Frontend Slides

Frontend Slides is an independent open-source agent skill by Zara Zhang. The official source is maintained as a public GitHub repository and also provides a Claude Code plugin with its own plugin metadata. According to the provider, the skill helps create animated HTML presentations from scratch, convert existing PowerPoint files into a web presentation, and improve existing decks in a controlled way. It is not a presentation hosting service, a hosted editor, or a substitute for editorial and legal approval. Its product boundary is a reusable set of instructions, templates, viewport CSS, animation patterns, and local helper scripts that a compatible coding agent can apply inside a user-selected project.

Purpose and workflow

The workflow starts by clarifying purpose, length, available content, and the intended presentation form. For new decks, the source proposes visual style discovery: the agent creates several small HTML previews, and the user compares them before a full deck is generated. This turns an abstract design choice into a visible comparison. For existing files, the workflow first checks element counts, image sizes, text density, and viewport behavior. When content is too dense, it should be split across slides instead of being compressed into an unreadable canvas.

The included guidance covers fixed presentation stages, slide navigation, keyboard and touch controls, reduced motion, responsive outer interfaces, and self-contained HTML output. The repository includes style presets and an optional bold template pack with many visual systems. Those templates are not corporate brand approvals and do not grant permission to use third-party logos, images, or quotations. Users must review fonts, images, SVGs, screenshots, quotations, and trademark rights for every resulting deck. The repository's MIT license permits use of the published source under its conditions; it does not automatically transfer rights to later presentation content or remotely referenced assets.

Files, conversion, and output

The core workflow aims for one HTML file with CSS and JavaScript embedded inline. The source also documents a Python script for extracting text, images, and notes from PPTX files, plus an export script that serves local slides over a temporary HTTP server and uses Playwright to render a PDF. These actions read local files, create previews, or write output files. They should run only on explicitly selected files, with a clear destination path, and never on production or confidential material without authorization. PDF export is a static snapshot; animations and interactive behavior are not fully preserved.

Data, network boundaries, and human acceptance

The basic output can work locally, but the reference architecture names Fontshare or Google Fonts stylesheets. That creates requests to third parties and may disclose an IP address, user agent, and request time. Optional paths also refer to npm, Playwright, Vercel, YouTube, and external template or font CDNs. Before use, review these requests, licenses, privacy effects, tracking risks, and provider availability. Company data, customer information, presentation content, and internal metrics belong in the context only with appropriate authorization. An agent must not present facts, quotations, brand use, or business claims as approved. Before publication, delivery, deployment, PDF export, or irreversible file changes, a person must review content, accessibility, security, sources, rights, and audience fit.