Curated Directory

Sparks of Weight-Space Programming

MIT CSAIL post showing GPT-6 Astra reading, fixing and writing neural network weights directly, with no gradients or training. One edit to 0.016% of Qwen3.5-0.8B's weights took MATH-500 from 7.9% to 48.1%.

djs

Small JavaScript and TypeScript engine in no_std Rust from Armin Ronacher, built mainly for wasm32 to replace QuickJS in Pi's code mode, where model-written scripts call tools and return data.

react-resizable-panels Workspace

Brian Vaughn's react-resizable-panels added a Workspace component in 4.16 for VS Code-style layouts, with tabs you drag between groups, reorder, or drop on an edge to split.

next-bundle-optimizer

Official Next.js agent skill that reads Turbopack's module graph to find client JavaScript that could be deferred, duplicated dependencies and failed tree shaking, then reports or applies the fixes.

agent-desktop

Computer use for any coding agent, from the agent-browser team at Vercel Labs. A Rust CLI that reads and operates desktop apps on macOS, Windows and Linux through the OS accessibility tree instead of pixels.

Delta for Claude Code

Zed's Delta CLI and Claude Code plugin make a Claude Code session shareable. A teammate joins with your conversation and uncommitted worktree as context, in their own Claude session.

TanStack Redact

A smaller, synchronous runtime for existing React apps. Add one build plugin for TanStack Start, Vite, Vinext or Next.js and ship less client JavaScript without changing application code.

Vercel CI

Vercel's CI service, in beta on all plans. It reads your turbo.json, runs only the tasks a change affects, gives each task its own machine and serves unchanged ones from Remote Cache.

Alchemy

Infrastructure as Effects. Cloud resources and the application code that uses them live in one type-safe Effect program in TypeScript, across Cloudflare, AWS, GCP, Kubernetes, Fly, Neon, PlanetScale, Stripe and more.

SPACE

Perplexity's open-source sandbox platform, the one behind every task in Perplexity Computer. It runs agent and user code in isolated Firecracker microVMs behind an HTTP API.

durable-harness

Vercel's adapters that run existing coding-agent harnesses (Pi, OpenCode, fx, Claude Code, Codex) as durable agents on a Workflow SDK World, so a crash or function timeout resumes the turn instead of losing it.

Dillon Mulroy's Coding Standards and Error Design docs

Two rule docs Dillon Mulroy carries from project to project in place of a pile of skills. One covers coding standards for Effect + Alchemy TypeScript on Cloudflare, the other covers how to design errors that humans and agents can act on.

MCP Events (SEP-3415)

Draft SEP that adds an Events extension to the Model Context Protocol. A client subscribes to upstream events like a Slack message, a repo push or a PagerDuty incident, and the agent reacts when one arrives instead of waiting for a cron job or a user message.

ThinkRail

Open-source agentic IDE from JetBrains, built around the pi coding agent. Every chat gets its own git worktree, and the agent keeps project intent in a graph of living SPEC.md files.

flavors

Kitze's gallery of copy-pasteable design flavors, each a DESIGN.md you hand to your coding agent so the app it builds doesn't look like every other AI-generated landing page.

shadcn/ui Skill

The official shadcn/ui agent skill. It gives coding agents project context, CLI and registry know-how and composition rules, and now has a DESIGN.md workflow that turns a preset plus a brand spec into a fully themed app with a design system page.

break-ui

Emil Kowalski's agent skill that feeds your UI the worst realistic data (long names, unbreakable emails, one-letter names, missing fields, huge counts, empty lists, non-Latin text), puts it behind a "Demo data / Worst case" toggle, and reports what broke along with a fix for each.

OpenSEO

Open-source, pay-as-you-go alternative to Semrush and Ahrefs covering keyword research, rank tracking, backlinks, site audits, and AI visibility, with an MCP server so agents can work with your SEO data.

Whistle

Cactus Compute's on-device speech-to-text model in a single 16.9 MB file that runs on the CPU with no dependencies and mostly beats Whisper base at a fraction of the size.

