What is Meta-Architect?

JustineDevs — Sat, 02 May 2026 16:53:44 +0000

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?”.

Meta-Architect is live on GitHub.
Repo link: https://github.com/JustineDevs/meta-architect

What Meta-Architect is (and isn’t)
Meta-Architect:
runs as skills inside Codex
ships as a package/plugin, not a new shell
tries to help with architecture, not just local code edits
It’s designed to plug into the Codex ecosystem rather than competing with it.

Here is what you get.

What Meta-Architect is
Meta-Architect gives your AI agents a real architecture workflow, not loose prompts.

You get six named roles with strict responsibilities.

\$arch designs the system and explains the stack
\$sage pulls proven OSS patterns and libraries
\$flow maps user flows, state, and edge cases
\$vet reviews security and stops unsafe builds
\$vibe reviews developer experience and user experience
\$build turns the approved plan into an execution roadmap

Each role is a skill with a clear contract.

How it fits into your workflow

The idea is not to replace your existing tools.
Instead, Meta-Architect gives you a set of skills you can call when you need to:
understand the shape of a system
explore alternative designs
identify weak boundaries and risky assumptions
turn requirements into more structured plans
You stay inside Codex; Meta-Architect just adds a set of focused capabilities on top.

Why this exists
Most AI dev tools feel helpful at first, then fall apart when you try to ship.

They hallucinate stacks that never reach production
They ignore architecture, security, and release gates
They treat your repo as plain text instead of a system with rules and states

Meta-Architect goes the other way.
You treat the repo as the source of truth for skills, prompts, gates, packaging, and CI from v0.1.0 onward.

What lives in the repo
You are not downloading a bag of prompts.
You are getting a shaped repository.

Skills

skills/meta-architect-*.md holds the core skills for Arch, Sage, Flow, Vet, Vibe, Build
each skill defines a trigger, responsibilities, expected outputs, and gate impact

Prompts

stable role prompts for Architect, Sage, Flow, Security Reviewer, Builder, and others
written as contracts so your agents keep the same behavior across sessions

Templates

templates/AGENTS.md defines how triggers, gates, and evidence should behave together
templates/catalog-manifest.json is the catalog for skills and prompts

Missions

scenario folders like collaborative whiteboard, fintech app, SaaS dashboard
each mission describes constraints, expected gate changes, and what artifacts you should see

Plugin and publishing

plugins/meta-architect provides a bundle you can install as a plugin in compatible runtimes
scripts and npm commands handle skills:manifest, skills:validate, skills:pack, skills:install

Repo discipline

CONTRIBUTING plus issue forms keep contributions aligned with release safety
CI workflows run skills manifest, validate, pack, and install on each change
release and npm publish workflows treat your skills as versioned artifacts
checks prevent .omx runtime state from leaking into git

How you use this on a real project
Say your idea is:
“I want an AI powered smart contract security review workspace for solo founders.”

You run the skills in this order.

\$arch gives you system boundaries, stack choices, and clear risks
\$sage finds frameworks, auth, and infra from curated OSS lists
\$flow maps user journeys, data flow, and failure paths
\$vet reviews security and stops the build if status is not GREEN
\$vibe reviews onboarding, complexity, and usability
\$build turns the approved plan into repo layout, environment setup, and v0.1.0 milestones

You move from idea to gated plan in a repeatable way.

Who this helps
For developers

You keep control of architecture while offloading repetitive design work
You get a build plan that respects security and release practice

For new architects

You see how decisions, patterns, trade offs, and gates connect
You get concrete examples instead of vague theory

For experienced architects

Why share this now?
This release is “ready enough” for people who actually want to use it in anger and tell me where it breaks:
architecture folks who want more structure around design discussions
engineers experimenting with AI-assisted reviews or design sessions
anyone tired of manually repeating “here’s how our system works…” to an LLM
If you want to kick the tires:
👉 https://github.com/JustineDevs/meta-architect

You get a checklist that encodes your process into skills and CI
You can standardize how teams and agents reason about architecture

For maintainers and tool builders

You start from a repo pattern that already has skills, prompts, missions, MCP config, CI, and plugin bundling
You can adapt this pattern for your own skills library

What comes next
Near term work

Make the Architect skill behave like a small architecture handbook
Expand missions with deeper real world scenarios
Improve docs for skills publishing and MCP setup
Wire this into Codex style workflows so agents help maintain the repo under gates

Longer term work

Turn Meta-Architect into a base template for new repos that want AI aware architecture from day one

If this is the type of discipline you want around your AI workflows, review the repo, open an issue, or start a fork.
Repo link again: https://github.com/JustineDevs/meta-architect

Go-MiroFish, lightweight and local-first

JustineDevs — Sat, 25 Apr 2026 15:01:30 +0000

Go-Mirofish Fast local Go AI swarm engine – predict anything with agents

I needed a fast offline tool to test reactions to documents. I upload an earnings report or trading thesis or product announcement. I want to see how hundreds of different AI agents respond in a social simulation. No data leaves my machine. No slow cloud delays.So I built go-mirofish.You feed the tool any document. It builds a knowledge graph. It creates hundreds of AI agents. Each agent has its own personality. The agents run a full social simulation. You get a prediction report. You can chat with any agent to ask more questions.I rewrote the control plane in pure Go. Python no longer runs the hot path. The system now delivers sub 2 ms p50 latency. It handles 198 requests per second with zero errors. It runs on a standard laptop. It also runs on a Raspberry Pi 5.

Here is how you start in minutes: https://go-mirofish.vercel.app/docs

Home / entry	Simulation run
Report generation	Report timeline / tools
Simulation history	Deep interaction
Split: graph, workbench & system terminal	Graph view & node details

Thanks to these projects. MiroFish created the original swarm engine. MiroFish-Offline added the local English version. OASIS powers the multi-agent simulation. Neo4j manages the graph memory. Ollama runs the local models.I built go-mirofish for real work. Traders test market reactions and liquidation forecasts. Product teams run PR war rooms.
Teams practice cyber drills.

Writers explore alternate story endings.
Try the demo here.
https://go-mirofish.vercel.app

See the full source here.
https://github.com/go-mirofish/go-mirofishWhat
document will you test first. Tell me in the comments.

Forem: JustineDevs

What is Meta-Architect?

How it fits into your workflow

Go-MiroFish, lightweight and local-first