fix: reject unknown subcommands on the host instead of forwarding them

[kapil-agnihotri]

The root command needs arbitrary args so code-vm -- <cmd> can forward a command to the guest, but cobra.ArbitraryArgs also handed any word it did not recognize as a subcommand straight into the VM. A typo, or a binary older than the subcommand being invoked, then surfaced as the guest shell's exec: profile: not found and exit 127 — after the VM had already been booted and entered — naming neither the real problem nor the host.

Validate args instead: ArgsLenAtDash reports how many args preceded -- (-1 when absent), so 0 is exactly the passthrough form. A bare invocation and code-vm -- <cmd> are unchanged; anything else fails on the host before the config is loaded, before Status, before boot.

[Copilot]

Pull request overview

This PR rejects unknown root-level arguments on the host while preserving explicit -- passthrough to the guest VM.

Changes:

• Added host-side argument validation.
• Added unit and integration coverage.
• Documented guest passthrough syntax.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Summary
test-vm-sandbox.sh	Adds an integration regression check.
README.md	Documents command forwarding behavior.
internal/cli/root.go	Rejects unknown commands before VM setup.
internal/cli/root_test.go	Tests rejection and explicit passthrough.

Suppressed comments (1)

internal/cli/root_test.go:90

• Could this test use a missing or invalid config instead of setupShellFixture? That helper calls loadConfig() itself and leaves a valid config in place, so the test only proves that no fake Lima calls occur; a regression that loads the config before rejecting the unknown word would still pass. Point configPath at a nonexistent config (and restore it) so this test also locks in the documented ordering before config loading.

	setupShellFixture(t)

	r := installFakeClient(t, "")

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.