I am developing a repo-native agent app for analyst-grade answers over complex private file.

The initial idea is: I want codex could help me to deep dive digest all my work files (tons of ppt, excel & doc), and then help me to design IT architecture, prepare slides outline for top management, or write a full report for company own huge IT platform for multi purpose (on-boarding or showcase)

I have developed an auto sync. Knowledge base in Codex and a set of skills + scripts. Then I realized, this could be a repo with an AGENTS.md to manage all pipelines.

Now all you need to do is download the repo, open it in Codex, and start chatting with your private working files.

I call this repo pattern is: The repo is the app. Codex is the runtime.

What do you think about this pattern? Maybe it could be the future :) due to we could fully leverage the tokens and AI Agent capability of Codex/Claude Code more than just coding.