Best Claude Code Tools & Skills

learn-claude-code

shareAI-lab/learn-claude-code · ★ 80k ▲371/7d · Python

Learn Claude Code is a 17-chapter Python course that builds a small coding-agent harness one mechanism at a time. It explains the machinery around a model, including tools, permissions, hooks, memory, teams, MCP, saved workflows, and goal checks; English is the default, with Chinese and Japanese translations.

Verdict Our run installed 50 packages in 19 seconds and passed all 482 tests in 45 seconds, making Learn Claude Code an unusually easy agent course to verify locally. Use it when you want the seams of an agent harness exposed in small Python chapters.

claude-code-best-practice

shanraisshan/claude-code-best-practice · ★ 68k ▲262/7d · HTML

This is a comprehensive collection of best practices, guides, and configuration examples for Claude Code, an advanced AI coding assistant. It serves as a community-driven manual to help developers master the tool's powerful features, from basic settings to complex 'agentic engineering' workflows.

Verdict This repository is an indispensable, S-tier resource for any serious Claude Code user. It's the missing manual that expertly bridges the gap between official documentation and real-world mastery.

goose

aaif-goose/goose · ★ 56k ▲276/7d · Rust

Goose is a versatile, open-source AI agent that runs locally on your machine. It provides a unified interface, as a desktop app, a command-line tool, or an API, to interact with over 15 different AI model providers, from OpenAI to local models via Ollama. It aims to be the central hub for all your AI-powered tasks, including coding, research, and workflow automation.

Verdict Goose is an exceptional tool that delivers on its promise of being a universal AI agent. Its multi-platform support (desktop, CLI, API) and vast provider compatibility make it a powerful central hub for anyone who uses AI seriously.

impeccable

pbakaus/impeccable · ★ 97k ▲9.4k/7d · JavaScript

Impeccable is a design skill and command set for AI coding tools such as Claude Code, Codex, Cursor, and Gemini CLI. It gives an agent a shared design vocabulary, stores product context in project files, and can check rendered interfaces for recurring visual and accessibility mistakes.

Verdict Our Impeccable run built in 11 seconds, but 1 of 449 tests failed, so I would use its design guidance while keeping live edits behind Git review. Its 24 commands and 61 detector rules give agent-heavy frontend work a much better vocabulary than a loose request to make a page look good.

netdata

netdata/netdata · ★ 81k ▲101/7d · Go

Netdata is an infrastructure monitor that collects, stores, charts, and alerts on per-second data from servers, containers, applications, hardware, and network devices. It gives operators an immediate local view, then scales through parent nodes and an optional cloud service for larger fleets and AI-assisted investigations.

Verdict Our Netdata run installed 52 Python packages and built in 6 seconds, but pytest collected 0 tests and pip-audit found 1 known vulnerability, so the measured packaging path is easy to build but not release-clean. Netdata is a strong choice for a lean operations team that wants useful charts and anomaly signals before assembling a Prometheus and Grafana stack.

system_prompts_leaks

asgeirtj/system_prompts_leaks · ★ 70k ▲293/7d · JavaScript

This repository is a massive, curated collection of system prompts, the secret, underlying instructions given to large language models like ChatGPT, Claude, and Gemini before they interact with a user. It solves the mystery of how these AIs are configured to behave by providing a raw, behind-the-scenes look at their operational rules, personalities, and limitations.

Verdict This repository is an essential, must-watch resource for anyone serious about understanding the current state of generative AI. While its 'leaked' nature means you should approach the content with critical analysis, the sheer volume, meticulous organization, and relentless updates make it an unparalleled treasure trove for researchers, developers, and the deeply curious.

context7

upstash/context7 · ★ 64k ▲304/7d · TypeScript

Use Context7 when your coding agent needs library documentation and you accept a hosted indexing service. Context7 retrieves documentation and code examples through CLI skills or MCP tools, with SDK integrations available. Our run on September 30, 2026, at commit 83e972e installed and built successfully, but the test command exited 1. We did not establish successful authenticated retrieval, editor setup, or deployment.

Verdict Context7 is worth evaluating for documentation retrieval inside coding agents when a hosted index fits your requirements. Our run verified installation and compilation, while the test command failed and authenticated retrieval remained unverified.

awesome-claude-code

hesreallyhim/awesome-claude-code · ★ 57k ▲395/7d · Python

Awesome Claude Code is an edited directory of guides, skills, plugins, agents, status lines, and other projects built around Claude Code. It gives you one categorized place to find candidates that would otherwise be scattered across GitHub and personal sites.

Verdict Our Awesome Claude Code run installed 51 packages in 26 seconds and passed all 112 tests, so its publishing machinery is unusually well checked for a link directory. Use it to build a Claude Code shortlist, then inspect each chosen project's code, permissions, activity, and license yourself.

codebase-memory-mcp

DeusData/codebase-memory-mcp · ★ 48k ▲393/7d · C

codebase-memory-mcp is a local code index that gives AI coding agents a map of functions, calls, imports, routes, and files instead of making them reopen the repository for every question. It stores that map in SQLite, exposes it through MCP and a CLI, and does not send the code to a hosted service.

Verdict Our sandbox installed 352 packages in 31 seconds, built in 17 seconds, and passed all 47 UI tests, so codebase-memory-mcp is a credible choice for developers who want a local structural map inside their coding agent. Use it for navigation, call tracing, and change-impact clues, then confirm consequential answers against the source.

Handy

cjpais/Handy · ★ 35k ▲787/7d · Rust

Handy is a desktop dictation app that records your speech, transcribes it on your computer, and pastes the words into the active text field. It supports Windows, macOS, and Linux with local Whisper or Parakeet models, so basic dictation does not require a speech API or send recordings to a hosted service.

