diff --git a/TODO.md b/TODO.md
new file mode 100644
index 0000000..a8850bd
--- /dev/null
+++ b/TODO.md
@@ -0,0 +1,31 @@
+# TODO
+
+* Fully audit the locker source to ensure that it cannot crash based on
+  error handling in its own code.
+* Add the (optional) ability to allow some key presses through, such as
+  screen brightness keys.
+* Add support for running screensaver programs ("hacks" in
+  `xscreensaver` parlance) that draw interesting things on
+  `bscreensaver`'s blanker windows.  Without this, `bscreensaver` is
+  more accurately called `bscreenlocker`.
+* Dig into the systemd service a bit more; I think it isn't always
+  locking properly before the system suspends (sometimes there is a
+  flash of screen contents on resume).
+* Build a second locker dialog using another toolkit.  Qt could be
+  interesting, as well as iced.  I do not believe it is possible to
+  use GTK4, as GTK4 no longer has `GtkPlug` and is intentionally too
+  platform-agnostic to support the kind of things we need.
+* Move the `xcb-xembed` and `xcb-async` sub-crates to their own project.
+  I may want to flesh out `xcb-xembed` so it supports the entirety of
+  the XEMBED spec, not just the features I needed for this project.
+* Consider using `x11rb` rather than `xcb`.  The former autogenerates a
+  full XCB protocol library in Rust, while the latter is a set of
+  bindings to the `libxcb` C library.  I would like to reduce the amount
+  of non-Rust code executed in the locker process, for safety/security
+  reasons.
+* Add support for non-Linux systems that are also UNIX-like.  PAM, I
+  believe, is different on other systems and the auth code might need
+  some changes.  Also, a method of disabling building, installing, and
+  running the systemd, and possibly DBus, services would be needed.
+* Figure out why forwarding key release events (in addition to key press
+  events) to the unlock dialog results in duplicate key presses.