improve-codebase-architecture

Scan a codebase for module deepening opportunities and present them as a visual HTML report.

9|Updated Jul 17, 2026
One-click install
npx skills add https://github.com/Yassimba/loom --skill improve-codebase-architecture-yassimba
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: improve-codebase-architecture
Source: https://github.com/Yassimba/loom/tree/main/skills/improve-codebase-architecture
Command: npx skills add https://github.com/Yassimba/loom --skill improve-codebase-architecture-yassimba

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Codebases accumulate shallow modules whose interfaces are nearly as complex as their implementations, making them hard to test and hard for humans and AI agents to navigate. This Skill surfaces that architectural friction and turns it into concrete, visualized refactoring candidates. ## Core Features & Use Cases - Architecture exploration: Walks the codebase (guided by recent commit hot spots, CONTEXT.md domain language, and existing ADRs) to find shallow modules, leaking seams, and untestable areas, applying the deletion test to each suspect. - Visual HTML report: Renders each candidate as a self-contained HTML file in the OS temp directory using Tailwind and Mermaid via CDN, with before/after diagrams, recommendation-strength badges, and ADR-conflict callouts. - Decision grilling loop: After you pick a candidate, it hands off to the grill-me skill to walk constraints, dependencies, and the shape of the deepened module, updating CONTEXT.md or proposing ADRs when decisions stick. - Use Case: Point it at a subsystem that keeps churning in git history; it produces a report showing, for example, that the Order intake pipeline is six shallow layers deep and proposes collapsing it into one deep module with a single testable interface. ## Quick Start Use the improve-codebase-architecture skill to review this repository and show me the top deepening opportunities in an HTML report.

Frequently Asked Questions about improve-codebase-architecture

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I find shallow modules in my codebase?

Run this skill and it spawns a sub-agent to walk the codebase looking for modules whose interface is nearly as complex as their implementation. It applies the deletion test to each suspect: if deleting the module would concentrate complexity rather than move it, it is flagged as a deepening candidate.

How to visualize architecture refactoring proposals?

The skill writes a self-contained HTML report to the OS temp directory using Tailwind and Mermaid from CDNs. Each candidate gets a card with files, problem, solution, benefits, a side-by-side before/after diagram, and a recommendation-strength badge of Strong, Worth exploring, or Speculative.

Does the architecture review respect existing ADRs?

Yes, it reads ADRs in ai-docs/adr/ before scanning and avoids re-litigating recorded decisions. If a candidate contradicts an ADR, it is only surfaced when the friction justifies reopening the decision, and the card is marked with a warning callout referencing the ADR.

What happens after I pick a refactoring candidate?

The skill hands off to the grill-me skill, which walks the decision tree with you: constraints, dependencies, the shape of the deepened module, and what tests survive. New domain terms are added to CONTEXT.md, and load-bearing rejections can be recorded as ADRs.

Where does the architecture report get written?

The report is written to the OS temp directory as architecture-review-<timestamp>.html, resolved from $TMPDIR with fallback to /tmp or %TEMP% on Windows. Nothing is written into the repository, and the file is opened automatically with the platform's default opener.