Verdict Our Handy run installed 432 packages in 52 seconds and built in 20 seconds, but it exposed no test target for our harness to run. Try the packaged release if private, system-wide dictation is worth a model download and a few permissions on your computer.

Anthropic-Cybersecurity-Skills

mukul975/Anthropic-Cybersecurity-Skills · ★ 35k ▲296/7d · Python

This is a massive, open-source library of over 800 cybersecurity skills designed to be plugged into AI agents. It solves the 'blank slate' problem, giving a generic AI the specialized, structured knowledge of a senior security analyst for tasks like incident response, threat hunting, and forensics.

Verdict Anthropic Cybersecurity Skills is an indispensable, foundational library for anyone building serious AI agents for security. It's not a complete solution, but it provides the most comprehensive, well-structured, and up-to-date knowledge base available.

mcp-for-blender

MCPBlender/blender-mcp · ★ 33k ▲490/7d · Python

MCP for Blender, formerly blender-mcp, lets an MCP client inspect and change a live Blender scene. A local Python server talks to a Blender add-on, so the model can create objects, edit materials, import assets, export files, and run Blender Python.

Verdict Our MCP for Blender run installed 62 packages in 13 seconds and passed 263 tests, so it deserves a supervised trial from technical Blender users. Keep the socket on localhost, turn on safe mode, save before prompting, and disable the default minimal telemetry if policy requires it.

mcp-for-blender

ahujasid/blender-mcp · ★ 33k ▲490/7d · Python

MCP for Blender, formerly blender-mcp, lets an AI client inspect and change a live Blender scene. Its local Python server passes requests to a Blender add-on for object edits, materials, asset imports, exports, screenshots, and Blender Python.

Verdict Our MCP for Blender run installed 62 packages in 11 seconds, used 68 MB, and completed pytest with 263 passed and 0 failed, so the pinned Python package is easy to trial. Use it for supervised scene work where a person can watch the viewport, save often, and reject a bad action.

gitleaks

gitleaks/gitleaks · ★ 31k ▲206/7d · Go

Gitleaks is a command-line scanner for passwords, API keys, and tokens in Git history, files, directories, or piped text. It gives developers a local and CI-friendly way to catch likely credentials before they spread, but it detects patterns rather than proving that every secret is valid or that none were missed.

Verdict Our Gitleaks run built in 35 seconds and passed all 14 Go tests in 14 seconds, making it an easy first secret-scanning gate for local development and CI. Use it if you will inspect warnings, seed test credentials, and review every suppression.

crush

charmbracelet/crush · ★ 29k ▲80/7d · Go

Crush is a terminal coding agent that can edit a project, run tools, use language servers, and connect to several model providers. It gives developers one text interface for coding sessions without tying the workflow to a single model company.

Verdict Our Crush run passed all 118 tests, but install, build, and test took 627 seconds combined, so this polished terminal client still asks for patience when you work on its source. Use it if model choice, MCP connections, and an interactive terminal matter to you.

awesome-artificial-intelligence

owainlewis/awesome-artificial-intelligence · ★ 17k ▲23/7d · Python

Awesome Artificial Intelligence is an English-language shortlist of books, courses, papers, tools, and operating advice for developers building AI systems. It solves the blank-page problem by making a small set of editorial choices instead of collecting every AI link it can find.

Verdict Our run installed 35 packages in 15 seconds and passed all 12 tests, but those checks validate the list machinery rather than the merit of every recommendation. Use this repository when you want a disciplined starting map for AI engineering and are comfortable with an opinionated editor.

transformers.js

huggingface/transformers.js · ★ 16k ▲12/7d · JavaScript

Choose Transformers.js 4.3.0 for a browser inference prototype if its supported pretrained models fit your task and you can test the intended devices. Transformers.js combines pretrained models with input preprocessing and output postprocessing through JavaScript pipelines. Our run on September 18, 2026, checked commit a1728dd. Installation and build succeeded in our sandbox, but the test command failed. We did not validate a deployed browser application, inference quality, WebGPU compatibility, offline packaging, or sustained workload memory use.

Verdict Transformers.js 4.3.0 is worth prototyping when a supported pretrained model and JavaScript pipeline fit your application. Our September 18, 2026, run at a1728dd established successful installation and build in our sandbox, with an unsuccessful test command and two SIGKILL-interrupted suites.

llm

simonw/llm · ★ 13k ▲16/7d · Python

LLM is a Python command-line tool and library for sending prompts to hosted or local language models through one workflow. It also stores conversations in SQLite, handles attachments and structured output, runs tools, creates embeddings, and grows through provider and feature plugins.

Verdict Our LLM install used 60 packages and 67 MB, but its tests stopped in 8 seconds because `llm_echo` was missing, so the source checkout is easier to use than to verify from our measured setup. LLM is the best fit here for an engineer who wants one scriptable command, durable SQLite history, and a broad plugin vocabulary across hosted and local models.

awesome-agentic-ai-zh

WenyuChiou/awesome-agentic-ai-zh · ★ 8.3k ▲314/7d · Python

awesome-agentic-ai-zh is a Traditional Chinese-first learning roadmap for AI agents, with complete English and Simplified Chinese editions maintained alongside it. It organizes concepts, projects, and small exercises into separate paths for using agent CLIs and building agent systems.

Verdict Our run installed 35 packages in 14 seconds, built the documentation in 4 seconds, and found 0 known vulnerabilities, making the curriculum easy to inspect locally. Choose it if you want a maintained, trilingual route through CLI agents, Claude Code, MCP, and multi-agent systems.

