Improve PR Architecture

Read one pull request through the deep-module lens and propose deepening opportunities: places where the PR introduced a shallow module, tested past an interface, or touched a module that was already shallow and could be deepened cheaply while the author is in there. The output is opportunities, never a verdict. improve- skills generate ideas; review- skills gate. This one never blocks a merge, never edits code, and never posts without an explicit yes.

Usage: /improve-pr-architecture <pr# | PR URL | git ref>, or no argument for the current branch.

Built on the same vocabulary as the other design skills:

• Call the Skill tool with "codebase-design" for the vocabulary (module, interface, depth, seam, adapter, leverage, locality) and its principles (the deletion test, "the interface is the test surface", "one adapter = hypothetical seam, two = real", replace-don't-layer testing). Use these terms exactly in every candidate.
• The domain language in CONTEXT.md names the modules; ADRs in docs/adr/ record decisions this skill honours rather than re-litigates.

What this skill is not: a third smell reviewer. code-review's Standards axis and review-panel's maintainability persona already run the Fowler smells, and Middle Man, Speculative Generality and Shotgun Surgery brush against shallowness from the smell side. This skill's lens is narrower and deeper: interface complexity against implementation complexity, where the seam sits, and whether the tests have locality. A finding that is a smell and nothing more belongs to those skills.

Process

1. Pin the diff

All three input modes end with BASE and HEAD pinned as SHAs, and PR set when there is a pull request to comment on.