[Agent]: Minimal Coding Agent — Best Practices, Fundamentals, Security, Minimalism

[julia-meshcheryakova]

Chat Mode/Agent Name

MinimalCodingAgent

Type

Custom Agent

Purpose

Enable contributors to use a minimal coding agent (native Github custom agent) that enforces engineering excellence, security, and best practices, but only produces the smallest functional code avoiding overengineering and unnecessary documentation, editing/adding only the necessary files.

Requirements

• Implement logic/code minimally and strictly to requirements
• Enforce repository conventional style and security rules
• Prohibit extra comments/documentation except for critical notes
• Discourage unnecessary abstractions, backwards compatibility, or non-essential features
• Is native GitHub custom agent

[julia-meshcheryakova]

I had positive experience with Custom agent defined as simple as:

You are a Python coding assistant for ML microservices. Generate minimal, working code only—no unnecessary abstractions. Focus on clarity and functionality.
Guidelines:
Data ingestion: load large datasets efficiently. Use batching if necessary. Handle missing values gracefully.
Embeddings: generate vector embeddings using OpenAI. Use efficient batching for large datasets.
Storage & retrieval: store embeddings for semantic search (in-memory, SQLite, or vector DB). Provide retrieval functions.
Inference: create functions or endpoints that take a natural language query, compute embeddings if needed, and return relevant items.
FastAPI: create minimal endpoints for ingestion, embedding creation, and inference. Keep endpoints clear and concise.
Streamlit: create thin UI to call FastAPI endpoints. Include text input, submit button, and result display.
Jupyter notebooks: generate small, reproducible EDA code using pandas and matplotlib/seaborn. Include basic statistics, visualizations, and missing value analysis.
Configuration: read API keys and other secrets from .env.
Python style: code should be modular, testable, Pythonic, and suitable for a senior Python engineer.
Minimalism: avoid over-engineering, extra classes, or unnecessary boilerplate. Comment only where clarity is needed.

But for the sake of this repo it needs to be refined to be generic enough.

[WilliamBerryiii]

@julia-meshcheryakova - hang tight on this because I do think there's a place for a minimal RPI Framework flow ... something between "Ask -> Ask -> Agent" and Task Researcher/Planner/Implementor. I'm still trying to think this through, so welcome any insights you have on easing the bar to entry to HVE Core.

[WilliamBerryiii]

@julia-meshcheryakova - circling back on this! We've since built out the RPI Agent which provides evidence-backed implementation with subagent delegation for complex tasks.

The minimalism principles you outlined are also now embedded in our base copilot-instructions.md:

• No extra tests/scripts/docs unless explicitly requested
• Backward-compatibility layers only when explicitly requested
• Comments remain brief and factual - no narrative or step-by-step reasoning

Does the RPI Agent workflow cover what you were envisioning, or is there still a gap for a lighter-weight "just code it" agent that skips the research/planning phases entirely?