Tools

Turns a Notion PRD or feature spec into an implementation plan page and tasks, flagging open questions and linking spec, plan and tasks while tracking progress. From OpenAI's official plugin repo (developer: Notion); needs the Notion MCP connected.

Open-source LLM evaluation platform: one opencompass command picks the model and datasets and runs the evaluation across hundreds of benchmarks. Works with local Hugging Face models and API models, with vLLM and LMDeploy for faster inference.

A more intuitive find-and-replace tool than sed: uses JavaScript/Python-style regex syntax, supports literal string mode and in-place file edits, which suits agents doing bulk text rewrites.

Converts raster images like JPG and PNG to SVG, handling color tracing, black-and-white line art and unevenly lit scans, for vectorizing logos, illustrations and old drawings. The CLI is a 1.0 preview installed with cargo (needs Rust); a desktop app is also available.

Tw93's lean pack of engineering-habit skills: /think to pressure-test plans, /hunt for root causes, /check to review diffs, /ui for frontend interfaces, /write to polish Chinese and English prose, and more (8 skills). Works with both Claude Code and Codex.

Hands your current browser session to the agent through a Chrome extension: read console logs and network requests, take screenshots, and run Lighthouse performance, accessibility and SEO audits without signing in again.

Converts and two-way syncs Jupyter notebooks with plain Python scripts or Markdown, so agents can edit research and teaching notebooks like ordinary text and diffs stay readable. Needs a local Python environment; install with pip.

Turns any folder into a Markdown task board: Claude Code or Codex break work into tasks with acceptance criteria for review before coding, managed from the terminal or a browser board. Works without Git for personal plans too. Runs locally with no account; needs Node.js or Homebrew.

Cleans a LaTeX project before arXiv submission: removes comments, auxiliary files and unused .tex files and images, and can compress images, so internal notes are not published with the source. From Google Research; installable with Homebrew on macOS.

Uses a Figma plugin and a local WebSocket server so agents can read and directly edit designs: create frames and text, adjust auto layout and styles, and replace copy in bulk. Requires Bun; now maintained by Grab.

Python library and CLI for extracting main web content: fetches pages, crawls via sitemaps and RSS feeds, strips headers, footers and other noise, and outputs TXT, Markdown, JSON, XML and more with metadata. Also processes downloaded HTML offline.

Gives the agent three exact Tailwind arbitrary-value shadow utilities (sm, md, lg) for layered, neutral elevation on cards, controls, popovers and hero media, with rules against mixing them with default shadows, tinting or stacking.

Adds a lightweight interactive WebGL globe with the cobe library: canvas sizing for device pixel ratio, location markers, rotation in onRender, resize handling and teardown, for plain pages or React/Next.js hero sections.

Helps the agent add or debug web animations with GSAP: tweens, timelines and position parameters, easing, stagger and ScrollTrigger pin and scrub, plus fixes for jank, cleanup with gsap.context() in React and refreshing after images load.

Has the agent use Iconify Simple Icons brand marks instead of typed company names in logo rows, integration grids and partner lists, sized in a 64x64 box, monochrome by default, grid-aligned and labeled for accessibility when needed.

General guidance for building UI with Tailwind CSS: responsive and state variants, dark mode, sparing arbitrary values, components before @apply, and pitfalls such as dynamically built class names missing in production, with button and hero examples.

Builds or debugs interactive 3D web scenes with Three.js: scene, camera, renderer, lights, materials, GLTF loading and controls, plus performance basics like capping pixel ratio, handling resize, disposing WebGL resources and reduced motion.

Steers WebGL-heavy landing pages toward a chosen look by picking one of four visual lanes per page type, passing message, performance, accessibility and reliability gates, and outputting a lane, visual spec, stack, fallback and build order.

Guides the agent to build a coherent, product-grade web motion system in the style of Stripe, Linear, Apple and Vercel: reusable primitives, duration and easing defaults, stagger choreography, scroll and hover patterns, performance rules and reduced-motion support.

Adds thin vertical guide lines at the content container edges plus small corner squares for a measured, structured layout, with CSS tokens and pseudo-element code that keeps guides behind content and out of pointer events.

Turns a fuzzy UI idea into a design-first prompt spec for UI generation: goal, format, layout, type system, color, imagery, exact copy, constraints and negative prompt, then iterates by changing one variable at a time.

Builds cinematic, scroll-driven product storytelling with GSAP ScrollTrigger: pinned scenes, scrubbed timelines, progressive UI reveals and section handoffs, with cleanup in React, simpler choreography on small screens and readable copy and CTAs.

Builds seamless infinite marquees for logos, testimonials, screenshots or tags by duplicating the item sequence and translating the track from 0 to -50%, with stable widths, faded edges, optional hover pause and reduced-motion handling.

Reveals headings and short copy word by word through an overflow mask as they scroll into view, using GSAP ScrollTrigger and a split helper that avoids the paid SplitText plugin, with aria-label, reduced-motion support and React cleanup.