test: reproduce #4285 — broken default agent commands for Claude and Codex

packages/shared/src/builtin-terminal-agents.test.ts

@@ -0,0 +1,33 @@
+import { describe, expect, test } from "bun:test";
+import { BUILTIN_TERMINAL_AGENT_COMMANDS } from "./builtin-terminal-agents";
+
+describe("issue #4285: out-of-the-box defaults must produce a working agent", () => {
+	test("claude default launches in auto mode", () => {
+		const [command] = BUILTIN_TERMINAL_AGENT_COMMANDS.claude;
+		// Per the issue: users expect Claude's default to be "auto mode" so
+		// the agent can run without per-action permission prompts. The two
+		// flags that disable prompting are `--dangerously-skip-permissions`
+		// and `--permission-mode bypassPermissions`. `acceptEdits` only
+		// auto-accepts file edits — it still prompts before shell commands,
+		// which the issue reports as broken-on-first-run behavior.
+		const enablesAutoMode =
+			command.includes("--dangerously-skip-permissions") ||
+			command.includes("--permission-mode bypassPermissions");
+		expect(enablesAutoMode).toBe(true);
+	});
+
+	test("codex default does not combine workspace-write sandbox with `--ask-for-approval never`", () => {
+		const [command] = BUILTIN_TERMINAL_AGENT_COMMANDS.codex;
+		// Per the issue: this exact pairing produces "sandbox doesn't allow
+		// me to run terminal command" errors. The workspace-write sandbox
+		// blocks shell operations needing network or out-of-workspace access,
+		// and `--ask-for-approval never` removes the agent's ability to
+		// escalate — so codex silently refuses instead of asking the user.
+		// Working alternatives: `--full-auto` (workspace-write + on-failure),
+		// or `--dangerously-bypass-approvals-and-sandbox` for full auto mode.
+		const isBrokenCombination =
+			command.includes("--sandbox workspace-write") &&
+			command.includes("--ask-for-approval never");
+		expect(isBrokenCombination).toBe(false);
+	});
+});