OpenDots

CopilotKit's open-source, self-hostable template for always-on AI coworkers ("Dots"), each with its own computer, document Spaces, voice calls, and Slack access.

fx

Vercel Labs' small open-source coding agent written in Zig. It ships as one native binary of about 6.7 MiB, needs no background service, works with any model, and its terminal output reads more like a Unix shell than an IDE.

gdp-ts

Guillermo Rauch's TypeScript take on Ghosts of Departed Proofs, made up of a library, lint presets and an agent skill. Sensitive functions require compile-time proofs that the caller actually ran the authorization or entitlement check.

Files SDK

Hayden Bleasel's unified storage SDK for TypeScript. One `Files` class for uploads, downloads, listing, signed URLs, and moving data across 48 storage backends, from S3 and R2 to GCS, Azure Blob, and the local filesystem.

Scrawl

Kit Langton's macOS menu-bar app for showing an agent a problem instead of describing it. Press a shortcut, talk, click and draw over the screen, and paste the result into your coding agent.

Yielded Auth

Composable authentication, sessions, and identity workflows for Effect v4 — one Effect Schema contract from server routes to client, with storage, hashing, and keys supplied as Layers so it fits your existing database.

Foldkit

TypeScript frontend framework built on Effect that uses The Elm Architecture. One immutable Model, typed Messages, a single update function, and side effects returned as Commands.

T3 Code

Theo's open-source control plane for coding agents. Drive Claude Code, Codex, Cursor, OpenCode, Pi and ACP agents on your own machine with your existing subscriptions, from a desktop, web or mobile app.

OpenCode

Open-source AI coding agent from Anomaly (formerly SST) that runs in the terminal, as a desktop app, or as an IDE extension, and works with 75+ LLM providers including local models.

Pi Durable

Durable harness for long-running agents from the team behind the Pi coding agent. It checkpoints every step so a run survives crashes and evictions, and it now runs inside a Cloudflare Durable Object through the Agents SDK's PiHarness.

e2e

e2e's open-source TypeScript test runner for web and mobile apps, where each test can be fully deterministic, fully agent-driven from natural-language goals, or a mix of both.

Mostik

Mostik's stealth-exit writeup on passing hidden states between frozen models through a trained bridge, so a 4B model decodes with a 753B model's latent context instead of its text.

Agent Skills Discovery RFC

Cloudflare's draft spec for publishing Agent Skills at a predictable URL — an index at /.well-known/agent-skills/ lists each skill as a single artifact with a type, a URL and a SHA-256 digest. Apache-2.0, v0.2.0, and a draft rather than a ratified IETF RFC.

PostHog MCP Analytics

Analytics for MCP servers — every tool call an agent makes lands as a normal PostHog event, so you get per-tool failure and latency, the agent's own account of what it was after, and the tools it went looking for and didn't find. Beta, behind a feature preview.

binpatch

Binary delta-update engine from BYK. Publish a bsdiff+zstd patch next to each release and a self-updating binary walks the chain from the installed version to the target instead of downloading itself again. Zero runtime dependencies, MIT.

harness-doctor

Claude Code / Codex skill that mines the last N days of Claude Code transcripts across every project on the machine into five fact tables — bash, errors, tokens, human corrections, fan-out — and returns a ranked list of fixes with the measured number behind each one.

video-use

Edit video by chatting with a coding agent — drop raw takes in a folder, get a cut, graded, subtitled final.mp4 back. The model never watches the footage; it reads a word-level transcript and a rendered timeline.

Cloudflare OS

Cloudflare's internal AI productivity platform, open-sourced. Agents build "gadgets" — private per-user apps written from a prompt — and reach company systems through Gatekeepers, a capability-based permission layer.

Modal

Serverless cloud for AI and data teams — run functions, batch jobs, and GPU inference from Python or JavaScript with no infra to manage.

Herdr

Open-source terminal multiplexer built for coding agents — run Claude Code, Codex, OpenCode and more from one terminal on any box, with persistent PTY sessions you can detach and reattach from any device, even a phone over SSH.

Effect Skill

Kit Langton's opinionated agent skill for building production TypeScript apps with Effect v4 — workflows, services, layers, schemas, config, schedules, caches, streams, HTTP clients, and tests, with current APIs and production defaults.

exe.dev

Cloud for you and your agents — disposable per-second sandboxes, persistent Linux VPS with SSH/HTTPS, and Devbox cloud dev environments.