Awesome-MCP-ZH

yzfly/Awesome-MCP-ZH · ★ 7.8k ▲22/7d

This is a comprehensive, curated directory of resources for the Model Context Protocol (MCP), an open standard that lets AI models interact with external tools and data. It's built specifically for a Chinese-speaking audience, collecting everything from beginner explanations and client software recommendations to a massive, categorized list of MCP servers that give AI new capabilities. It solves the discovery problem for anyone in the Chinese tech community looking to build with or use AI agents.

Verdict For any Mandarin speaker interested in the AI agent ecosystem, Awesome-MCP-ZH is an indispensable and brilliantly executed resource. It's more than a list; it's a curated gateway that provides context, practical advice, and a comprehensive map of a complex new field.

Unity-MCP

IvanMurzak/Unity-MCP · ★ 4.6k ▲26/7d · C#

Unity-MCP connects an MCP-capable coding agent to the Unity Editor and, through custom C# tools, to a compiled game. The agent can inspect scenes, change assets and scripts, run tests, read profiler data, and call project-specific methods instead of stopping at code suggestions.

Verdict Our Unity-MCP run installed 55 packages in 7 seconds and passed 605 of 611 tests, with the other 6 skipped and none failed. That makes the CLI an easy trial for a Unity team that wants an agent inside the editor, though it does not prove the agent will make good scene or code changes.

kubernetes-mcp-server

containers/kubernetes-mcp-server · ★ 2.2k ▲27/7d · Go

kubernetes-mcp-server lets MCP clients inspect and operate Kubernetes and OpenShift through a native Go server. It talks directly to the Kubernetes API and exposes tools for resources, pods, logs, metrics, Helm, multiple clusters, and optional ecosystem integrations.

Verdict Our kubernetes-mcp-server run installed 393 Go packages, built successfully, and passed all 36 tests in 102 seconds, the strongest clean build result in this batch. Use it for supervised cluster diagnosis with a dedicated read-only ServiceAccount, a narrow toolset, and sensitive kinds denied.

ui-ux-pro-max-skill

nextlevelbuilder/ui-ux-pro-max-skill · ★ 143k ▲2.7k/7d · Python

UI UX Pro Max is a design-reference skill for Claude Code and other coding agents. It turns a product description into a structured design brief with layout patterns, styles, palettes, type pairings, stack guidance, and checks that the agent can use while writing interface code.

Verdict Our UI UX Pro Max run installed 32 packages in 15 seconds, built in 5 seconds, and returned 0 audit findings, making it cheap to trial in an agent-driven frontend project. Use it to produce a design brief and reusable project rules, especially when you otherwise start from a vague prompt.

agent-skills

addyosmani/agent-skills · ★ 112k ▲1.5k/7d · JavaScript

Agent Skills is a collection of 25 Markdown workflows that tell AI coding agents how to specify, plan, build, test, review, and ship software. It gives teams reusable engineering rules instead of making each developer repeat the same instructions in every chat.

Verdict Our sandbox installed 0 packages in 6 seconds and passed 1 test in 2 seconds, so Agent Skills is cheap to inspect but only lightly proven by the path we ran. Use it as an editable set of working agreements, starting with the two or three workflows tied to failures your team already sees.

pytorch

pytorch/pytorch · ★ 105k · Python

PyTorch is a Python framework for working with multidimensional arrays, calculating gradients, and building neural networks on CPUs and several kinds of accelerators. It lets researchers change model code with ordinary Python while giving engineering teams a path into compiled and distributed execution.

Verdict Our run installed 62 packages in 747 seconds, then the source build failed and pytest stopped at 200 collection/setup errors, so application teams should use a pinned official binary unless framework contribution is the job. PyTorch is still the sensible choice when a model or production stack already depends on its APIs and accelerator support.

Front-End-Checklist

thedaviddias/Front-End-Checklist · ★ 75k ▲45/7d · MDX

Front-End Checklist is a catalog of 385 English rules for checking the parts of a website that teams forget before release, including accessibility, performance, SEO, and security. You can read it on the web, use the generated README, or let an AI agent query the same rules through its MCP server and installable skills.

Verdict Our Front-End Checklist run built in 124 seconds and passed all 20 Jest tests in 39 seconds, but the contributor install still occupied 1,421 MB. Use the README or web catalog as a release memory aid, and add MCP when searchable agent access to 385 rules earns its service boundary.

ppt-master

hugohe3/ppt-master · ★ 71k ▲3.2k/7d · Python

PPT Master is a Python workflow that lets an AI coding agent turn documents, topics, or existing decks into editable PowerPoint files. It solves the usual AI-slide problem by producing native shapes, text, charts, tables, masters, notes, and motion that you can continue editing in PowerPoint.

Verdict Our PPT Master run installed 110 packages in 32 seconds and built in 15 seconds, but it exposed no test target, so it is easier to install than to admit into a strict release gate. Use it when editable PowerPoint structure is worth running an agent and checking the deck yourself.

ultralytics

ultralytics/ultralytics · ★ 63k ▲160/7d · Python

Ultralytics is a Python package and command-line tool for training, validating, exporting, and running YOLO computer-vision models. It puts detection, segmentation, classification, pose, depth, oriented boxes, and video tracking behind one interface.

Verdict Our Ultralytics run installed 109 packages and occupied 5,982 MB, then its tests stopped on missing `libGL.so.1`, so the friendly Python API still needs a deliberate runtime image. Use it when one maintained interface for several YOLO tasks saves more work than a custom vision stack, and settle the AGPL-3.0 or Enterprise licensing choice before the prototype becomes a product.

humanizer

blader/humanizer · ★ 63k ▲2.1k/7d · Python

Humanizer is an English-language skill that tells an AI agent how to remove common signs of machine-written prose without changing the underlying claims. Its 26-pattern checklist works in Claude Code, Codex, and other agents that support skills.

