[Bug] kimi upgrade exits 0 without upgrading in non-TTY environments — automation cannot detect the no-op

Summary

In non-TTY environments (piped stdin/stdout), kimi upgrade prints the manual-update message and exits with code 0 without performing any update. Automation (CI scripts, desktop apps, wrappers) that only checks the exit code cannot distinguish "upgrade succeeded" from "upgrade was a no-op", and will silently report success while the CLI is still on the old version.

Environment

• OS: macOS (darwin-arm64)
• Install method: official script (~/.kimi-code/bin, native install)
• Version: 0.31.1 -> 0.32.0

Steps to reproduce

kimi upgrade < /dev/null        # stdin/stdout not a TTY

Output:

A newer version of @moonshot-ai/kimi-code is available (0.31.1 -> 0.32.0).
Detected install source: native (windows). Auto-update is not supported on this platform.
To update manually, run: curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash

echo $?   # → 0, but `kimi --version` is still 0.31.1

For comparison, under a PTY (script -q /dev/null kimi upgrade) the CLI works as expected: it shows the interactive update menu (Source: native, Command: curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash) because canAutoInstall("native", "darwin") returns true.

Root cause

apps/kimi-code/src/cli/sub/upgrade.ts: the manual branch is taken when !canAutoInstall(...) || !isInteractive, where isInteractive = process.stdin.isTTY && process.stdout.isTTY. For piped/automation calls isInteractive is false, so the CLI always falls into the manual-message branch and returns 0.

Note this is different from (and not covered by) #2391: install-source detection is not broken on macOS — the misleading native (windows) text is just a hardcoded template in renderManualUpdateMessage. The real blocker for automation is the TTY gate combined with exit code 0.

Expected behavior

For automation, one of the following would work:

1. A non-interactive flag (e.g. kimi upgrade --yes) that runs the install without prompting, or
2. A non-zero exit code (or machine-readable output) when the command did not actually update, so callers can detect the no-op.

Impact

Any tool that invokes kimi upgrade programmatically (CI pipelines, desktop apps like myTauri app, package managers) will falsely report "upgrade completed" while the CLI remains outdated — with no error surfaced to the user.

Related: #2391 (misleading platform text in the manual message), #2409 (text-only fix; does not address the non-TTY exit-code behavior).

Start in apps/kimi-code/src/cli/sub/upgrade.ts by reading the manual branch, canAutoInstall, and the isInteractive check. Reproduce kimi upgrade with stdin redirected from /dev/null, then define the non-interactive behavior so callers can distinguish an actual update from a no-op. Done means automation no longer silently reports success when the version remains unchanged.