bug: Idiosyncratic instructions for the same action across prompts. #819

Problem Statement

Across the commands/prompts the LLM is given idiosyncratic instructions for the same task.

This is not conducive to consistent agent behavior across different commands/prompts.

Solution

Adopt consistent instructions for the same task.

• Consistent phrasing.
• Consistent and distinctive terminology, e.g. change-name. Specifically, we do not want the LLM to calculate that it is being instructed to alter the value of name.

Scope

• **Input**: <psuedo-random> should be: **Input**: The argument after ``/opsx:archive`` is "$@". This is the change-name.
  • opsx*apply command.
  • opsx*propose command. There is an OR qualifier, but before that the same instructions can be used.
  • opsx*archive command.
  • opsx*continue command.
  • opsx*explore command. This should be consistent with the opsx*propose command.
  • opsx*ff command. This should be consistent with the opsx*propose command.
  • opsx*new command. This should be consistent with the opsx*propose command.
  • opsx*sync command.
  • opsx*verify command.
• **Select the change**: make consistent across commands.
  • as above
• **Steps**: make consistent across commands.
  • as above

Related

#818