Stop hand-writing AI workflow config for every new project.

Adopting a structured AI coding workflow on a new project means hand-authoring rules, memory-bank documentation, and configuration files — across whatever stack the team picked. RooCode Generator scans the project, infers context, and asks an LLM to draft the rules, memory bank, and Cursor-style configs the project actually needs. One interactive CLI, multiple LLM providers, consistent output across every stack.

The challenge

Generating useful, project-specific AI workflow configuration without locking the user into a single tech stack — and without hallucinating rules that don't match the actual codebase.

The outcome

A single CLI replaces hours of hand-authoring RooCode rules, memory-bank docs, and Cursor configs for new projects — using LangChain-powered LLM calls grounded in real workspace analysis.

Technical approach

• Interactive Commander-based CLI — `generate` and `config` commands with progress indicators (ora) and chalk logging
• Modular generator architecture — AiMagicGenerator dispatches to memory-bank, roo, or cursor sub-generators
• Project context analyzer — scans the workspace and feeds normalised metadata into prompts
• LangChain LLM integration — pluggable OpenAI, Anthropic, and Google GenAI providers via configurable adapters
• Memory-bank generation — produces structured project documentation grounded in the analyser output
• Result-type error handling — explicit success/failure objects instead of thrown exceptions
• DI container architecture — modular service registration in @core/di/modules for testability
• Template-driven generation — file templates rendered with project-aware substitutions
• Configuration management — roocode-config.json and llm.config.json drive deterministic regeneration

Results at a glance

★ 13

GitHub Stars

OpenAI · Anthropic · Google

LLM providers

Any (analyser-driven)

Tech stack