Verdict Our Humanizer run installed 35 packages in 10 seconds, built in 4 seconds, and found no test target, so setup is trivial while rewrite reliability remains a manual judgment. Use it as an inspectable editing checklist when a person will review the diff.

litellm

BerriAI/litellm · ★ 62k ▲379/7d · Python

LiteLLM gives a Python application or a shared gateway one interface for calling many model providers. It centralizes routing, fallbacks, credentials, virtual keys, spending controls, and request records so each application does not need its own provider integrations.

Verdict Our LiteLLM run installed 379 packages in 22 seconds with 0 known vulnerabilities, but the root project offered no build or test target, so it proved dependency resolution rather than release readiness. Use LiteLLM when several teams and providers need one policy point for keys, spend, routing, and guardrails.

chrome-devtools-mcp

ChromeDevTools/chrome-devtools-mcp · ★ 55k ▲363/7d · TypeScript

Chrome DevTools MCP lets a coding agent control and inspect a live Chrome browser through the Model Context Protocol or a standalone command-line tool. It gives the agent console messages, network requests, screenshots, performance traces, and page controls so it can investigate browser problems directly.

Verdict Our Chrome DevTools MCP run installed 1,130 packages and used 615 MB, then its tests failed after 131 seconds in Puppeteer's Chrome launch and connection path. Use it when an agent needs Chrome-specific console, network, trace, or memory evidence, and give it a dedicated browser profile.

agentic-awesome-skills

sickn33/agentic-awesome-skills · ★ 49k ▲349/7d · Python

Agentic Awesome Skills is a catalogue of reusable instruction packs for coding agents, plus local tools for finding and recording a chosen set. It helps Codex, Claude Code, Cursor, and other supported clients reuse task-specific playbooks while keeping the selected files open for review.

Verdict Our run installed 11 packages in 3 seconds, then the build failed on a missing Python `yaml` module and the tests hit the 900-second cap, so Agentic Awesome Skills is easier to consume than to develop from a bare Node container. Use a selected skill or focused plugin when you need breadth across coding agents and will inspect what the agent chooses.

claude-howto

luongnv89/claude-howto · ★ 46k ▲1.2k/7d · Python

Claude How To is an English-language course for learning Claude Code, with Vietnamese, Chinese, Ukrainian, and Japanese translations. It turns commands, memory, skills, subagents, MCP, hooks, plugins, checkpoints, and CLI use into a guided sequence with files you can copy into a project.

Verdict Our run installed 35 packages in 24 seconds, built in 8 seconds, and found no test target, so Claude How To is easy to inspect but did not prove its examples through an automated suite. Use it when you need a sensible course and editable starting files.

Fabric

danielmiessler/Fabric · ★ 45k ▲78/7d · Go

Fabric is a Go command-line tool and local server for running reusable AI instructions called Patterns. It keeps task prompts in editable Markdown, then lets you run them with local or hosted models instead of rebuilding the same workflow in each chat app.

Verdict Our Fabric run built in 266 seconds, but 3 of 71 tested packages failed in the extension template area, so adopt it around a few reviewed Patterns before making it shared infrastructure. It is a good fit when the task instruction is the asset you want to keep while models change.

kong

Kong/kong · ★ 44k ▲31/7d · Lua

Kong Gateway sits between clients and APIs, LLM providers, or MCP servers, then routes traffic and applies shared rules. It handles jobs such as load balancing, authentication, rate limits, health checks, logging, and provider-specific AI traffic through configuration and plugins.

Verdict Our Kong fixture built in 22 seconds and passed all 6 tests, but those results cover a small Rust workspace rather than the running gateway. Put Kong on the shortlist when APIs, several LLM providers, and MCP servers genuinely need one policy boundary.

gradio

gradio-app/gradio · ★ 44k ▲50/7d · Python

Gradio turns a Python function into a browser app with controls for text, images, audio, and other inputs. It lets people try a model or internal tool without learning your Python code, and can expose the same functions through client libraries or a Model Context Protocol server.

Verdict Our Gradio source build passed in 74 seconds, but its test command reported an error and stayed in watch mode until the 900-second timeout. I would choose Gradio for a Python model interface or internal tool, while requiring separate checks for the actual inputs and deployment settings.

playwright-mcp

microsoft/playwright-mcp · ★ 39k ▲243/7d · TypeScript

Playwright MCP lets an AI agent inspect and operate a web browser through Playwright, using page structure as its main interface. It solves the awkward job of giving an MCP client a persistent browser session without requiring a vision model for every click.

Verdict Our Playwright MCP run installed 112 packages and built in 7 seconds, but testing stopped because the process reported Node 18.20.8 against a Node 20 minimum. Use it for stateful, exploratory browser work where accessibility snapshots help an agent reason across several actions.

netron

lutzroeder/netron · ★ 34k ▲27/7d · JavaScript

Netron opens machine-learning model files as interactive graphs, so you can inspect layers, connections, shapes, and stored values without writing a loader for each framework. It runs in a browser, as a desktop app, or from Python, which makes it useful for checking an unfamiliar model before you wire it into an application.

Verdict Our Netron run installed 327 packages in 26 seconds, but its build and tests both failed, so using the viewer is easier than developing the repository on a generic Linux box. Use Netron when you need the quickest readable view of one of its main model formats, especially through the hosted app or a packaged release.

watermarks-remover

guillaumemeyer/watermarks-remover · ★ 26k ▲780/7d · Python

watermarks-remover inspects and removes provenance signals from text and files you own. It covers deterministic Unicode cleanup, file metadata, and optional attempts against statistical text, image, audio, and video watermarks, while stating that no universal clean result is possible.

