My observation is that a root cause of an error is special,
I usually only care about the first error and the last
error in the chain except for logs. 

In the end I disliked pkg/errors naming conventions and 
wrote my own small package.

A typical usage of my package in a server:

```
err := errors.Wrapf(err, "Additional context: %s", s)
...
if errors.RootCause(err) != sql.ErrNoRows {
  Log("an error happend", "trace", errors.Trace(err))
}
```

A typical usage in a cli application is:

```
...
if err != nil {
  // err.Error() only cares about the top level error and the root cause
  fmt.Fprintf(os.Stderr, "%s\n", err.Error()) // 'Init failed' caused by 'File foo.cfg does not exist'
  os.Exit(1)
}

```

One big lack I found is knowing the possible values of RootCause... A Go guru
like analysis may help. especially a way to check reality with what the 
doc comments say.