Skip to content

context.md

Latest commit

 

History

History
15 lines (11 loc) · 890 Bytes

context.md

File metadata and controls

15 lines (11 loc) · 890 Bytes

Context: Memory Layer (RAG & Patterns)

Location: universal-refiner/src/memory/

Purpose

This module provides the "Long-term Memory" for the refiner. it uses RAG (Retrieval Augmented Generation) to inject project-specific code snippets and patterns into prompts.

Key Files

  • neural-snippets.ts: Implements the RAG system. Searches the local codebase for relevant snippets.
  • local-brain.ts: Manages persistent project-level engineering mandates and learned patterns.

Module Instructions

  1. Snippet Precision: Prioritize high-signal files (main entry points, core logic) over boilerplate when retrieving snippets.
  2. Pattern Persistence: All patterns learned via discover_rules must be saved to local-brain.ts.
  3. Consistency: Ensure injected mandates match the current project's GEMINI.md where possible.