Verdict Our watermarks-remover run installed 55 packages in 6 seconds and passed its 96-second test run, making the core a credible trial for authorized file hygiene. Use it when you can keep originals, inspect detailed detector status, and validate each result independently.

python-sdk

modelcontextprotocol/python-sdk · ★ 25k ▲49/7d · Python

The MCP Python SDK is the official Python library for building Model Context Protocol servers and clients. It turns typed Python functions into tools, resources, and prompts, then handles stdio, Streamable HTTP, or SSE communication with compatible AI applications.

Verdict Our MCP Python SDK run installed 61 packages and built in 38 seconds combined, but its tests never collected because `logfire` was absent. Use it when Python and official protocol coverage matter, especially for a server or client that must track the current MCP specification.

go-micro

micro/go-micro · ★ 23k · Go

Go Micro lets Go developers build ordinary network services, AI agents, and repeatable workflows in the same runtime. It can expose service methods as MCP tools, connect agents over A2A, and keep model-driven steps beside code paths whose order is fixed.

Verdict Our Go Micro run built in 102 seconds and passed all 162 tests in 97 seconds, so v6 is worth a serious trial for Go teams that want agents and services in one process model. Use the deterministic flow API for business steps and place approval checks around model-selected tools.

onnxruntime

microsoft/onnxruntime · ★ 22k ▲24/7d · C++

ONNX Runtime runs exported machine-learning models on CPUs, GPUs, browsers, mobile devices, and vendor accelerators. It gives applications one runtime API while execution providers handle the differences between deployment hardware.

Verdict Our ONNX Runtime run installed 66 packages in 16 seconds and built in 3 seconds, but it found no test target, so the result supports a quick package trial rather than a claim about the full native stack. Put it on the shortlist when one exported model must run across different hardware families.

typescript-sdk

modelcontextprotocol/typescript-sdk · ★ 14k ▲38/7d · TypeScript

The official MCP TypeScript SDK supplies the client, server, transport, schema, and authentication pieces needed to build Model Context Protocol software. Version 2 splits those jobs into focused packages that run on Node.js, Bun, and Deno.

Verdict Our TypeScript SDK checkout installed 877 packages in 25 seconds and occupied 665 MB, but its default build and test stages were both skipped because the root exposed neither target. Use the published v2 packages for new TypeScript MCP work, especially when you need both client and server behavior.

firecrawl-mcp-server

firecrawl/firecrawl-mcp-server · ★ 7.7k ▲44/7d · JavaScript

Firecrawl MCP Server gives MCP-compatible agents tools for web search, page scraping, site crawling, document parsing, research, and browser interaction. It handles the MCP adapter layer while Firecrawl's cloud API, or a separate self-hosted Firecrawl backend, does the web work.

Verdict Our Firecrawl MCP Server run installed 339 packages in 21 seconds and passed all 123 tests, so the adapter itself is easy to trial. Use it when Firecrawl is already an acceptable service boundary and you can restrict an agent's tools, paths, and credit use.

browser-tools-mcp

AgentDeskAI/browser-tools-mcp · ★ 7.3k ▲9/7d · TypeScript

BrowserTools MCP lets an AI coding client inspect the Chrome tab a developer is already using. A local MCP server and unpacked extension expose console output, network traffic, selected elements, screenshots, browser storage, and Lighthouse audits without moving the session into an automated browser profile.

Verdict Our browser-tools-mcp run built in 8 seconds and passed 355 of 356 tests, with the one failure in degraded attachment handling and 0 audit findings. It is a strong choice for debugging the authenticated Chrome tab in front of you, especially when console and network evidence matter more than browser automation.

genkit

genkit-ai/genkit · ★ 6.6k ▲10/7d · TypeScript

Genkit is an open-source framework for building server-side AI features in JavaScript and TypeScript, Go, Python, or Dart. It gives application developers one interface for model calls, structured output, tools, retrieval, workflows, local debugging, and production traces instead of making them assemble each part separately.

Verdict Our Genkit build succeeded, but it took 407 seconds after an 8-second install, so source contributors should budget for a slow loop even though dependency setup was quick. Use Genkit when an application team wants typed AI workflows, provider plugins, local traces, and a documented route into Firebase or Cloud Run.

apfel

Arthur-Ficial/apfel · ★ 6.5k ▲20/7d · Swift

apfel is a command-line tool for macOS that unlocks the free, on-device AI model built into Apple Silicon Macs. It lets you use this powerful, private AI directly in your terminal for scripting and automation, or run it as a local server that mimics the OpenAI API for your development projects.

Verdict apfel is a must-have utility for any developer or power user on a modern Mac. It brilliantly bridges the gap between Apple's powerful, private on-device AI and the command-line workflows where it can be most useful.

awesome-ai-agents

slavakurilyak/awesome-ai-agents · ★ 2.8k ▲88/7d · Go

Awesome AI Agents is a curated directory of public agent projects, frameworks, memory systems, evaluation tools, guardrails, and MCP software. It helps builders find candidates by category while showing repository ownership, submission provenance, maintainership evidence, stars, and recent star growth.

Verdict Our Awesome AI Agents run installed 3 packages in 2 seconds, built in 26 seconds, and passed 6 of 6 tests in 5 seconds, so the catalog tooling gave us a clean result. Use it for broad discovery when category, provenance, and fresh activity matter more than a ranked recommendation.

ECC

affaan-m/ECC · ★ 293k ▲4.6k/7d · JavaScript

ECC is a large kit of skills, specialist agents, rules, hooks, and memory tools for coding assistants, with Claude Code as its best-supported home. It gives an assistant a repeatable process for planning, testing, reviewing, and handing off work instead of making you rebuild that process in every prompt.

