2.2K skills · sorted by relevance
Implementação educacional do GPT em ~300 linhas. Reproduz o GPT-2 (124M) em OpenWebText. Código limpo e hackeável para aprender transformers. Por Andrej Karpathy. Perfeito para entender a arquitetura do GPT do zero. Treine em Shakespeare (CPU) ou OpenWebText (multi-GPU).
The English summary is being prepared. Showing the original description.
Use this Canvas-owned clean-room skill when working with Remotion or React-based video generation. Trigger for Remotion projects, compositions, sequences, frame-based animation, rendering, captions, audio, video assets, templates, and debugging programmatic video output.
The English summary is being prepared. Showing the original description.
Use this Canvas-owned clean-room skill whenever a PowerPoint `.pptx` deck is an input or output. Trigger for creating, editing, reviewing, repairing, templating, combining, splitting, rendering, or extracting slide decks, including pitch decks, reports, speaker notes, template-following, and Google Slides-targeted handoff.
The English summary is being prepared. Showing the original description.
Load this skill whenever the project contains SVG graphics — inline SVGs, external SVG files, SVG icons, SVG illustrations, or SVG-based data visualizations. Under no circumstances use SVG without proper accessible titles, descriptions, and ARIA roles where required. Absolutely always add <title> and <desc> to meaningful SVGs and set role="img" with aria-labelledby pointing to those elements. Also apply this skill whenever the project accepts untrusted or user-uploaded SVG content, which requires sanitization before rendering.
The English summary is being prepared. Showing the original description.
Load this skill whenever the project contains charts, graphs, data visualizations, infographics, or any visual representation of data (bar charts, line charts, pie charts, scatter plots, dashboards). Under no circumstances render a chart without a text alternative, data table, or summary. Prioritize conveying the same information through non-visual means.
The English summary is being prepared. Showing the original description.
Load this skill before implementing a fix for a barrier that may originate in a shared component, design system, renderer, or external dependency rather than this project's own code. Decide whether the responsible fix is local, an existing installed capability, or an upstream contribution before writing a workaround. Absolutely always record any temporary local divergence with an owner and removal condition. Under no circumstances open, comment on, or submit anything in an external project without explicit human approval.
The English summary is being prepared. Showing the original description.
Load this skill when writing, reviewing, or interpreting behavioral accessibility automation — tests that manipulate a rendered page (resizing the viewport, pressing real keys, capturing and comparing screenshots) rather than only inspecting static markup or a single computed accessibility tree. Covers Reflow risk (SC 1.4.10) and Focus Visible risk (SC 2.4.7) checks. Absolutely always report indicator results separately from confirmed failures. Under no circumstances treat a clean behavioral scan as WCAG conformance, or collapse `cantTell`/test-error/indicator outcomes into a single pass/fail signal.
The English summary is being prepared. Showing the original description.
Default material, information-density hierarchy, clean-rendering, negative-prompt hygiene, and evidence-preserving photo-repair layer for IM2 / GPT Image 2 / gpt-image-2 / image_gen. Use for clean generation/redraws, artifact cleanup, focal-detail control, atmospheric depth, physically coherent light paths, air scattering, volumetric light, light ratio, material realism, architecture/terrain/hard-surface scene mass and structural grounding, and 企业/公司活动/门店/团队合影/商品/餐饮/证件文档照片修复. Preserve identity, headcount, placement, product facts, text, logo, QR, date, price, and scene evidence; prefer residual blur over invented restoration detail.
The English summary is being prepared. Showing the original description.
Use when choosing, reviewing, or implementing SF Symbols in an Apple-platform app — picking meaningful symbols, keeping the same concept on the same symbol everywhere, and handling variants, rendering modes, and accessibility.
The English summary is being prepared. Showing the original description.
Research-grounded typography for any text-bearing surface — web, PDF, email, documents. GENERATIVE: apply correct line length, font sizing, contrast, dark mode, mobile, and Japanese rules from the start when building any UI, page, layout, or document. REVIEW: audit existing CSS/HTML/templates for typography violations. Use this skill whenever you are writing CSS that touches font-size, line-height, max-width, color, or font-family. Use it when building any frontend component, page, or layout — even if the user doesn't mention typography. Use it when reviewing any rendered output for readability. Use it when generating PDFs, emails, or any formatted document. If text will be read by a human, this skill applies.
The English summary is being prepared. Showing the original description.
Extract clean markdown or text content from specific URLs via the Tavily CLI. Use this skill when the user has one or more URLs and wants their content, says "extract", "grab the content from", "pull the text from", "get the page at", "read this webpage", or needs clean text from web pages. Handles JavaScript-rendered pages, returns LLM-optimized markdown, and supports query-focused chunking for targeted extraction. Can process up to 20 URLs in a single call.
The English summary is being prepared. Showing the original description.
Converts Markdown to styled HTML with WeChat-compatible themes. Supports code highlighting, math, Mermaid (rendered to PNG via headless Chrome), PlantUML, footnotes, alerts, infographics, and optional bottom citations for external links. Use when user asks for "markdown to html", "convert md to html", "md 转 html", "微信外链转底部引用", or needs styled HTML output from markdown.
