A repository just for ideas, since I'm having them faster than I can finish them...

• A debugger where you select a couple of lines and a static analyzer determines and allows you to provide values for in-scope variables in normal code
• A command where you run it, e.g. my_tool ping www.google.com, and it pops up an interface that's constructed on the fly from the man page or online help. It shows a field for each variable along with the variable's name, information, etc. So you can run any commands from a GUI just by running the tool followed by the commands. It might even open up a row of screens for pipelines that would allow you to customize each. Then you click "OK" and it drops you back off at the command line with the altered command.
• A wiki that is just some additional Git porcelain (if this doesn't already exist). Just a few tools to handle automatically linking, e.g. wikilinks are translated to/from normal markdown link syntax, moving files works gracefully, etc. I think all of the logic could exist within individual files and all of the relevant data could exist in HTML comments within individual files that would be transparently added/removed.
• A reactive, semantic knowledge database runtime. Structured data (e.g. French verb conjugations, programming concepts) is stored relationally (SQL), transformed functionally (Lisp), interconnected and updated reactively (FRP/RxJS/ReactiveCocoa), and expressed declaratively (in a DSL), all within a JSON-compatible, type-safe, introspectable environment.