Latitude.sh

Global AI cloud — deploy bare metal servers and GPU clusters in seconds across 25 locations, with Kubernetes, storage, and managed databases.

context-doctor

Claude Code / Codex skill that audits the fixed context a coding agent auto-loads every session — tool/MCP definitions, connectors, plugins, skills, subagents, memory/rules files — and proposes ranked cuts to prune, gate, or route what's loaded but unused.

integrations.sh

A registry of 5,700+ integration specs from Rhys Sullivan — the same builder behind Executor. Every service normalized into MCP, OpenAPI, GraphQL, and CLI form, each surface leading with a grounded, cited credential guide agents can act on.

Logging Sucks

Boris Tane's argument that traditional logging is broken for distributed systems and that "wide events" — one comprehensive, high-cardinality structured event per request per service — are the fix. Log what happened to the request, not what your code is doing.

memory-view

Claude Code / Codex skill that reads a project's auto-memory vault and generates a self-contained HTML explorer to visualize what the agent has remembered — MEMORY.md plus topic files — without editing or managing it.

session-report

Claude Code / Codex skill that generates a self-contained HTML report debugging what is in a session's context window and how every token is spent — context budget, retained thinking, the dumb-zone cutoff, loaded CLAUDE.md and skills, and full history.

It's Time To Rethink Everything

Theo Browne's CascadiaJS 2026 talk arguing that AI is a "new cloud moment" — just as the cloud removed the cost of provisioning servers, agents remove the cost of building, so the sacred rules of software (file systems, codebases, packages, git, deployment) are worth tearing down and rebuilding from first principles.

Eve

Open-source agent framework from Vercel — define agents as directories of TypeScript and Markdown config and deploy them as standard Vercel projects.

Flue

TypeScript framework for building durable AI agents and workflows with no vendor lock-in — write once, deploy anywhere, run on any LLM.

teach

Matt Pocock's skill that turns a coding agent into a personal tutor — it builds a stateful learning workspace and designs lessons within your zone of proximal development to teach you any skill or concept over multiple sessions.

fff

The fastest and most accurate file search toolkit for AI agents, Neovim, Rust, C, and Node. Keeps an in-memory index for sub-10ms queries instead of spawning ripgrep/fzf per search.

Infisical

Open-source security platform for developers and AI agents — secrets management, certificate management (PKI), and privileged access (PAM) under a single identity model with unified auditing.

Agent Registration with auth.md

Open protocol for autonomous agents to self-register for services without human intervention — a standardized auth.md file + HTTP endpoints replace OAuth and sign-up forms.

netxjs-monorepo

Opinionated Next.js + Turborepo starter with Biome linting, TypeScript, shadcn/ui, and pre-loaded Claude Code agent skills and hooks.

Mediabunny

TypeScript library for reading, writing, and converting media files directly in the browser — like FFmpeg, but for the web. Zero dependencies, tree-shakable, and hardware-accelerated via the WebCodecs API.

deepsec

AI-powered vulnerability scanner that uses coding agents to perform deep security analysis of codebases — regex-based candidate matching followed by LLM-driven investigation, with support for distributed execution and PR diff review.

Learn Harness Engineering

A 12-lecture curriculum on building effective harnesses that enable AI agents to complete complex tasks reliably — covering architecture, state management, session continuity, and observability patterns.

npm-deps-cleanup

AI agent skill that audits and reduces JavaScript package dependency footprint — removes unused deps, deduplicates workspace versions, analyzes transitive closures, and applies e18e replacement recommendations. Works with npm, pnpm, Yarn, and Bun.

How We Test TanStack AI Across 7 Providers

How TanStack AI runs 137 deterministic tests across 7 LLM providers in ~2 minutes on every PR using a mock server instead of real API calls.

Accept Markdown

Protocol and implementation guide for serving Markdown to AI agents via HTTP content negotiation — same URL delivers HTML to browsers, Markdown to AI.

Is It Agent Ready?