Verdict Our ECC run installed 264 packages in 17 seconds, but its tests never finished within 900 seconds. Use it if you are a heavy Claude Code user who will select a narrow profile, inspect the hooks, and treat the 4 audit findings as release work.

tensorflow

tensorflow/tensorflow · ★ 201k · C++

TensorFlow is an open-source machine-learning platform for building and running models through stable Python and C++ APIs. It gives teams one established project for model work across CPUs, GPUs, and device-specific runtimes, but contributing to the framework is far harder than installing its published package.

Verdict Our TensorFlow source install stopped after 6 seconds because the sandbox had no Bazel or Bazelisk, so framework contributors should expect a real toolchain rather than a pip-style setup. Use TensorFlow when an existing TensorFlow estate or its broad platform support solves a named compatibility problem.

graphify

Graphify-Labs/graphify · ★ 139k ▲4.2k/7d · Python

Try Graphify on one repository if you need to trace code relationships, but keep assistant integration optional until you reproduce the failed checks in our run. Graphify parses code locally with tree-sitter and builds a queryable knowledge graph. Our run on September 30, 2026 tested commit 1cd9a36; it recorded successful install and build commands, followed by a test command that exited 1. We did not test graph accuracy, an assistant workflow, or a shared MCP deployment.

Verdict Graphify is worth a limited trial when tracing relationships is the problem you need to solve. Our run supports a narrower conclusion about readiness: installation and the build command succeeded, while the test command failed.

codex

openai/codex · ★ 132k ▲1.1k/7d · Rust

Codex CLI is OpenAI's coding agent for working on a local repository from the terminal. It reads files, edits code, and runs commands under configurable approval and sandbox rules, using either a ChatGPT account or an OpenAI API key for model access.

Verdict Our Codex run installed one pnpm package in 27 seconds and used 170 MB, yet it never reached a build or test target, so the released CLI is the sensible path for users and the source tree is work for specialists. Use Codex when OpenAI model access, local repository tools, and explicit approval rules fit your development process.

generative-ai-for-beginners

microsoft/generative-ai-for-beginners · ★ 123k ▲363/7d · Jupyter Notebook

Generative AI for Beginners is an English-language Microsoft course, with automated translations into more than 50 languages. Its advertised 21 lessons pair explanations with Python and TypeScript examples for prompts, chat, search, images, RAG, agents, security, and model selection.

Verdict Our run installed 283 packages in 34 seconds with 0 known npm vulnerabilities, but it exposed no build or test target, so this is a reading-led course rather than a verified application. Use it for a broad first pass if you already know basic Python or TypeScript and can check each chosen lesson against current provider documentation.

TradingAgents

TauricResearch/TradingAgents · ★ 119k ▲4.1k/7d · Python

TradingAgents uses language models in analyst, researcher and risk roles to turn market data into written reports and a trading decision. It gives researchers a way to inspect how those roles reach a conclusion and compare decisions across tickers and dates.

Verdict TradingAgents installed in 50 seconds and passed 1,262 tests in our sandbox, but open issue #1518 questions whether its backtest credits decisions with price moves explained by after-close inputs. Use it to study agent reasoning and inspect written reports.

whisper

openai/whisper · ★ 112k ▲349/7d · Python

Whisper is OpenAI's speech-recognition model family and Python code for turning recordings into text, identifying the spoken language, and translating speech into English. It lets developers process audio locally through a command-line tool or Python, without sending recordings to a hosted speech API.

Verdict Our Whisper installation occupied 5,700 MB, and its tests failed after 202 seconds with exit 137 despite a successful build. Choose the reference implementation for local batch work when direct access to OpenAI's code matters and you can validate transcripts on your own recordings.

Agent-Reach

Panniantong/Agent-Reach · ★ 107k ▲4.3k/7d · Python

Agent Reach is a Chinese-first Python project, and the repository includes a detailed English guide. It gives command-running AI agents one place to select and check tools for websites, social networks, video platforms, GitHub, and feeds.

Verdict Our Agent Reach run installed 90 packages in 54 seconds and passed all 607 tests, so its Python core is a credible base for an agent that needs several internet sources. Use it when you accept that each social or video route still depends on an outside tool, account, and platform rule.

claude-mem

thedotmack/claude-mem · ★ 106k ▲3.1k/7d · TypeScript

Claude-Mem records what coding agents do, condenses that history, and brings relevant observations into later sessions. It began around Claude Code and now connects to several agent hosts, with local search through SQLite, Chroma, and MCP tools.

Verdict Our Claude-Mem run installed 800 packages and used 1,033 MB, then its test command failed in 1 second because `bun` was not found, so this is useful agent infrastructure with a real runtime footprint. Use it if lost coding context costs you more than operating and auditing a local memory worker.

awesome-mcp-servers

punkpeye/awesome-mcp-servers · ★ 97k ▲224/7d

This is not a single piece of software, but a massive, curated directory of open-source servers built on the Model Context Protocol (MCP). MCP is a standard that lets AI models and agents safely connect to external tools like databases, APIs, and local files. This 'awesome list' solves the critical discovery problem for developers, providing a central catalog for finding pre-built capabilities to give their AI agents.

Verdict Awesome MCP Servers is the definitive, indispensable catalog for the agentic AI revolution. While its massive popularity has led to clear maintenance challenges, evidenced by the thousands of open issues, its comprehensive scope and excellent organization make it the essential first stop for any developer looking to empower their AI with real-world capabilities.

VoiceStudio

debpalash/VoiceStudio · ★ 96k ▲8.9k/7d · Python

VoiceStudio is a local desktop workbench for cloning voices, creating speech, dubbing video, transcribing audio, dictating, and making audiobooks. It puts several speech engines behind one Electron app and API, so you can keep routine audio jobs on your own machine instead of assembling separate tools.

