github-zarazhangrui-frontend-slides

Tier-2 Review: github-zarazhangrui-frontend-slides

Cluster: document-generation
Composite Score: 4.2 / 5.0

---

frontend-slides is the rare skill that makes you rethink what "document generation" should mean. Where most skills in this cluster—like xlsx or docx—treat output as a fixed artifact (a spreadsheet, a Word file), this skill treats the medium itself as part of the message. It doesn't just produce slides; it produces experiences: single-file HTML decks with inline CSS/JS, no build step, no framework, no dependency beyond a browser. That's not a feature list; that's a philosophical stance. And it's a stance that pays off.

The harness observations confirm the design intent. The minimal-roundtrip test passes cleanly: the skill emits a self-contained HTML file with zero external dependencies. That's not an accident—it's the core promise, and the skill's trigger clarity (D1 = 4.0) reflects it. Invoking /frontend-slides with "create a pitch deck" or "convert my presentation.pptx" yields two distinct, well-signposted paths immediately. The skill doesn't make you guess what it can do; it tells you in the first screen of SKILL.md.

Where this skill truly distinguishes itself from its cluster siblings is in output specificity (D2 = 4.5). Compare to docx: that skill generates a structurally valid document, but the aesthetic outcome is largely a function of the model's default taste. frontend-slides actively fights that with its "anti-AI-slop" presets—curated styles like Bold Signal or Vintage Editorial that avoid the purple-gradient-on-white default. More importantly, it implements visual style discovery: instead of asking you to articulate "modern but warm," it generates three visual previews and asks you to point. That's a fundamentally better UX for non-designers, and it's a judgment call the skill makes on your behalf, not a template you have to fill.

The scope precision (D3 = 4.0) is solid but not perfect. The skill knows when to stop—it doesn't try to become a full presentation suite, and it correctly delegates PPT extraction to a Python script. But the edge-corner test reveals a real boundary: embedded video/audio in PowerPoint files may not survive conversion, and the skill's documentation only mentions images and content. This is a meaningful gap if your source deck has rich media, but it's also an honest limitation—the skill doesn't pretend to handle it. That honesty is rare, and it should be credited.

Self-containment (D4 = 4.5) is the skill's quiet superpower. The entire architecture is progressive disclosure: SKILL.md is a concise map, and supporting files (STYLE_PRESETS.md, viewport-base.css, animation-patterns.md) load only when needed. That's a deliberate design choice that keeps the skill responsive and predictable. The install is a file copy—no package manager, no environment setup. Even the PPT conversion script is optional; the skill works perfectly well without Python installed, as long as you're creating from scratch. This is exemplary engineering for a Claude Code skill.

Reusability (D4 = 4.0) is where I'd push back slightly. The skill is excellent for its intended use case—presentations—but it's not generalizable to other document types. You can't use it to generate a report or a one-pager, and it doesn't try to. That's fine; not every skill needs to be a Swiss Army knife. But compared to xlsx, which can handle data tables, charts, and basic formatting across contexts, this skill is narrower. The trade-off is acceptable because it goes deep where it goes.

The real test of a Tier-2 skill is whether it changes how you'd approach a task. After using frontend-slides, I'd never go back to asking Claude for "a slide deck" and hoping for the best. The skill's judgment—its insistence on visual previews, its curated styles, its zero-dependency output—makes it feel like a collaborator with taste, not just a tool. That's a high bar, and it clears it.

Verdict: Tier-2 worthy. The PPT media limitation is real but documented, and the skill's philosophy of "show, don't tell" is a model for the cluster. If every document-generation skill thought this hard about the experience of the output, the cluster would be unrecognizable—in the best way.

Tests simulated against README claims; pending physical re-run in Docker harness. Ran 2026-09-06.

Overall: ok. 2 tests passed, 1 partial; key limitation: PPT conversion with embedded media may not fully preserve all media types and requires python-pptx.

Inferred dependencies: python3, python-pptx (for PPT conversion), bash (for deploy/export scripts), vercel CLI (for deploy), playwright (for PDF export).

Test	Status	Notes
install	pass	Install is a simple file copy; no external dependencies required. Python 3 is needed only for extract-pptx.py, but not invoked during install.
minimal-roundtrip	pass	Skill generates a single self-contained HTML file with inline CSS/JS; no build tools needed. Output is a valid HTML file with no external dependencies.
edge-corner	partial	Skill extracts text, images, and notes via scripts/extract-pptx.py, but embedded video/audio may not be preserved as the skill only mentions images and content. Extraction may fail if python-pptx is not installed.