Skip to content

PRD.md

Latest commit

 

History

History
67 lines (46 loc) · 6.24 KB

PRD.md

File metadata and controls

67 lines (46 loc) · 6.24 KB

📘 PRODUCT REQUIREMENT DOCUMENT (PRD)

🧱 Epic: GPT-Aware Middleware for GitHub Codebase Scanning and Writing

🎯 Goal

Enable custom GPT agents (e.g., The Architect, The Polyglot) to:

  • Autonomously read, navigate, write to, and scan a live GitHub repository.
  • Fetch files, recursively crawl directory trees, and access raw content.
  • Write to branches by updating or creating files.
  • Assess system completeness, inconsistencies, missing layers, unused code, etc.
  • Generate accurate PRDs, TECH_SPECs, and Codex-ready tasks using real source code as context.

💡 Why It Matters

Current GPT agents depend on manual context injection (copy/paste) or limited summaries. That:

  • Causes blind spots and inconsistencies.
  • Prevents deep architectural reasoning.
  • Adds overhead for the human orchestrator (you).

To act like software engineers, GPTs need a shared, live source of truth (the real repo) and the ability to propose or apply changes directly.

🧰 Features (Functional Scope)

Feature Description Status
Recursive File Tree Scan Read and list all files under a folder (depth N). ✅ Shipped
Read File Contents Base64-decode and return raw content of any file. ✅ Shipped
Write File Contents Update or create a file in a specified branch. ✅ Shipped
Plugin Manifest & OpenAPI Provides ai-plugin.json and openapi.yaml for GPT Actions ✅ Shipped
Planned: Detect Module Boundaries Identify domain modules from paths. 🚧 Planned
Planned: Track Missing Interfaces Compare usecase interfaces to existing implementations. 🚧 Planned
Planned: Scan for Project Health Report dead code, stubs, TODOs, broken layers. 🚧 Planned
Planned: Enable GPT Planning GPTs generate next features/tasks directly from structure. 🚧 Planned

Features marked Planned are for future versions.

🧑‍💻 Users

Role How They Use It
The Architect Loads full backend structure to assess usecase/repo alignment.
The Polyglot Loads frontend trees and schemas for UI state/data structure planning.
Codex Uses tasks written by GPTs based on real repo data.
All GPTs Can propose or commit changes by writing/updating files safely.

🔒 Constraints

  • Must use authenticated GitHub API via personal token.
  • All writes must target non-protected branches (e.g., gpt-agent/*).
  • Must not expose GitHub PAT to GPTs or frontend.
  • Should cache or paginate results to handle large repos.
  • Optional: Require commit message templates from GPTs.