AI agents

microsoft/AI-Engineering-Coach

A local dashboard that scores how well you work with AI coding tools — AI Engineer Coach parses local AI coding session logs from any harness into a single dashboard, with 45 rules to spot anti-patterns in prompting, session hygiene, and context use. It also surfaces reusable skills from repeated prompts and tracks code output by model and workspace—all without sending data off your machine.

better agentic engineering

KnockOutEZ/wigolo

Local-first web search and crawling for AI agents, zero API keys — wigolo gives coding agents search, fetch, crawl and research via MCP without cloud APIs or per-query costs. It plugs into Claude Code, Cursor, Codex, VS Code and more, positioning itself as a local alternative to Exa, Firecrawl and Tavily during its public beta.

The go-to web for your AI coding agent — local-first search, fetch, crawl & research over MCP. No API keys, no cloud, $0/query. Public beta.

razorpay/slashdocs

Turn agent output into reviewable HTML docs, not chat dumps — slashdocs is a self-hosted server, CLI, and remote MCP endpoint where AI agents publish self-contained HTML reports, specs, and diagrams. Humans comment directly on the text, agents read feedback and push new versions, and every version is kept immutable. Built for teams mixing Claude Code or other agents with human reviewers.

Give your agents the power to publish reviewable HTML documents. Self-hosted server + CLI + remote MCP; agents push, humans review with comments and versions.

BuilderIO/agent-native

A framework pairing autonomous agents with purpose-built UIs — Agent-Native is an open-source TypeScript framework from Builder.io for building agents that work alongside a dedicated UI. Developers define each capability once as an action, usable both as an agent tool and as a UI function, sharing validation and permissions. Useful for teams building knowledge-work agents that need inspectable, editable results, not just a chat box.

A framework for building agentic apps

weave-os/router

A fast routing layer that cuts LLM costs without code changes — Weave Router sits between agentic apps and multiple LLM providers, picking the right model per prompt in under 50ms. Written in Go and OpenAI-compatible, it works with tools like Claude Code and Codex, letting teams cut inference costs 40-70% by just swapping an endpoint.

Model router for agentic systems. Routes every prompt to the right model in <50ms. Cut costs 40-70% with just an endpoint change.

Amal-David/awesome-jev

A curated map of Jev demos, SDKs, and agent tooling — This awesome-list gathers demos, SDKs, browser/desktop tools, and reading material around Jev, TypeSafe's probability-based yes/no judgment model. It's aimed at developers building typed classifiers or agents, with source links and notes on setup, privacy, and evidence limits rather than just hype.

Jev demos, projects, SDKs and skills, with source links and a curated X gallery.

Significant-Gravitas/AutoGPT

AutoGPT now ships a visual builder for self-hosted AI agents — AutoGPT lets users describe a desired outcome or design every step visually, then run agents on demand, on a schedule, or via triggers. With over 185,000 stars and a self-host option, it suits developers and non-coders who want autonomous task completion without relying on a hosted platform alone.

AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.

OpenBMB/PilotDeck

OpenBMB's WorkSpace-based platform for AI agents with memory — PilotDeck is a task-oriented productivity platform from OpenBMB organizing AI agent work into WorkSpaces, with native MCP support for tool integration. It focuses on operational boundaries and memory evolution, aimed at developers building persistent, context-aware agent workflows rather than one-off chat tasks.

Task-oriented AI Agent productivity platform

techjarves/ClaudeCode-Portable

A self-contained Claude Code workspace that runs straight off a USB stick — ClaudeCode-Portable bundles Anthropic's official Claude Code CLI/Agent SDK with a portable Node.js runtime, letting you launch it from a USB drive or folder with no install. It supports nine AI providers (OpenRouter, Gemini, NVIDIA NIM, Ollama, custom APIs, etc.) and adds a browser dashboard with streaming chat and shared conversation history.

Run Claude Code from a USB drive on any PC — no installation required.

eyaltoledano/claude-task-master

A task-management layer that plugs into any AI coding chat — Taskmaster brings structured task management to AI-driven development, integrating with Cursor, Lovable, Windsurf, Roo and other AI chat tools. It's built for developers coordinating AI-assisted coding workflows who need a shared, trackable task list across sessions and tools rather than ad-hoc prompts.

An AI-powered task-management system you can drop into Cursor, Lovable, Windsurf, Roo, and others.

aipoch/open-science

A local-first desktop workbench for running reproducible science with AI agents — AIPOCH Open-Science is an open-source desktop app pairing AI agents with Python/R execution, MCP tools and connectors to keep research reproducible and traceable. Model-agnostic and local-first, it targets researchers wanting agent workflows without giving up control over data or tooling, running on macOS, Windows and Linux.

