checking the command string cannot prevent unknown risks. It is recommended to start with the user / group executing the command and be a safe sandboxie.

I changed my approach,

after each query command, such as ps/ss, query, generally do the grep operation

We limit the types of commands that users can execute. For example, when a ps/ss, user post, select the type of command to execute.
and can only add the Filter keyword in a format similar to

<cmd> key1 key2 key3

check key1., in turn limits that key can only be words or contain a limited number of special characters. Splicing will not affect the system.
this kind of check is relatively simple. It can be determined by a rule.

as follows

word_re = re.compile(r'([a-zA-Z0-9\.]+)')

greps = []
for key in keys:
    matches = word_re.findall(key)
    -sharp key 
    if len(matches) == 0 or len(matches) > 1 or matches[0] != key:
        return send_msg(from_user, f': {key}')
    greps.append(f'grep -i {key}')

cmd = f'ps -ef | {"|".join(greps)}'

after splicing commands, the final execution format is as follows

ps -ef | grep key1 | grep key2 | grep key3

this makes it possible that the verification command will only query after the user post keyword and will not affect the system