How to set up Roo Code for code translation using prior files, example patterns, retrieval — and validate via TDD?

[yayayru]

Hi Roo Code team,

I'm using Roo Code for the first time and would like to configure it for a more advanced code translation workflow.

🔍 Goal:

I want to translate code not just based on simple source-to-target examples, but by leveraging previous files and similar code patterns across the project — even when those examples aren't direct pairs.

Essentially, I want to create a system that:

• Translates new code using retrieved examples from the codebase (like a RAG workflow).
• Maintains consistent structure and naming based on prior work.
• Validates the translated code using TDD (Test-Driven Development) — i.e., running tests after translation, fixing broken output, and continuing.

---

❓ What I'm looking for:

1. How can I set up Roo Code to use prior files and examples for translation?

   • Is this possible with the Memory Bank or should I integrate a retrieval system (e.g., vector DB)?
   • Can I define a custom Roo Mode that retrieves similar code, uses it in the prompt, and performs a translation?

2. How to structure a dataset of examples?

   • What's the recommended format — .yaml, .json, .md?
   • What metadata should be included for effective retrieval (e.g., language, tags, context)?
   • How should the examples be chunked (e.g., per function, class) for better embedding/search?

3. Can Roo Code support a TDD-like loop?

   • After translating a file, I want to automatically run its associated test(s).
   • If the test fails, the assistant should analyze the output and fix the code, repeating until it passes.
   • Is there any support or existing mode for this sort of feedback-correction loop?

---

If there are any config examples, YAML templates, or workflows for this kind of setup, I’d really appreciate being pointed in the right direction.

Thanks so much — really excited to work with Roo Code!