Tools
The web apps, APIs, terminal utilities and open-source projects we use alongside Claude Code and Claude Cowork, or keep an eye on. We update it as we work.
Showing 23 of 23
No tools match that.
Try a different search or clear the filter.
Claude Code skills and plugins, the ones we install into our agents, and the tooling around them.
A curated directory of agent skills for design engineers.
A searchable catalogue of SKILL.md skills that give agents like Claude Code, Cursor and Codex frontend and design-engineering expertise. Each entry covers a focused area (accessibility, animations, design systems) and installs by copying one command.
Equipping a coding agent with production-grade UI craft without writing the skill yourself.
A repository of engineering skills for coding agents.
Twenty-four skills maintained by Addy Osmani covering the full development cycle across six phases: define, plan, build, verify, review and ship. It ships four specialist personas (code review, testing, security and web performance), seven reference checklists and eight commands (`/spec`, `/plan`, `/build`, `/test`, `/review`, `/webperf`, `/code-simplify`, `/ship`). Install it with `npx skills add addyosmani/agent-skills`; it works in Claude Code and in any other agent that reads SKILL.md.
Giving an agent a phased working method instead of handing it one loose task.
A security scanner that vets agent skills before install.
Scans agent skills from Git repos, URLs, zips or directories against 64 vulnerability patterns, such as prompt injection or data exfiltration. It returns a 0 to 100 risk score in terminal, JSON, Markdown or SARIF.
Vetting third-party skills for malicious or vulnerable code before installing them into Claude Code, including inside CI via SARIF.
A CLAUDE.md ruleset that sharpens how agents write code.
Packages four coding principles into a CLAUDE.md ruleset that steers how Claude Code and Cursor approach tasks, targeting common failure modes like overengineering and needless edits. It ships as a plugin or as a standalone markdown file.
Getting more focused, reliable agent behaviour without writing your own guideline file from scratch.
An agent skill that makes AI coding agents write far less code.
Makes a coding agent stop before writing code and climb a ladder of cheaper options first: skip it (YAGNI), standard library, native platform feature, already-installed dependency. Its published benchmarks report 80-94% less code and 3-6x faster runs.
Stopping a coding agent from over-engineering, cutting dependencies, boilerplate and token cost on everyday tasks.
A collection of composable skills for coding agents.
An open-source set of modular skills that extend agents like Claude Code, Codex and Cursor across planning, review and execution, with visual outputs in interactive MDX (diagrams, annotated code). An interactive CLI installs them one at a time.
Adding focused planning, review and orchestration skills to a coding agent without building the workflow yourself.
An agent skill that strips AI-writing tells from text.
Detects 33 documented signs of AI writing (rule-of-three constructions, excessive em dashes, hedging language) and rewrites passages to remove them, with a final audit. An optional calibration mode matches your voice from writing samples.
Making AI-drafted copy read as human-written before publishing, optionally in your own voice.
A local git proxy that validates your branch with agents before pushing.
It sits in front of your real remote: push to no-mistakes and it validates the branch in a disposable worktree with agents (review, tests, docs, lint), forwarding it only once every check passes and opening a clean PR. Safe fixes are applied automatically; the rest is escalated to you.
Keeping slop out of PRs: every push goes through agent review, tests and lint without interrupting your work.
Supabase's official skills for coding agents.
Two skills maintained by Supabase: one covers every Supabase product (database, Auth, Edge Functions, Realtime, Storage, vectors, Cron, Queues) plus client-library integration, the other collects Postgres performance practices across eight categories. They install with npx skills add supabase/agent-skills or as a Claude Code plugin from their marketplace, and follow the Agent Skills Open Standard format, which 18+ agents support.
Getting an agent to write Supabase queries, indexes and auth flows from the official documentation instead of improvising.
OpenAI's official plugin that brings Codex into Claude Code.
Adds commands to review code with Codex without leaving Claude Code: /codex:review for a normal read-only review, and /codex:adversarial-review for a tougher pass that questions the design and the choices made. It also lets you hand off tasks to Codex in the background with /codex:rescue, and manage them with /codex:status, /codex:result, /codex:transfer and /codex:cancel. It needs a ChatGPT subscription (or an OpenAI API key) and Node.js 18.18 or later.
Getting a second opinion from Codex on a change, or delegating an investigation or fix to it, without switching tools.
Scrapers, headless browsers and agents that read and drive the web.
A search API that scrapes AI engines and Google.
A unified API that queries AI search engines (ChatGPT, Perplexity, Copilot, Gemini, Grok) and traditional ones (Google Search, News, AI Overviews) and returns clean structured JSON. Each response carries parsed markdown, HTML or text plus extracted sources, citations, query fan-out and shopping blocks. It supports geographic targeting, async batch requests and webhooks.
Monitoring brand visibility and citations across AI answer engines at scale for GEO and AEO work, without integrating each provider.
A lightweight Rust headless browser for automation.
An open-source headless browser engine written in Rust that runs real JavaScript via V8 and speaks the Chrome DevTools Protocol, acting as a drop-in replacement for headless Chrome with Puppeteer and Playwright. It offers a CLI for fetching and extracting page content, parallel scraping, built-in stealth and anti-fingerprinting, and proxy support. It uses far less memory and starts faster than Chrome, and exposes an MCP integration for agents.
Web scraping, browser testing and agent automation at scale where a lighter, faster alternative to headless Chrome helps.
A browser agent framework where the LLM writes Playwright code.
