What loopholes should be assumed when a name-specific file is used for code golf?

Question

I have posted this golfing tip for Bash before:

In Pure Bash (or any other pure shellscript), use . for looping.

It's a tip that recomments using a file whose the program has. I have ever posted these answers that uses the tip:

• Implement a Truth-Machine
• 1, 2, Fizz, 4, Buzz
• Swap every two elements in a list
• Tips for golfing in Bash
• Print 0 to 100 without 1-9 characters
• Uncomment a COBOL program!

But I have thought that these situations can be thought:

• The program's file is placed on / (root directory), which prevents from using cd / to end the loop.
• The program's file is read-only, which prevents from using >x to end the loop.
• The program is invoked from different directory, such as invoked with bash /path/to/x, bash ../../x
• The program's directory has several files such as bash, zsh, yes, rm -Rf / that can have arbitrary data such as different shell script so invoking things such as for x in *; do ...; done, *, eval *, . * results in several behaviours.
• The shell script is invoked with the flag such as -u and -e.

What assumptions must be given for such programs?