map((event) => ({
3450-
runId: event.runId,
3451-
type: event.type,
3452-
toolName: event.toolName,
3453-
toolCallId: event.toolCallId,
3454-
})),
3455-
).toEqual([
3456-
{
3382+
emitDynamicToolStartedDiagnostic({
3383+
call,
34573384
runId: "run-1",
3458-
type: "tool.execution.started",
3459-
toolName: "echo",
3460-
toolCallId: "call-echo-unobserved-terminal",
3461-
},
3462-
{
3463-
runId: "other-run",
3385+
sessionId: "session-1",
3386+
sessionKey: "agent:main:session-1",
3387+
});
3388+
emitTrustedDiagnosticEvent({
34643389
type: "tool.execution.completed",
3390+
runId: "other-run",
3391+
sessionId: "session-1",
3392+
sessionKey: "agent:main:session-1",
34653393
toolName: "echo",
34663394
toolCallId: "call-echo-unobserved-terminal",
3467-
},
3468-
{
3395+
durationMs: 1,
3396+
});
3397+
expect(
3398+
testing.hasPendingDynamicToolTerminalDiagnostic({
3399+
call,
3400+
runId: "run-1",
3401+
sessionId: "session-1",
3402+
sessionKey: "agent:main:session-1",
3403+
}),
3404+
).toBe(false);
3405+
3406+
emitDynamicToolTerminalDiagnostic({
3407+
call,
34693408
runId: "run-1",
3470-
type: "tool.execution.completed",
3471-
toolName: "echo",
3472-
toolCallId: "call-echo-unobserved-terminal",
3473-
},
3474-
]);
3409+
sessionId: "session-1",
3410+
sessionKey: "agent:main:session-1",
3411+
durationMs: 1,
3412+
response: {
3413+
success: true,
3414+
contentItems: [{ type: "inputText", text: "echo done" }],
3415+
},
3416+
});
3417+
3418+
await flushDiagnosticEvents();
3419+
3420+
const toolDiagnosticEvents = diagnosticEvents.filter(
3421+
(
3422+
event,
3423+
): event is Extract<
3424+
DiagnosticEventPayload,
3425+
{ type: "tool.execution.started" | "tool.execution.completed" | "tool.execution.error" }
3426+
> => event.type.startsWith("tool.execution."),
3427+
);
3428+
expect(
3429+
toolDiagnosticEvents.map((event) => ({
3430+
runId: event.runId,
3431+
type: event.type,
3432+
toolName: event.toolName,
3433+
toolCallId: event.toolCallId,
3434+
})),
3435+
).toEqual([
3436+
{
3437+
runId: "run-1",
3438+
type: "tool.execution.started",
3439+
toolName: "echo",
3440+
toolCallId: "call-echo-unobserved-terminal",
3441+
},
3442+
{
3443+
runId: "other-run",
3444+
type: "tool.execution.completed",
3445+
toolName: "echo",
3446+
toolCallId: "call-echo-unobserved-terminal",
3447+
},
3448+
{
3449+
runId: "run-1",
3450+
type: "tool.execution.completed",
3451+
toolName: "echo",
3452+
toolCallId: "call-echo-unobserved-terminal",
3453+
},
3454+
]);
3455+
} finally {
3456+
unsubscribeDiagnostics();
3457+
}
34753458
});
34763459

34773460
it("does not duplicate terminal diagnostics for wrapped dynamic tool blocks", async () => {

‎extensions/codex/src/app-server/run-attempt.ts‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5644,6 +5644,7 @@ export const testing = {
56445644
shouldForceMessageTool,
56455645
shouldReleaseTurnAfterTerminalDynamicTool,
56465646
resolveTerminalDynamicToolBatchAction,
5647+
hasPendingDynamicToolTerminalDiagnostic,
56475648
buildCodexPluginThreadConfigEligibilityLogData,
56485649
withCodexStartupTimeout,
56495650
setOpenClawCodingToolsFactoryForTests(factory: OpenClawCodingToolsFactory): void {

0 commit comments

Comments
 (0)