Advanced Configuration for Better Standardized Outputs through Simulated Interactions

[nyaoouo]

In order to improve the standardization of the output generated by the LLM, I propose we implement an advanced configuration that uses simulated interactions. This would allow for a more refined, context-aware generation of commit messages based on specific interaction patterns.

The intended operational flow is as follows:

system: [basic prompt]
user: [sample diff]
assistant: [sample commit message]
user: [real diff]

In this scenario, the LLM would then generate a commit message that is more accurately formatted according to the provided sample. This could potentially lead to increased efficiency in crafting commit messages that adhere to the preferred style of the individual or team, thereby improving overall project workflow.

Looking forward to your feedback.

[Blarc]

Hey @nyaoouo,

your suggestion makes a lot of sense. I will need to check how OpenAI API supports this. It would also probably make sense to preserve the old simple configuration alongside the new, because it's easier to use.

You're very welcome to dig in yourself and open a PR. Otherwise I'll try to look into it when I have time.

Thanks 🙏