Get an AI coding agent to produce an approved design before it touches any code.
Turn an approved design into a sealed, task-by-task execution plan.
Execute a large refactor or new feature in bounded, logged steps instead of one big diff.
Prevent an agent from inventing speculative abstractions for features that do not exist yet.
| vortezwohl/architect | avbiswas/sam2-mlx | chanthruu/wind_energy_monitor | |
|---|---|---|---|
| Stars | 27 | 27 | 27 |
| Language | Python | Python | Python |
| Setup difficulty | easy | moderate | moderate |
| Complexity | 3/5 | 4/5 | 3/5 |
| Audience | developer | researcher | data |
Figures from each repo's GitHub metadata at analysis time.
Installs as a plugin marketplace entry for Codex or Claude Code, requires starting a new agent session after install.
Architect is a plugin and skill set for AI coding agents such as Claude Code and Codex, built to stop those agents from making silent structural decisions while they write code. The problem it addresses is that a request that sounds simple, like adding a second payment provider or refactoring a service, actually forces decisions about ownership, boundaries, compatibility, and rollback behavior. When those decisions are left implicit, a codebase tends to drift into messy patterns: features landing in whatever file is nearest, speculative abstractions built for a future that never arrives, or existing behavior changed without anyone deciding on purpose to change it. Architect forces those decisions into the open through a strict, manual three stage workflow: design, then propose, then build. In the design stage, you describe the change you want in plain language, and Architect reads the repository, asks what needs to stay compatible, and produces an approved design bundle without writing any code yet. In the propose stage, you give the approved design a short plan name, and Architect turns it into a sealed package containing task files, a verification plan, and an execution log. In the build stage, you run that same plan name again, and Architect executes the recorded tasks in order, staying inside the boundaries the design stage set, and updates its logs with what actually happened. Each stage refuses to do the next stage's job. The design stage will not write files or implement anything. The propose stage will not redesign the solution or edit application code. The build stage will not reopen the design or invent new structure partway through. The tool is intentionally manual: it will not guess which stage to run next for you, and the README is explicit that you should not skip straight from a big request to the build stage. You can install Architect as a plugin for Codex or Claude Code, or add its skills directly to a compatible tool. After installing, you start a new session so the agent can discover the added commands. The project is licensed under MIT.
A three stage plugin (design, propose, build) that forces AI coding agents like Claude Code and Codex to make architecture decisions explicit before writing code.
Mainly Python. The stack also includes Python, Claude Code, Codex.
Free to use, modify, and distribute for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.