(I knew I'd written this up but couldn't find it...the reason was because it was a git commit message, not a forum post. Reprinting here.)

This is a fun usage of the new @-types, which lets you refer to Rebol values more easily in JS-DO and JS-EVAL expressions. By doing so you not only save on typing, but it finesses what would be hairy escaping situations if your only currency was text:

>> text: { ykcirT" a s'tI}

>> js-eval [{reb.Spell("reverse",} @text {) + "String" + '"'}]
== {It's a "Tricky String"}

So what's going on there is it's generating a temporary API handle to refer to the value fetched from TEXT. This API handle is then used literally as one of the arguments to the variadic/evaluating
extraction routines in the API (mixed here with some JavaScript + string concatenation). The variadic knows to free this handle after it is done processing it, (so it is like a reb.RELEASING() or reb.R()

This is more powerful than just treating a block as running JS-EVAL UNSPACED [...]. But if you want that, you can still use a plain GROUP!:

>> text: "gnirob"

>> js-eval [{"} (reverse text) { group"}]
== "boring group"

You can use @path/values, @(computed values). Although INTEGER! Rebol values aren't the best example (since you could pass them fine via their string representations), they'll work too (just easier to demo in a PR than Blobs for BINARY! and such...)

>> x: 1000
>> y: 20
>> js-eval [{reb.Unbox(} @(x + y) {)}]
== 1020

And to make things even more succinct, WORD!s in this dialect can be
used to generate the reb.XXX() calls for you:

>> x: 1000
>> js-eval [{20 +} unbox @x]
== 1020

Currently there's support for UNBOX and SPELL, but this could expand.

(Note: JS-DO and JS-EVAL are similar, except JS-EVAL tries to return a result to Rebol translated from the JavaScript it is given...and cannot declare variables in global scope. See technical notes in the implementation if interested.)