Verdict Our VoiceStudio run installed 1,166 packages and built in 20 seconds, but 1 of 924 tests failed, so this is ready for a serious trial rather than blind adoption. Use it when local dubbing, cloning, transcription, and agent access genuinely belong in one workspace.

Understand-Anything

Egonex-AI/Understand-Anything · ★ 89k ▲857/7d · TypeScript

Understand Anything turns a codebase, documentation set, or linked LLM wiki into a searchable knowledge graph and local browser dashboard. It combines Tree-sitter's structural analysis with LLM-written summaries so a developer can trace files, functions, dependencies, business flows, and likely change impact without opening every file first.

Verdict Our sandbox installed 585 packages and passed 789 of 793 Vitest cases, with 4 skipped, so Understand Anything is unusually easy to verify for a tool that asks an LLM to map a whole codebase. Use it for onboarding, guided exploration, and better questions about unfamiliar code.

rtk

rtk-ai/rtk · ★ 85k ▲545/7d · Rust

RTK is a local command-line tool that shortens terminal output before an AI coding agent reads it. It groups repetitive results and trims file or command details so shell work takes less space in the conversation.

Verdict Our RTK run built in 78 seconds but finished with 4 failed tests out of 8,256, so automatic command rewriting deserves a limited trial before a team-wide rollout. Use it when repetitive shell output crowds your coding sessions and you can compare filtered results with their originals.

deer-flow

bytedance/deer-flow · ★ 85k ▲324/7d · Python

Evaluate DeerFlow if you need a self-hosted workspace for agent research and coding and can investigate setup problems before deployment. DeerFlow orchestrates subagents, memory, and sandboxes through extensible skills. Our run on September 27, 2026, at commit 230d8f0 left setup unresolved: the recorded install outcome conflicts with packaging errors, and the test command returned a nonzero exit code.

Verdict Evaluate DeerFlow if its combined agent workspace fits your needs and your team can investigate setup problems. Our September 27, 2026 run at commit 230d8f0 does not justify a clean setup recommendation: the install summary conflicts with packaging errors, and pytest returned errors.

screenshot-to-code

abi/screenshot-to-code · ★ 81k ▲199/7d · Python

Screenshot to Code turns visual references, including screenshots and screen recordings, into frontend code using AI. Its current README lists HTML, React and Vue output options, with a React/Vite interface and a FastAPI backend for local operation. Our sandbox checked repository setup, not the fidelity or behavior of generated pages.

Verdict Screenshot to Code reached 42 Jest passes in our sandbox before the backend test command stopped at poetry: not found. It merits a trial for developers who can finish the Python setup and inspect generated interfaces, provided their reference material may go to the selected model provider.

hyperframes

heygen-com/hyperframes · ★ 76k ▲3.9k/7d · TypeScript

Choose HyperFrames for an HTML-based video workflow if your team can validate the rendered output before release. HyperFrames turns HTML, CSS, media, and seekable animations into MP4 video using headless Chrome and FFmpeg. Our run on September 30, 2026, at commit efd3c51 installed and built successfully, but the test command exited 1. That supports a controlled pilot, not a conclusion that we verified a finished video or cloud deployment.

Verdict Pilot HyperFrames if HTML is a useful video authoring format for your team. Our dated sandbox record supports installation and compilation, while the unsuccessful test command and absent export verification leave adoption work unfinished.

docling

docling-project/docling · ★ 70k ▲304/7d · Python

Docling turns PDFs, Office files, images, audio, video, email, and several specialist formats into a common document structure. You can export that structure as Markdown, HTML, plain text, WebVTT, DocLang, DocTags, chunks, or lossless JSON for search and AI pipelines.

Verdict Our Docling run installed 182 packages and used 7,131 MB, then finished with 79 failed tests and 5 errors, so adopting it means owning a purpose-built runtime image. Use it when mixed formats, local processing, and recoverable document structure justify that weight.

mem0

mem0ai/mem0 · ★ 68k ▲322/7d · Python

Mem0 is a memory layer for assistants and agents that stores useful facts from earlier conversations and retrieves them later. It gives developers a Python or TypeScript library, a self-hosted server, and a hosted service, so an app can remember a user without stuffing the full chat history into every prompt.

Verdict Our Mem0 run installed 76 packages in 37 seconds and built in 14 seconds, but the 987-test run ended with 25 failures and 93 collection or setup errors. Use it when you need a well-documented memory layer across Python, TypeScript, self-hosting, and cloud, and you can gate your exact provider and vector-store combination with integration tests.

last30days-skill

mvanhorn/last30days-skill · ★ 66k ▲559/7d · Python

Last30Days is an agent skill that searches recent discussion, video, code activity, prediction markets, papers, and web pages, then asks your agent to turn the findings into a sourced brief. It cuts down the tab-by-tab work of checking what people have said about a topic this month.

Verdict Our Last30Days run installed 35 packages and built successfully, but its test step failed after 829 seconds on UID lookup errors in keychain tests. Try it when recent community evidence is central to your work and you already use a supported agent host.

keras

keras-team/keras · ★ 64k ▲11/7d · Python

Keras is a Python library for building and training neural networks through a shared model interface. It supports JAX, TensorFlow and PyTorch for training, with OpenVINO available for inference, so you can choose an execution engine without starting every model definition again.

Verdict Our Keras run installed 197 dependencies in 102 seconds, occupied 4129 MB, and hit the 900-second test limit. Use it when keeping model code across training backends solves a real maintenance problem, provided you can check your custom components on each intended engine.

voicebox

jamiepine/voicebox · ★ 59k ▲544/7d · Python

Voicebox is a local voice studio for speech generation, voice cloning, transcription, and dictation. Its README describes a multitrack Stories editor, REST API, and MCP server for using voice profiles in applications and agent conversations. These are documented capabilities; our sandbox did not assess their audio output.

