interface-review

Reviews git diffs, branches, and pull requests for interface regressions across UI, accessibility, typography, and layout.

Updated Aug 22, 2026
One-click install
npx skills add https://github.com/squidllee/skills --skill interface-review-squidllee
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: interface-review
Source: https://github.com/squidllee/skills/tree/main/interface-review
Command: npx skills add https://github.com/squidllee/skills --skill interface-review-squidllee

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? When you change front-end code, it is hard to know whether the change broke the interface somewhere else. This Skill reviews a change (a branch, pull request, or working tree) rather than a single screen, expanding changed files to the surfaces they affect and classifying every finding as Introduced, Regression, or Pre-existing. ## Core Features & Use Cases - Scope Resolution: Automatically resolves the review target from git state, handling branches, PRs, uncommitted changes, shallow clones, and mid-rebase states, while excluding lockfiles, snapshots, and generated output. - Removed-Signal Detection: Reads the deleted side of every diff hunk to catch removed accessibility attributes, focus styles, motion preferences, and text signals that a post-change-only review would miss. - Classified Findings Report: Produces a scope block, coverage table, and findings table with severity, status, and before/after evidence, ending in an Approve or Block verdict. - Use Case: Before merging a pull request that refactors a shared Button component, run the review to confirm no consumer lost its focus indicator, aria-label, or disabled state. ## Quick Start Ask the assistant to review the interface impact of your current branch or a specific pull request, for example by requesting an interface review of PR 482.

Frequently Asked Questions about interface-review

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

FAQPage Schema
How do I review a pull request for UI and accessibility regressions?▼

Provide the PR number as the review target, and the Skill fetches the pull request ref, diffs it against the merge base, expands changed files to their consumer surfaces, and reports findings with severity and status. It never checks out the PR, so your working tree stays untouched.

How does change-based interface review differ from a full site audit?▼

A change review asks whether this diff made things worse, classifying findings as Introduced, Regression, or Pre-existing and capping pre-existing findings at three. A whole-repository audit is a separate review handed to the better-interface skill without scope blocks or statuses.

What happens when there is no change to review?▼

When the tree is clean and HEAD is not ahead of the merge base, the Skill states the repository facts and offers three routes: review the last commit by SHA and subject, review a named target, or run a whole-repository audit. It never silently falls back to reviewing the last commit.

Does the review modify my working tree or check out branches?▼

No. The review is strictly read-only: it fetches pull request refs into remote-tracking refs and reads files with git show. Checkout, switch, and stash commands are never used, and optional rendered verification runs in an isolated temporary worktree.

Which files are excluded from the interface review scope?▼

Lockfiles, snapshots, generated output, vendored code, and binaries are excluded and named in the scope block. Two exceptions stay in scope: font files, which affect typography, and images added to components, which affect alt text and accessibility.

Why does the review read the deleted lines of a diff?▼

Regressions are invisible in the post-change state, so the removed side of every hunk is searched for dropped aria attributes, focus styles, reduced-motion handling, and text signals. Each removal is routed to the owning domain skill and only reported if confirmed as a real regression.