Scanner that evaluates website compliance with AI agent standards — checks robots.txt, Markdown negotiation, MCP servers, OAuth, and agent protocols.

noqa

AI-powered QA platform that automates mobile app testing with visual AI agents — write tests in plain language, run across iOS and Android on 1000+ real devices.

SEO Lens

Chrome extension that audits SEO metadata on any webpage with instant scoring, structured data validation, and AI-friendly copyable reports for Claude Code and Cursor.

tfjs-zero-to-hero

Re-implementing classic ML models in TypeScript with @tensorflow/tfjs-node, following Andrej Karpathy's Neural Networks. Zero to Hero. Final destination — a from-scratch two-tower retrieval model on MovieLens.

Using Claude Code: The Unreasonable Effectiveness of HTML

Thariq makes the case for HTML over Markdown as Claude Code's output format — richer visualizations, color, diagrams, and easier sharing once agents are the ones authoring and humans rarely hand-edit.

Pullfrog

AI PR review agent from Zod's creator. Dynamically scales effort, uses subagent swarms for large PRs, and runs inside GitHub Actions with batteries-included MCP tools.

quality-code

Rhys Sullivan's TypeScript coding skill — branded types, discriminated unions, end-to-end types, real tests over mocks, OpenTelemetry observability, and picking the right abstractions instead of premature ones.

tabular-nums should be the default for any updating number

Soren makes the case for `font-variant-numeric: tabular-nums` as the default for timers, counters, prices, percentages, scores, and any live data — equal-width digits stop the layout from twitching when numbers change.

Executor

Open-source tool gateway for agents from Rhys Sullivan. Normalizes MCP, OpenAPI, GraphQL, and custom sources into one typed SDK that agents call in a sandbox — built on Cloudflare's Code Mode pattern.

Hex

Open-source, fully native macOS voice-to-text app using local Parakeet TDT v3 models with zero network calls. Press-and-hold a hotkey to record, release to paste the transcript anywhere.

improve-codebase-architecture

Matt Pocock's skill for deepening modules — restructuring prototype code into something long-lived with cleaner layer dependencies and easier navigation.

Google Search (AI SDK)

Vercel AI SDK tool that grounds Gemini model responses with live Google Search results and citations.

Amazon SES

AWS cloud email service for sending transactional and marketing messages via API or SMTP at pay-per-use pricing.

Axiom

Observability platform for logs, traces, and AI engineering telemetry with petabyte-scale ingestion and zero sampling.

Bloom.diy

AI-powered mobile app builder that turns ideas into native Android apps without writing code.

Browser Use

Open-source browser automation harness that lets AI agents navigate and interact with websites at scale.

Cloudflare AI

Serverless GPU inference and agent infrastructure running on Cloudflare's global edge network.

Cloudflare Email Routing

Free email routing service that forwards custom domain addresses to any inbox with optional Workers processing.

Drizzle ORM

TypeScript ORM with a SQL-like query API, type-safe schema definitions, and multi-database support.

Effect Institute

Interactive online learning platform with narrated video tutorials for the Effect-TS library.

Effect Solutions

Prescriptive guide and reference for writing idiomatic Effect-TS programs with setup, patterns, and ecosystem docs.

Effect

TypeScript library for building robust apps with typed errors, structured concurrency, retries, and first-class OpenTelemetry observability.

Exa

Search API built for AI agents, returning fast, high-quality web results with structured content extraction.

Fallow

CLI tool that analyzes TypeScript and JavaScript codebases to find unused exports, dead code, and complexity hotspots.

Firecrawl

Web crawling and scraping API that converts any website into LLM-ready markdown or structured JSON.

Google CheckMX

Google Apps Toolbox utility that checks MX, SPF, and DKIM DNS records to diagnose email delivery configuration issues.

GPTZero

AI content detection platform that identifies text generated by ChatGPT, Claude, Gemini, and other LLMs.

Groq

API provider for open-weight LLMs and Whisper speech-to-text at extremely high inference speeds.

Inboxy

Email warmup service that uses AI and a private inbox network to boost domain reputation and deliverability.

Inkless

Free e-signature platform for uploading, signing, and sending unlimited documents without a subscription.

