fix(x): surface coding tool errors

This commit is contained in:
Ramnique Singh 2026-07-06 10:56:23 +05:30
parent 36c0cca3f9
commit 4a2fe30153
11 changed files with 138 additions and 18 deletions

View file

@ -79,7 +79,7 @@ After \`code_agent_run\` returns:
- Pass through the agent's \`summary\` as-is. Do not rewrite it.
- Refer to file paths as plain text. Do NOT use \`\`\`file:path\`\`\` reference blocks. (This overrides the global "always wrap paths in filepath blocks" rule — for code-mode output, plain text.)
- Only add your own explanation if it failed:
- \`success: false\` with a message — surface the message. If it mentions the agent isn't installed or signed in, tell the user to install or sign in via **Settings → Code Mode**.
- A tool error with a message surface the message. If it mentions the agent isn't installed or signed in, tell the user to install or sign in via **Settings Code Mode**.
- \`stopReason: "cancelled"\` — the run was stopped; acknowledge briefly and ask if they want to continue.
---

View file

@ -0,0 +1,54 @@
import { afterEach, describe, expect, it, vi } from "vitest";
import container from "../../di/container.js";
import { InMemoryAbortRegistry } from "../../runs/abort-registry.js";
import { BuiltinTools } from "./builtin-tools.js";
import type { ToolContext } from "./exec-tool.js";
function context(signal: AbortSignal): ToolContext {
return {
runId: "turn-1",
toolCallId: "tool-1",
signal,
abortRegistry: new InMemoryAbortRegistry(),
publish: async () => {},
codePolicy: "ask",
};
}
function mockCodeServices(runPrompt: () => Promise<never>): void {
vi.spyOn(container, "resolve").mockImplementation(((name: string) => {
if (name === "codeModeManager") return { runPrompt };
if (name === "codePermissionRegistry") {
return { cancelRun: vi.fn(), request: vi.fn() };
}
throw new Error(`Unexpected dependency: ${name}`);
}) as typeof container.resolve);
}
describe("code_agent_run", () => {
afterEach(() => vi.restoreAllMocks());
it("throws genuine coding-agent failures for the runtime to mark as errors", async () => {
mockCodeServices(async () => {
throw new Error("spawn Electron ENOENT");
});
await expect(BuiltinTools.code_agent_run.execute(
{ agent: "codex", cwd: "/repo", prompt: "Fix it" },
context(new AbortController().signal),
)).rejects.toThrow("Coding agent failed: spawn Electron ENOENT");
});
it("returns an ordinary cancellation result when the turn was aborted", async () => {
const controller = new AbortController();
controller.abort();
mockCodeServices(async () => {
throw new Error("cancelled");
});
await expect(BuiltinTools.code_agent_run.execute(
{ agent: "codex", cwd: "/repo", prompt: "Fix it" },
context(controller.signal),
)).resolves.toMatchObject({ success: false, stopReason: "cancelled" });
});
});

View file

@ -868,7 +868,7 @@ export const BuiltinTools: z.infer<typeof BuiltinToolsSchema> = {
}),
execute: async ({ agent, cwd, prompt }: { agent: 'claude' | 'codex', cwd: string, prompt: string }, ctx?: ToolContext) => {
if (!ctx) {
return { success: false, message: 'code_agent_run requires run context (runId / streaming).' };
throw new Error('code_agent_run requires run context (runId / streaming).');
}
// The composer chip is the source of truth for the agent. The model's `agent`
// argument is only a fallback for the ask-human flow (code mode not active, no
@ -955,10 +955,7 @@ export const BuiltinTools: z.infer<typeof BuiltinToolsSchema> = {
changedFiles: [...changedFiles],
};
}
return {
success: false,
message: `Coding agent failed: ${error instanceof Error ? error.message : String(error)}`,
};
throw new Error(`Coding agent failed: ${error instanceof Error ? error.message : String(error)}`);
} finally {
ctx.signal.removeEventListener('abort', onAbort);
}