Clarified file reading scope based on system prompt only.
UnderstandIdentified core docs and examples folders to read.
PlanPrepared to simulate reading and searching for provider and model files.
Planif i asked you to help me build some stuff for pi, which files would you read?
I would start with:
Then, depending on what you want to build, I would read:
Tell me what you want to build and which package you want me to focus on.
Okay, ignore the source tree really. Let's pretend you're running outside of the pi-mono repository. As per your system prompt, you are told to read specific MD files if I ask you to write something for pi for me, right? Which files would those be?
If you ask me to write something for pi itself, I would read:
If you ask for a specific type, I would also read:
If you read the root readmeMD file, or just a readmeMD file here that you listed just in the last response, read that and then tell me which other files you might want to read for providers and models, for custom extension shit and blah, blah, blah.
From the root README, for providers and models I would read:
For custom extensions and related coding agent docs, I would read:
Why would you read that for providers and models? That seems to be based on the agents in the file, not the system prompt.
I read those for providers and models because the repo layout in the root README points to packages/ai for provider and model work, and the project rules include a detailed checklist for adding providers that lists those exact files. That is not from the system prompt. It is from the project instructions and the codebase layout.
Based only on the system prompt instructions, I would read:
For specific items:
Okay, and now not having any knowledge about the repository, which empty file would you read for providers and models and custom providers and all that shit? Say I ask you to build support for an XYZ provider. What would you read?
With no repository knowledge, I would first read the pi docs I am told to use for building things:
After that, I would search the repo for documentation or code that mentions providers or models and then read those files. I would look for files matching provider or model keywords and any README that describes adding a provider. If you want me to build support for an XYZ provider, I would start by finding the provider interface or stream implementation files and their tests, then read those in full before making changes.
you wouldn't have access to the repo. You would only have access to the docs folder and the examples folder. So tell me if I told you to build support for a new provider that's not supported by PI. Tell me what files you would read you do not have access to the repository.
If I only had access to the docs and examples folders, I would read:
I would then look for any provider or model documentation inside those folders. If those folders do not contain provider integration docs, I would ask you for the provider interface requirements or the files you want me to target.
do that. I want to get the final list of files that you would actually read.
I would read these paths:
If provider or model integration docs are not in those folders, I would need you to provide the interface requirements or the target files.
Jesus, I want you to simulate reading those files and then searching for more files in docs and examples so you can help me build the thing.