I built an open-source framework for creating consistent workspace structures when working with AI coding assistants. It supports Gemini, Claude, and Codex.
The problem: AI assistants are great at generating code but provide no organizational structure. After a few months of "vibe coding," you end up with inconsistent project layouts, AI agents repeatedly asking "where is this file?", and invisible technical debt.
The solution: A tiered scaffolding system (Lite/Standard/Enterprise) that creates predictable patterns for both humans and AI.
What it does:
> Creates provider-specific config files (GEMINI.md, CLAUDE.md, or CODEX.md)
> Sets up standardized directory structures
> Includes reusable skills and workflows
> Adds Makefile-based session management
Usage: python bootstrap.py -t 2 -n myproject --provider claude
Why this matters: Same structure across all projects means lower cognitive overhead. AI agents recognize patterns and act faster. Code is written once but modified many times—structure helps with the modifications.
Single Python file (~5K lines), MIT licensed.
Has anyone else thought about sustainable patterns for AI-assisted development? Most content focuses on "build X in 10 minutes" but ignores what happens when you maintain that code for 6 months.
thomas-jamet•2h ago
The problem: AI assistants are great at generating code but provide no organizational structure. After a few months of "vibe coding," you end up with inconsistent project layouts, AI agents repeatedly asking "where is this file?", and invisible technical debt.
The solution: A tiered scaffolding system (Lite/Standard/Enterprise) that creates predictable patterns for both humans and AI.
What it does:
> Creates provider-specific config files (GEMINI.md, CLAUDE.md, or CODEX.md) > Sets up standardized directory structures > Includes reusable skills and workflows > Adds Makefile-based session management
Usage: python bootstrap.py -t 2 -n myproject --provider claude
Why this matters: Same structure across all projects means lower cognitive overhead. AI agents recognize patterns and act faster. Code is written once but modified many times—structure helps with the modifications.
Single Python file (~5K lines), MIT licensed.
Has anyone else thought about sustainable patterns for AI-assisted development? Most content focuses on "build X in 10 minutes" but ignores what happens when you maintain that code for 6 months.