The open-source AI research workbench for scientific research and agent workflows. Local-first, model-agnostic desktop app with extensible skills, MCP tools and connectors, Python/R execution and traceable artifacts for reproducible research on macOS, Windows and Linux.

anthropics/claude-quickstarts

Starter templates for building apps on Anthropic's Claude API — A set of ready-to-customize TypeScript projects for building with Claude, including a customer support agent, a financial data analyst with chat-driven visualizations, and demos of computer-use agents (containerized and native macOS). Useful for developers who want a working foundation rather than starting from scratch.

A collection of projects designed to help developers quickly get started with building deployable applications using the Claude API

feder-cr/invisible_playwright_mcp

A Playwright MCP agent built to slip past anti-bot defenses — This MCP server drives a stealth Firefox browser that AI agents like Claude Code or Gemini CLI can use to browse without triggering anti-bot systems or captchas. It plugs into assistants via MCP for tasks like job hunting, scraping, and general computer use.

Playwright MCP server undetected by anti-bots and captchas: AI agent browses the web on anti-detect stealth Firefox, Python, undetected browser automation, scraping, computer use.

stepfun-ai/Step-Code

A terminal agent built around Step's own models and workflow — Step Code is a terminal-based coding agent tied to StepFun's Step models, handling file edits, test runs, and long tasks via /goal and /cron. It adds one-command static site publishing (StepPage) and claims Claude Code plugin compatibility plus token-efficient subagent orchestration for long-horizon work.

pacifio/atlas

Git-style checkpoints for every coding agent you run — Atlas tracks coding-agent sessions as checkpoints, linking commits back to the prompts, tool calls, and reasoning that produced them. Run Claude Code, Codex, or any ACP-registry agent side by side on the same codebase with shared memory, then query exactly which agent did what.

Source control for agents. Use multiple coding agents, track their changes and query them in one place

outcrawlai/scout

An open-source agent that drives a real browser for you — Scout takes a plain-language task, controls a real browser via CDP to click, type and scroll through pages, then returns the result or reports what's missing. It runs locally or on Outcrawl's cloud, which adds captcha solving and run replays, and supports bringing your own LLM models.

Scout: the open-source browser agent, by Outcrawl. Mirrored automatically from Outcrawl's own repository.

HelpCode-ai/anythingmcp

Self-hosted gateway that wraps 517 APIs as MCP tools — AnythingMCP connects REST/OpenAPI, SOAP, GraphQL, OData and SQL systems to Claude, ChatGPT and Copilot without writing a custom MCP server. It ships 517 ready-made connectors, including SAP, Etsy and Amazon Seller, with 23 needing no API key, plus RBAC, SSO and audit logs for enterprise use.

MCP gateway for AI agents: turn any REST/OpenAPI, GraphQL, SOAP or SQL API into MCP tools. 562 ready-made connectors (SAP, ERP, e-commerce) with RBAC, SSO, audit logs and a knowledge graph.

fhshaik/typesafe-mario

Letting a language model play Mario using only structured JSON state — Instead of feeding screenshots to an AI, this project converts NES RAM and telemetry into compact object-centric JSON (Mario's motion, enemies, terrain) for TypeSafe's Jev model to choose controller inputs. A live dashboard shows action probabilities, confidence, and latency, making it useful for anyone exploring structured-state game agents.

A TypeSafe/Jev agent that plays Super Mario Bros. from structured emulator state.

christianmat/jev-pokemon

An AI model clears Pokémon Red with no scripts or cheats, just RAM reads — Jev, TypeSafe AI's decision model, finished Pokémon Red in 37h 40m over 16,150 decisions, costing about $1.65. A harness only reads memory and presses buttons; every menu, battle, and route choice is Jev's. Interesting for anyone curious how cheaply an LLM-driven agent can beat a full game.

Jev, an AI decision model, plays Pokémon Red. It beat the game in 37h 40m.

dukesun99/Corpus2Skill

Turning document corpora into navigable skill trees for agents — Official EMNLP 2026 Findings code for Corpus2Skill, which compiles enterprise document corpora into a hierarchy of Anthropic Skills. Agents navigate SKILL.md/INDEX.md files and fetch documents on demand at query time, replacing vector-search or BM25 serving. Useful for researchers and builders exploring alternatives to embedding-based RAG for enterprise QA.

