Thanks for the targeted fix — the immediate pain of the invalid choice: 'hook' error is a real blocker and it's good to get the subcommand wired up.

A flag on the implementation: the current stub replaces hooks_cli.run_hook() with print statements and exits, which means hook run --hook stop now succeeds without actually persisting anything to the palace. The PR description references #328 — if I'm reading the history right, the concern in #328 was that the hook subcommand was missing, not that run_hook itself was broken.

The effect of this PR is that Claude Code's stop/save hooks will now exit 0 (and show a confirmation message) without writing to MemPalace. That's better than crashing, but it's a silent no-op — users will see success output and assume their session was saved when it wasn't.

Would it make sense to:

1. Keep the try/except wrapper (good defensive practice around the import), but fall through to calling run_hook() when it's available?
2. Only fall back to the no-op stub if hooks_cli isn't importable?

Something like:

def cmd_hook(args):
    try:
        from .hooks_cli import run_hook
        run_hook(hook_name=args.hook, harness=args.harness)
    except ImportError:
        print(f"[MemPalace Hook] hooks_cli not available, hook '{args.hook}' skipped.", file=sys.stderr)
        sys.exit(0)
    except Exception as e:
        print(f"[MemPalace Hook] Error: {e}", file=sys.stderr)
        sys.exit(1)

That way the subcommand exists (fixes the crash), uses the real implementation when available, and degrades gracefully if it isn't. Happy to be corrected if there's a reason run_hook itself is the problem here.