Instantly

Cold email and outreach platform with built-in email warmup and AI-powered campaign automation.

IPWarmup

Provides pre-warmed SMTP servers with established sender reputation so campaigns can start at full volume.

Langfuse

Open-source LLM engineering platform for tracing, prompt management, evaluations, and production monitoring.

Lighthouse HQ

Immigration platform that guides international professionals through U.S. visa applications with document prep and expert support.

MailReef

Dedicated cold email sending infrastructure with private SMTP servers, dedicated IPs, and unlimited mailboxes.

Mem0

Persistent memory infrastructure for AI agents that compresses context and reduces token usage across sessions.

Models.dev

Open-source database of AI model specs, pricing, and capabilities queryable via a JSON API.

OpenPhone

Business phone system that provides virtual numbers, shared calling, and texting for teams in one app.

OpenRouter

One API to access 400+ LLMs from 60+ providers with automatic fallback and unified billing.

Perplexity API

API platform providing real-time web search and grounded LLM responses with citations via Sonar models.

PlanetScale

Managed MySQL and PostgreSQL database platform with branching workflows for zero-downtime schema changes.

Ragie

Managed RAG-as-a-service that ingests documents and provides hybrid search APIs for AI applications.

React Doctor

Static analysis tool that evaluates React codebases across security, performance, correctness, and architecture, producing a health score with diagnostics.

Reply.io

Multichannel sales engagement platform that automates outreach across email, LinkedIn, SMS, and calls.

Resend

Developer-first email API for sending transactional and marketing email, with multi-language SDKs, a CLI, and React Email templating.

Scira

Open-source agentic research platform that plans, retrieves, and returns cited answers across 17 search modes.

ScrapingBee

Web scraping API that handles headless Chrome rendering, proxy rotation, and anti-bot bypassing automatically.

Smartlead

Cold email platform with unlimited mailbox warmup, AI-driven sending patterns, and deliverability monitoring.

Stagehand

Open-source SDK for building browser agents using natural language instructions instead of brittle selectors.

Supermemory

Context and memory infrastructure for AI agents with hybrid vector search, user profiling, and connectors for Notion, Slack, Google Drive, and more.

telegram-claude

Telegram bot interface for Claude Code on a VPS. Message the bot, it spawns Claude Code in the selected project directory and streams results back — voice notes, streaming responses, session continuity, plan-mode interception.

Turso

SQLite at the edge with unlimited lightweight databases, concurrent writes, and no cold start penalties.

Ultracite

Zero-config preset that wires up linters, formatters, editor settings, and AI-agent rule files for JavaScript and TypeScript projects.

Unipile

Unified API for reading and sending messages across LinkedIn, email, WhatsApp, Telegram, and calendars.

agent-browser

Rust CLI tool for browser automation designed for AI agent workflows with JSON output and screenshot support.

agent-eval

Testing framework for measuring how well AI coding agents perform against your framework or configuration.

Vercel AI Gateway

Unified endpoint to route requests across hundreds of AI model providers with built-in failover and observability.

emulate

Local API emulator that provides stateful, production-fidelity stubs for services like GitHub, Google, and Slack.

json-render

Generative UI framework that lets AI produce dynamic interfaces constrained to a developer-defined component catalog.

just-bash

Simulated bash environment with a virtual filesystem for use in AI agent workflows.

portless

Replaces localhost port numbers with stable named URLs and enables HTTPS and HTTP/2 for local dev servers.

zero-native

Framework for building desktop and mobile apps with Zig backends and web frontends via platform WebViews.

Verifalia

Email validation service that identifies invalid and undeliverable addresses to reduce bounce rates and protect sender reputation.

VERT

Free, privacy-first file converter that runs entirely in the browser using WebAssembly. Supports images, videos, audio, documents, and more with no uploads required.

Zenblog

Headless CMS with a Notion-style editor, hosted media uploads, and a type-safe API client for developer-managed blogs.

auto-pr

Local Node.js loop that turns GitHub issues into pull requests — label an issue `auto-pr` and a multi-step Claude pipeline (research → plan → implement → review) spawns Claude Code and opens a PR.