When we talk about AI coding agents, most tools are great at “fill in this function” and pretty weak at “is this system actually well-designed?”. Most coding agents are optimized for “write code for this function”. That’s useful, but it leaves a gap at the level where the highest‑impact decisions are made: system boundaries, architecture trade-offs, integration risks, and “should we even build this?”. I’ve been working on a small but opinionated project: Meta-Architect, a set of Codex skills aimed at architecture, system design, and “should we actually build this?” questions. Is this architecture sane? Where are the weak boundaries? What are the trade-offs if we change X vs Y? Meta-Architect ships as a Codex skills bundle, not a separate CLI. Once installed, you stay in your existing Codex workflow and call dedicated skills for: - understanding your current system - exploring alternative designs - identifying risks and edge cases - turning vague requirements into clearer plans The v0.1.6 release focuses on making this practical: - clean package identity (@jstn-sdk/ma) - consistent install/uninstall + usage docs - skills centered on in-session Codex usage - packaging and tests wired so npm installs behave like running from source If you’re working with agents, complex systems, or multi-service architectures and want more structure than “chat + hope for the best”, I’d love your feedback.

Meta-Architect

Codex skills for architecture and system design

Codex skills for architecture and system design