What it does
Looks for architectural friction and proposes deepening opportunities, with testability and AI-navigability as the aim. It scopes first by the hot spots in your git history, reads GLOSSARY.md and any ADRs, then spawns a sub-agent to walk the code looking for shallow modules, leaky seams and hard-to-test parts, applying a deletion test. It writes a self-contained HTML report (Tailwind and Mermaid from a CDN) to the OS temp folder and opens it: each candidate gets files, problem, solution, wins, a custom before and after diagram and a strength badge, ending with a top recommendation. It uses a fixed vocabulary (module, interface, depth, seam, adapter, leverage, locality). After you pick one, it grills the decision tree and keeps GLOSSARY.md and ADRs up to date.
Output
An HTML architecture review, then an interactive design conversation about the candidate you choose.
Good for
Teams who want a structured, visual review of where their code is shallow, and who are happy to adopt a shared design vocabulary.
Medium risk: instructions only, no scripts, but it runs git log, spawns a sub-agent that reads your code, writes an HTML file to the temp folder and opens it with a command, and later may edit GLOSSARY.md or add ADRs in your repo. The report loads Tailwind and Mermaid from public CDNs. It relies on other Skills by the same author (codebase-design, grilling, domain-modeling). Not test-run: it needs a real repository and sub-agents, so there is no demo and the front end marks it as unverified. By Matt Pocock, MIT.