good practice

I review a lot of code from students, clients or colleagues and keep giving the same basic recommendations over and over again.
Now I collect important things here so I can easily point to them. Expect this to be expanded :).

source scripts

Before you send someone any code whatsoever, do the following:

CTRL+SHIFT+F10 to restart R with a clean workspace. Remember to have the Rstudio settings at Tools - Global Options - General set to: Restore .Rdata into workspace at startup OFF and Save workspace to .RData on exit NEVER
copy the script to some other folder than your working place
source() the entire script with CTRL+SHIFT+S
If something fails, fix it! E.g. stop messing with setwd (see below), copy data files to be read, rename all instances of a renamed object, …
Only now send the file (or folder with scripts, data files, etc. as zip file)

Rproj

Use Rstudio projects. They set the wd upon opening and keep projects separate.
Never use setwd() because others won’t have that exact path.
Even you yourself might not, after rearranging folders.
Use relative path names, e.g. read.table("datafolder/file.txt") instead of "C:/Users/berry/Desktop/Project/datafolder/file.txt"

packages

Follow the package recommendations in the packages section.

saveload

Store the results of long-running computations on disc.
The next time a script is run, they are loaded quickly.

if(  file.exists("objects.Rdata")  )
  {
  load("objects.Rdata") # load previously saved objects
  }  else
  {
  obj1 <-   mean(rnorm(2e7))             # in the first run,
  obj2 <- median(rnorm(2e7))             # compute the objects
  save(obj1, obj2, file="objects.Rdata") # and write them to disc
  }

If you need to rerun an analysis if the last run is older then 6 hours, this could be the condition:

difftime(Sys.time(), file.mtime("objects.Rdata"), units="h") > 6

For a single object, a good alternative to save and load is:

saveRDS(one_single_object, "object.Rdata")
explicit_object_name <- readRDS("object.Rdata")