Files
llmx/codex-cli/tests/agent-terminate.test.ts
Michael Bolin ae5b1b5cb5 add support for -w,--writable-root to add more writable roots for sandbox (#263)
This adds support for a new flag, `-w,--writable-root`, that can be
specified multiple times to _amend_ the list of folders that should be
configured as "writable roots" by the sandbox used in `full-auto` mode.
Values that are passed as relative paths will be resolved to absolute
paths.

Incidentally, this required updating a number of the `agent*.test.ts`
files: it feels like some of the setup logic across those tests could be
consolidated.

In my testing, it seems that this might be slightly out of distribution
for the model, as I had to explicitly tell it to run `apply_patch` and
that it had the permissions to write those files (initially, it just
showed me a diff and told me to apply it myself). Nevertheless, I think
this is a good starting point.
2025-04-17 15:39:26 -07:00

181 lines
4.9 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import { describe, it, expect, vi } from "vitest";
// --- OpenAI stream mock ----------------------------------------------------
class FakeStream {
public controller = { abort: vi.fn() };
async *[Symbol.asyncIterator]() {
// Immediately ask for a shell function call so we can test that the
// subsequent function_call_output never gets surfaced after terminate().
yield {
type: "response.output_item.done",
item: {
type: "function_call",
id: "callterminate1",
name: "shell",
arguments: JSON.stringify({ cmd: ["sleep", "5"] }),
},
} as any;
// Turn completion echoing the same function call.
yield {
type: "response.completed",
response: {
id: "respterminate1",
status: "completed",
output: [
{
type: "function_call",
id: "callterminate1",
name: "shell",
arguments: JSON.stringify({ cmd: ["sleep", "5"] }),
},
],
},
} as any;
}
}
vi.mock("openai", () => {
class FakeOpenAI {
public responses = {
create: async () => new FakeStream(),
};
}
class APIConnectionTimeoutError extends Error {}
return { __esModule: true, default: FakeOpenAI, APIConnectionTimeoutError };
});
// --- Helpers referenced by handleexeccommand -----------------------------
vi.mock("../src/approvals.js", () => {
return {
__esModule: true,
alwaysApprovedCommands: new Set<string>(),
canAutoApprove: () =>
({ type: "auto-approve", runInSandbox: false } as any),
isSafeCommand: () => null,
};
});
vi.mock("../src/format-command.js", () => {
return {
__esModule: true,
formatCommandForDisplay: (cmd: Array<string>) => cmd.join(" "),
};
});
// Stub logger to avoid filesystem sideeffects
vi.mock("../src/utils/agent/log.js", () => ({
__esModule: true,
log: () => {},
isLoggingEnabled: () => false,
}));
// After dependency mocks we can import the modules under test.
import { AgentLoop } from "../src/utils/agent/agent-loop.js";
import * as handleExec from "../src/utils/agent/handle-exec-command.js";
describe("Agent terminate (hard cancel)", () => {
it("suppresses function_call_output and stops processing once terminate() is invoked", async () => {
// Simulate a longrunning exec that would normally resolve with output.
vi.spyOn(handleExec, "handleExecCommand").mockImplementation(
async (
_args,
_config,
_policy,
_additionalWritableRoots,
_getConf,
abortSignal,
) => {
// Wait until the abort signal is fired or 2s (whichever comes first).
await new Promise<void>((resolve) => {
if (abortSignal?.aborted) {
return resolve();
}
const timer = setTimeout(resolve, 2000);
abortSignal?.addEventListener("abort", () => {
clearTimeout(timer);
resolve();
});
});
return { outputText: "shouldnothappen", metadata: {} } as any;
},
);
const received: Array<any> = [];
const agent = new AgentLoop({
model: "any",
instructions: "",
config: { model: "any", instructions: "" },
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: (item) => received.push(item),
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),
onLastResponseId: () => {},
});
const userMsg = [
{
type: "message",
role: "user",
content: [{ type: "input_text", text: "run long cmd" }],
},
];
// Start agent loop but don't wait for completion.
agent.run(userMsg as any);
// Give it a brief moment to start and process the function_call.
await new Promise((r) => setTimeout(r, 10));
agent.terminate();
// Allow promises to settle.
await new Promise((r) => setTimeout(r, 50));
const hasOutput = received.some((i) => i.type === "function_call_output");
expect(hasOutput).toBe(false);
});
it("rejects further run() calls after terminate()", async () => {
const agent = new AgentLoop({
model: "any",
instructions: "",
config: { model: "any", instructions: "" },
approvalPolicy: { mode: "auto" } as any,
additionalWritableRoots: [],
onItem: () => {},
onLoading: () => {},
getCommandConfirmation: async () => ({ review: "yes" } as any),
onLastResponseId: () => {},
});
agent.terminate();
const dummyMsg = [
{
type: "message",
role: "user",
content: [{ type: "input_text", text: "noop" }],
},
];
let threw = false;
try {
// We expect this to fail fast either by throwing synchronously or by
// returning a rejected promise.
await agent.run(dummyMsg as any);
} catch {
threw = true;
}
expect(threw).toBe(true);
});
});