Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
Architect-loop provides a series of specialized AI workflows designed for technical leads and architects. It moves codebase design through a rigorous pipeline of research, architectural drafting, and adversarial review to ensure system integrity before implementation begins.
The framework focuses on converting abstract requirements into actionable specs and issues, reducing the risk of architectural drift while optimizing token consumption during complex design loops.
WHO IT'S FOR
Technical leads / architects
designing high-quality codebase architectures with AI
AI agent / automation builders
optimizing token usage in complex AI loops
Backend / infrastructure engineers
implementing features using strict TDD cycles
QA / test automation engineers
hardening specifications through adversarial review
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Follow the documented setup, then try a first task.
I’m using
Bash user-profile installation
Installs: Complete architect-loop skills bundle installed into the user profile: Claude under $HOME/.claude/skills and Codex under $HOME/.agents/skills. · Claude, Codex
Before you start
Git
Bash
Codex CLI >= 0.133 when using Codex default Goal Mode; the Codex builder backend is otherwise optional.
Platform: Bash environment
Run from: A parent directory where the architect-loop repository should be cloned
Change into the cloned repository and run the Bash installer. The trailing comment records the corresponding Windows command and is preserved verbatim from the documented code unit.
cd architect-loop && ./install.sh # Windows: .\install.ps1
Use the codebase-design skill to review [target module or function] in [repository path]. Identify its interface and seam, assess whether it is deep or shallow using leverage, locality, and testability, and suggest one small redesign. Do not edit the code. Replace the bracketed placeholders with the local repository path and the specific module or function to review.
Use the tdd skill to implement [one small behavior] at the pre-agreed seam [seam named by the issue or spec] in [repository path]. Write one failing behavior test, add only the minimal implementation needed to pass it, run the test, and report the command output and exit code. Do not refactor while RED. Replace the bracketed placeholders with the behavior, the existing named seam, and the local repository path.
Claude
Codex