You can think of Shell as [] + IO + Managed. In fact, you can embed all three of them within a Shell:

select ::        [a] -> Shell a
liftIO ::      IO a  -> Shell a
using  :: Managed a  -> Shell a

Those three embeddings obey these laws:

do { x <- select m; select (f x) } = select (do { x <- m; f x })
do { x <- liftIO m; liftIO (f x) } = liftIO (do { x <- m; f x })
do { x <- with   m; using  (f x) } = using  (do { x <- m; f x })

select (return x) = return x
liftIO (return x) = return x
using  (return x) = return x

... and select obeys these additional laws:

select xs <|> select ys = select (xs <|> ys)
select empty = empty

You typically won't build Shells using the Shell constructor. Instead, use these functions to generate primitive Shells:

• empty, to create a Shell that outputs nothing
• return, to create a Shell that outputs a single value
• select, to range over a list of values within a Shell
• liftIO, to embed an IO action within a Shell
• using, to acquire a Managed resource within a Shell

Then use these classes to combine those primitive Shells into larger Shells:

• Alternative, to concatenate Shell outputs using (<|>)
• Monad, to build Shell comprehensions using do notation

If you still insist on building your own Shell from scratch, then the Shell you build must satisfy this law:

-- For every shell `s`:
_foldIO s (FoldM step begin done) = do
    x  <- begin
    x' <- _foldIO s (FoldM step (return x) return)
    done x'

... which is a fancy way of saying that your Shell must call 'begin' exactly once when it begins and call 'done' exactly once when it ends.