A terminal-native browser agent framework from Microsoft Research. Instead of predicting individual clicks, the model emits bash commands and writes Playwright scripts to drive the browser, observe pages and iterate. The agent's work becomes reusable, inspectable code (scripts, logs, screenshots) you can rerun and repair when it breaks.
Building browser-using agents with a transparent, debuggable alternative to heavy frameworks for long web tasks.
An adaptive Python web-scraping framework.
A Python library that bundles HTTP requests, stealth mode and full browser automation behind one API. Its parsers adapt automatically when a site's structure changes, and its fetchers can bypass anti-bot systems like Cloudflare Turnstile. It also ships a spider framework for concurrent crawls, a CLI with an interactive shell, and an MCP server for AI-assisted scraping.
Extracting data reliably, from one-off scrapes to large crawls that survive layout changes and anti-bot defences.
Tools to measure and cut token spend and compress context.
A CLI proxy that compresses command output for AI agents.
Intercepts terminal commands and strips verbose boilerplate from their output before it reaches a model's context window, while preserving the parts that matter like errors and test results. It installs a hook so every command is transparently rewritten to use its compression. It claims around 89% noise removal across 30+ commands such as cargo test, git, grep and find.
Cutting token use and cost in terminal coding agents (Claude Code, Cursor, Aider) and running longer sessions before context overflows.
A local terminal dashboard for AI coding token costs.
An open-source CLI and TUI that reads coding-session data straight from disk to show where your tokens and money go across Claude Code, Cursor, Codex and 20+ other tools. It breaks usage down by task, model, tool and project, tracks one-shot success and cache-hit rates, and its optimise scan flags waste patterns with estimated savings. Everything runs locally, no proxy or API keys.
Seeing and reducing token spend, comparing model performance, and staying inside a subscription budget.
A context-compression toolkit for AI agents and LLMs.
Compresses tool outputs, logs, files, RAG chunks and conversation history before they reach the model, claiming 60-95% fewer tokens while preserving answer quality. It ships in several forms: an inline compress() function for Python and TypeScript, a zero-code HTTP proxy, an MCP server, and direct wrappers for agents like Claude Code, Cursor, Codex and Aider. Reversible compression caches the original content for on-demand retrieval, and a shared memory layer dedupes context across agents.
Cutting token use and API cost across coding agents and LLM pipelines without rewriting existing code.
Document conversion and retrieval to feed models.
A utility that converts many file types to Markdown.
Converts documents such as PDF, Word, PowerPoint, Excel, HTML, images and audio into clean Markdown, preserving structure like headings, tables and lists. It runs as a CLI command, a Python library or via Docker, and supports optional plugins. The point is to prepare mixed source files for language models and text-analysis pipelines.
Turning a folder of PDFs and spreadsheets into Markdown to feed an LLM or an indexing pipeline.
A visual RAG system that retrieves document screenshots instead of parsed text.
Renders pages to image tiles with Playwright/CDP and embeds them with a fine-tuned Qwen3-VL model into a FAISS index, so tables, charts and layout survive retrieval instead of being flattened to text. It ships a rendering CLI, an indexing pipeline and a FastAPI search endpoint, plus a hosted index of 8.28M Wikipedia pages to query out of the box. A bundled Claude Code skill lets an agent screenshot and read pages directly rather than parsing HTML.
Semantic search and agent retrieval over visually structured documents, charts and tables that text-based RAG loses.
A LangChain CLI that writes and maintains codebase documentation built for agents.
It generates a wiki of your repository in an openwiki/ folder and keeps it current as the code changes; an optional GitHub Action opens a daily PR with documentation updates. It appends instructions to your AGENTS.md or CLAUDE.md so your agent consults the wiki when hunting for context. Supports Anthropic, OpenAI, OpenRouter, Fireworks and Baseten, with any model you pick.
Giving coding agents an up-to-date map of the repo without maintaining the docs by hand.
Tracking models and their versions.
A live dashboard tracking AI model releases and versions.
Monitors the release status and current versions of frontier models from every major lab (Claude, OpenAI, Google, xAI, Meta, DeepSeek and more) in one place. It shows a timeline of shipped versions, recent and rumoured releases, and per-model summaries with benchmark and pricing data. You can subscribe to be alerted the moment a new version launches.
Knowing instantly whether a model has shipped, without trawling changelogs or social media.
Interface design references for naming things correctly, in conversation and in prompts.
A visual dictionary that gives you the correct name for any interface element.
You describe the element in your own words ("the dark see-through layer behind a popup") and it returns the standard name, the API symbol and a prompt ready to paste. It documents 76+ web and macOS elements, each with a visual example, labelled anatomy, HTML/CSS snippets and tables mapping the plain name to AppKit, SwiftUI and web. It also compares lookalike patterns (modal, drawer, sheet) and searches by description with ⌘K.
Asking an agent for the component you actually mean, and keeping design and development on the same vocabulary.
Self-hosted mail and campaigns.
A self-hosted mail server and email marketing platform.
A self-hosted system that runs your own mail server and sends, tracks and manages email marketing campaigns from it. It combines a full mail server (webmail, DNS management, automatic SSL) with newsletter tooling: templates, contact lists, and analytics for opens, clicks and delivery. It runs on your own infrastructure via Docker, so subscriber data stays on your servers with no per-email fees.
Running newsletters and campaigns without recurring SaaS costs or vendor lock-in, keeping data in-house.
We use analytics cookies to understand how the site is used. Nothing is sent until you accept. You can read the cookie policy.