pr-review-canvas

pr-review-canvas generates an interactive, self-contained HTML walkthrough of a GitHub pull request that reads like a peer explaining what matters. It solves the problem of dense, hard-to-scan PR diffs by fetching the PR's metadata, changed files, and review comments through the gh API, categorizing files into core versus mechanical/boilerplate changes, adding reviewer annotations, and rendering diffs with moved-code detection so a reviewer can focus on the substantive changes first.

The skill ships a prebuilt dark-themed toolkit — a styles.css of CSS classes (headers, stat pills, collapsible file cards, sticky reviewer notes, boilerplate cards, verdict checklist) and a renderer.js of JS utilities (toggle helpers, a renderDiff function that filters imports, collapses whitespace-only changes, and tints moved code, plus an HTML-escape helper) — injected into a template.html. The assistant writes the PR body HTML with creative freedom, optionally adding plain-English pseudocode summaries for verbose code, diagrams, before/after flowcharts, and callout boxes. Notably, the skill includes careful, security-conscious guidance for safely embedding diff patch strings: because patches can contain a script-closing sequence plus newlines and quotes, it instructs saving patches to a JSON file via jq and injecting them with Unicode-escaped JSON through a small Python step so a raw patch cannot break out of an HTML script tag.

It targets developers and reviewers who want a richer, shareable review artifact than a plain diff. It is read-only with respect to the repository — it only fetches PR data via gh and produces a local HTML file — making it a benign documentation/visualization tool.