If any executable can be used as fixer, the fixer could also run as service to be communicated with via sockets:

• Unix sockets (--fix filename) (use -S for test) - see IO::Socket
• TCP Sockets (--fix tcp://server:port) - see IO::Socket::INET
• Webservices (--fix http://server:port/path) - see HTTP::Tiny or any other HTTP client module
• WebSockets (--fix ws://server:port/path) - see Protocol::WebSocket::Client

Sure serializing JSON, transmission, and parsing JSON takes some time but one could use transformation services as fixes independent from Catmandu. For instance a WebSocket fixer could also be used in a JavaScript web application.

Using arbitrary executables as fixes is very useful because users of Catmandu can use any programming language of their choice. For this reason Catmandu::Fix::cmd should be part of Catmandu core, but without adding more possibly instable dependencies. Maybe this module can be modified to not use IO::Pty::Easy?

By environment variable or catmandu config value to ensure that only trusted executables are called.

validate by passing JSON (or another format) to an external program and check its return value.

If myscript is an executable, the Catmandu command line client should support this:

catmandu convert CSV --fix myscript to YAML < input.csv

to actually do this:

catmandu convert CSV --fix "cmd('myscript')" to YAML < input.csv

I already raised the issue at LibreCat/Catmandu#68

As far as I understand the code has to be modified here: https://github.com/LibreCat/Catmandu/blob/master/lib/Catmandu/Env.pm#L139

I would not add Catmandu::Fix::cmd to the core of Catmandu unless it uses a more stable module (IO::Pty::Easy is useful but has many failures on CPANTS).