Official Findings of EMNLP 2026 implementation of Corpus2Skill: compile a document corpus into a navigable skill hierarchy that LLM agents explore at query time, with document lookup instead of a serving-time vector-search service.

brumar/chess-postmortem-skills

Turn a chess game plus your spoken thoughts into a narrated post-mortem — These Claude Code skills combine a Stockfish-backed annotated PGN, HTML viewer, and narrated video to explain your chess mistakes in plain language. It even matches a recorded voice memo of your in-game thinking to each move via whisper.cpp transcription and clock times, then answers your own questions with the engine.

Claude Code skills for chess post-mortems: Stockfish-backed annotated PGN, HTML viewer, narrated video

lowenbjer/claude-terse

A Claude Code plugin that strips the essay tone out of replies — This plugin targets Claude Code's tendency to answer every question with essays, metaphors, and

For those who struggle with how Claude speaks to them. No more essays. No more slogans, metaphors, "it's not X, it's Y". This plugin makes Claude speak normally.

sagochiko/aws-drawio-diagram-skill

A Claude Code skill that drafts clean AWS diagrams in draw.io — Describe an AWS setup as bullet points and this Claude Code skill produces a .drawio file with official AWS icons, skipping decorative titles or legends. Layout rules (nesting, labels, line routing, colors) live in the skill itself, so requests only need to state the architecture. It can also edit existing .drawio files, e.g. adding a WAF.

AWS の構成図を draw.io の形式で描く Claude Code の Skill / Claude Code Skill that draws AWS architecture diagrams as draw.io files

bertaye/agentic-cuda-optimizer

An LLM agent that iteratively rewrites and benchmarks CUDA kernels — A LangGraph-driven agent that generates, compiles, and benchmarks CUDA kernels in a loop, using a C++/NVRTC harness for correctness checks and Nsight Compute counters for optimization clues. It tunes both kernel code and launch configs, keeping the fastest validated version. Useful for GPU developers exploring automated kernel tuning.

A langgraph based workflow with a C++ CUDA harness to optimize CUDA kernels

microsoft/work-iq-samples

Multi-language client samples for Microsoft's Work IQ API — A sample collection showing how to connect to Microsoft's Work IQ Gateway, the AI-native interface to Microsoft 365 work intelligence, using C#, Rust, and Swift clients over A2A protocol or REST. Useful for developers building agents against Copilot Chat and Work IQ who need working auth and protocol examples across platforms.

Provides samples for use with the work-iq apis

AlterAIDev/openclaw-alter-plugin

A shared OpenClaw agent that keeps everyone's own accounts separate — This plugin lets one OpenClaw bot in Slack or Discord act on behalf of each team member individually, pulling from their own Gmail, calendar, or other connected accounts with per-user OAuth and human-in-the-loop approval. Alter Vault holds the credentials, so the agent itself never stores passwords or tokens. Useful for teams wanting a single shared bot without pooling everyone's identity.

Multi-user identity, OAuth grants and HITL approvals for OpenClaw, backed by Alter Vault

tidbcloud/ti-cli

A CLI that gives AI agents persistent storage and instant MySQL — ti is TiDB Cloud's preview CLI for managing a serverless POSIX filesystem and MySQL-compatible database clusters. It targets AI coding agents and sandboxes, letting them persist state, share files across sessions, and spin up HTAP SQL clusters in a few commands without managing infrastructure.

callstack/agent-device

A Mobile Device Feedback Loop Built for AI Coding Agents — agent-device gives coding agents like Claude Code, Codex, and Cursor a live feedback loop on iOS, Android, HarmonyOS, TV platforms, web, macOS, and Linux. Agents read token-efficient accessibility snapshots instead of screenshots, act via refs/selectors, and save evidence, with support for parallel worktrees and remote device clouds. Teams at Expensify and Shopify already use it for bug verification.

Open source device layer for AI coding agents: a CLI, MCP server, and typed Node.js API to control mobile, web, TV, and desktop apps on devices and simulators, locally or in the cloud.

firebase/firebase-tools

Firebase CLI now doubles as an MCP server for AI coding agents — The long-standing Firebase CLI for deploying, testing, and managing Firebase projects now bundles an official MCP Server, letting AI assistants query and act on Firebase resources directly. It also integrates with Firebase agent skills, giving coding agents built-in knowledge of CLI commands and best practices.

The Firebase Command Line Tools

amosblomqvist/learnharness

An AI tutor that maps your reasoning before teaching you anything — LearnHarness is an agent harness that diagnoses exactly where your understanding ends, plans a dependency path to your goal, and teaches step by step using your own answers and reasoning. A persistent knowledge graph tracks progress across sessions, and separate practice sessions adapt problems to how you think, aimed at learners of hard technical material.

Agent harness for learning difficult things.