It depends on what you want to do. For operations within
the specific tool set, the internal scripting language probably
offers more power over data objects with less effort (and,
perhaps, a decent base of reusable code and mid-level
scripts). But for working between two different vendors'
tools (or vendor and user) a UNIX filesystem manipulation
of data may be the only way to go (unless you want to
teach SKILL to play nice with whatever Mentor uses, etc.
and use UNIX file I/O to move data still).
Now I have seen useful scripting done from Cadence menus
modified to involve PERL scripts (like Spectre netlist
optimization, scanning and replacing multiple parallel FET
instances with one single that matches total geometry,
in the interest of solution convergence and speed).
Of course the right answer from the hiring party's point
of view is "all of the above, ready to rock, get busy coding".
I'd declare that plan and proceed with any of these as a
first step along the lifelong learning journey. Interviewers
like to see a plan as much as a legacy (as long as that
plan looks useful and nonthreatening).