How did you learn to use GitHub Copilot CLI effectively for larger/multi-file projects?
I’ve had a GitHub Copilot Pro annual subscription since December, but so far I’ve mostly used it for basic autocomplete and chat inside PhpStorm and PyCharm.
Recently I started experimenting more with Copilot CLI for larger and multi-file tasks, but the results have been mixed. Sometimes it works surprisingly well, and other times it ignores instructions, changes unrelated code, or loses context.
I’m trying to understand how experienced users are getting consistently good results.
Some questions I’d really like feedback on:
- How did you learn to prompt effectively?
- What kinds of instructions produce the best results?
- How do you structure requests for multi-file changes?
- How do you keep Copilot focused and following instructions?
- Do you use project instruction files, conventions, or templates?
- What workflows made the biggest difference for you?
I’ve mostly used GPT-5 Mini and GPT-5.4 so far. Some of the other models have much higher usage multipliers, so I haven’t experimented much with them yet.
For people using the more expensive models:
- What tasks make them worth the higher multiplier cost?
- What do they do noticeably better?
- Are there specific prompting styles or workflows that help justify using them?
For context, most of my work is PHP, WordPress, Linux/server scripting, and general IT automation.