The English summary is being prepared. Showing the original description.
Use when the task involves reading, creating, or editing `.docx` documents, especially when formatting or layout fidelity matters; prefer `python-docx` plus the bundled `scripts/render_docx.py` for visual checks.
The English summary is being prepared. Showing the original description.
GodoMaster — Godot 4.x game development intelligence: GDScript, nodes/scenes, 2D/3D rendering, physics, animation, UI, audio, input, shaders, networking, performance, export, file I/O, testing, localization, AI behavior, asset pipeline, and architecture tooling. Complete workflow for indie and professional game development based on official Godot documentation.
The English summary is being prepared. Showing the original description.
GStack design-html method for producing or planning quick HTML prototypes, design boards, or rendered visual explorations for UI review.
The English summary is being prepared. Showing the original description.
GStack design-review method for reviewing rendered UI, landing pages, decks, or visual artifacts for hierarchy, layout, accessibility, and polish.
The English summary is being prepared. Showing the original description.
Explicit compatibility adapter for the collection's legacy PDF workflow and delivery conventions. Prefer the installed official PDF skill for generic PDF reading, generation, rendering and forms.
The English summary is being prepared. Showing the original description.
Create visually rich PowerPoint (.pptx) presentations from academic papers, research notes, or any content the user wants in slide format. Uses PptxGenJS + LaTeX formula rendering. Always use this skill when the user wants PPT/PPTX output instead of Beamer/LaTeX slides. Trigger on: powerpoint, pptx, PPT, make a ppt, 做PPT, 做幻灯片, make slides (non-LaTeX), prepare a presentation (when context implies PPT), 做个报告, presentation slides, help me prepare a talk (when not Beamer), convert paper to slides (when PPT implied). Even if the user just says "make slides" or "做个演示" without specifying format, trigger this skill and ask whether they want PPT or Beamer — don't silently default to Beamer. Do NOT trigger on: beamer, latex slides, .tex files, tikz — those belong to the beamer skill.
The English summary is being prepared. Showing the original description.
Create, restyle, or rebuild Chinese academic PowerPoint presentations from papers, theses, reports, figures, notes, templates, screenshots, or visual mockups. Use for presentation planning, Template DNA extraction, slide mockups, template-based editable PPTX generation, editable reconstruction, rendering, and final QA.
The English summary is being prepared. Showing the original description.
Build a complete presentation deck from a brief. Generates storytelling plan, slide operations, renders the deck, and runs QA. Requires extracted template contracts from /slides-extract. Use when the user wants to create slides, build a deck, generate a presentation, write a strategy deck, or says things like "make me a 10-slide deck on X", "create a presentation about Y", "build slides for the board meeting".
The English summary is being prepared. Showing the original description.
Creates and modifies evolving-scene browser presentations. Use when the user asks to "create a presentation", "build a talk", "add steps", "modify a presentation", "update/resync the scene kit", or scaffold presentation infrastructure. Gathers requirements interactively, bootstraps missing scaffolding (monorepo-aware), registers presentations at their own routes, resyncs a project's vendored scene kit with the latest snapshot, and self-verifies build + render before reporting done.
The English summary is being prepared. Showing the original description.
Create professional, consultant-grade PowerPoint presentations from scratch using MckEngine (python-pptx wrapper) with McKinsey-style design. Use when user asks to create slides, pitch decks, business presentations, strategy decks, quarterly reviews, board meeting slides, or any professional PPTX. AI calls eng.cover(), eng.donut(), eng.timeline() etc — 67 high-level methods across 12 categories (structure, data, framework, comparison, narrative, timeline, team, charts, images, advanced viz, dashboards, visual storytelling), consistent typography, zero file-corruption issues, BLOCK_ARC native shapes for circular charts (donut, pie, gauge), production-hardened guard rails for spacing, overflow, legend consistency, title style uniformity, dynamic sizing for variable-count layouts, horizontal item overflow protection, chart rendering, and AI-generated cover images via Tencent Hunyuan 2.0 with professional cutout, cool grey-blue tint, and McKinsey-style Bézier ribbon decoration.
The English summary is being prepared. Showing the original description.
Generate editable learning PPTX decks by first creating HTML slides, or convert local HTML slide decks and single-page HTML presentations into editable PPTX files by rendering in a browser, reading DOM element positions, extracting inline SVG assets, and rebuilding text, shapes, gradients, bullets, and icons as separated PowerPoint objects. Use when a user asks to generate PPT/PPTX from a topic, outline, course material, study notes, or structured content, or asks for pixel-level/browser-positioned HTML-to-PPT/PPTX conversion with editable/separate elements instead of one full-slide screenshot.
The English summary is being prepared. Showing the original description.
Build, edit, redesign, render, and verify content-led, structurally diverse, clean, editable, reproducible PowerPoint `.pptx` decks. Use for scientific, lab, clinical, consulting, board, investor, editorial, policy, and operational presentations where alignment, readability, contrast, and balanced density matter.
The English summary is being prepared. Showing the original description.