Verdict Our Voicebox dependency install succeeded in 18 seconds, then the build stopped after 4 seconds because the sidecar script could not find python. Speech quality remains unassessed.

marketingskills

coreyhaines31/marketingskills · ★ 58k ▲996/7d · JavaScript

Marketing Skills is a comprehensive library of instructions that turns general-purpose AI coding assistants into specialized marketing experts. It provides structured playbooks and frameworks as simple text files, allowing AI agents to perform complex tasks like SEO audits, conversion rate optimization, and ad campaign planning with domain-specific knowledge.

Verdict For any technical marketer, founder, or growth engineer already using an AI coding agent, this project is an absolute must-have. It's a massive force multiplier, transforming your generalist AI assistant into a seasoned marketing strategist on demand.

CLIProxyAPI

router-for-me/CLIProxyAPI · ★ 57k ▲839/7d · Go

CLIProxyAPI is a Go gateway that lets OpenAI, Gemini, Claude, Codex, and Grok clients talk to several model providers through compatible APIs. It can pool OAuth accounts and API credentials, translate request formats, and route retries so one client does not need a separate integration for every provider.

Verdict Our CLIProxyAPI run passed all 158 Go tests after a 57-second install and 144-second build, making it a credible gateway for operators who need several coding-model protocols behind one endpoint. Use it if account pooling and format translation remove more work than credential care adds.

goose

block/goose · ★ 56k ▲277/7d · Rust

Goose is a local AI agent available as a desktop app, command-line tool, and embeddable API. It can edit code, run shell tools, automate other work, and connect to outside capabilities through MCP extensions while letting you choose the model provider.

Verdict Our Goose run installed 1,389 packages in 66 seconds, but both build and tests reached their 900-second caps while Rust was still compiling, so source contributors should expect a slow first feedback loop. The packaged app is the better route for people who want a provider-flexible local agent with MCP tools.

scientific-agent-skills

K-Dense-AI/scientific-agent-skills · ★ 55k ▲824/7d · Python

Scientific Agent Skills is a library of instructions, references, and helper scripts that teaches coding agents how to use scientific packages, databases, and research workflows. Its 177 current skills work with Claude Code, Codex, Cursor, and other Agent Skills hosts across biology, chemistry, medicine, engineering, data analysis, and scientific writing.

Verdict Our scientific-agent-skills run installed 120 packages and built in 10 seconds, but the test command exited 4 after 8 seconds because 106 skill suites cannot share one pytest process. Use the collection as a shelf: choose a small pinned subset, run each skill's isolated checks, and verify its scientific output against known data.

whisper.cpp

ggml-org/whisper.cpp · ★ 55k ▲138/7d · C++

whisper.cpp is a C/C++ implementation of OpenAI's Whisper speech recognition system that can turn audio into text locally. It is built for people who need offline transcription across desktops, phones, browsers, and small computers without carrying a large software runtime.

Verdict Our build succeeded in 20 seconds, but all 7 Go test targets failed to build in the supplied sandbox run. whisper.cpp is still a strong default for teams that specifically need local Whisper inference inside native, mobile, browser, or heterogeneous hardware deployments.

academic-research-skills

Imbad0202/academic-research-skills · ★ 55k ▲819/7d · Python

Academic Research Skills is an English-documented set of Claude Code skills for literature research, paper drafting, citation checks, revision, and simulated peer review. It turns those jobs into staged workflows with human approval points, but it does not validate that experiments happened or that underlying data are genuine.

Verdict Our npm install finished in 11 seconds with 0 packages, but there was no build or test target, so the clean audit says little about the 10-stage research workflow. Use Academic Research Skills if you already work in Claude Code, accept the noncommercial license, and want checkpoints that slow down paper drafting.

DeepTutor

HKUDS/DeepTutor · ★ 52k ▲9.1k/7d · Python

DeepTutor is a self-hosted learning workspace that joins AI chat, document retrieval, guided study, quizzes, writing, books, and persistent memory. It solves the fragmentation of keeping study material, progress, and model conversations in separate tools, while letting you choose local or hosted model providers.

Verdict Our DeepTutor run installed 218 packages and occupied 728 MB, then reached only 9% of the test suite before a 900-second timeout, so it fits technical users who want the whole learning workspace and can own its operations. Use it when shared context across reading, practice, writing, and agents is the point.

nature-skills

Yuan1z0825/nature-skills · ★ 51k ▲924/7d · Python

Nature Skills is a Chinese-first collection of agent instructions and helper code for scientific writing, literature work, figures, presentations, and research administration. A full English README and English pages for individual skills are available, so English-speaking researchers can evaluate and install it without translating the main documentation themselves.

Verdict Our Nature Skills run installed 49 packages in 23 seconds and passed all 10 tests, which makes the tested downloader component easy to trial. Use the repository when you want agent instructions that cover the paperwork around research as well as papers, especially if your lab works in Chinese and English.

supervision

roboflow/supervision · ★ 51k ▲74/7d · Python

Choose Roboflow Supervision when you already have a computer-vision model and need Python utilities around its predictions. Supervision provides model connectors, annotators and dataset utilities. Our run on October 10, 2026, at commit 0802a0e installed and built successfully, but its test command failed. That result supports trying the library with your own inputs; it does not establish deployment readiness or correct evaluation results.

Verdict Use Supervision when model connectors, annotations and dataset operations solve recurring work in your Python application. Our October 10, 2026 run at commit 0802a0e completed installation and the build step, but the test command failed with dependency-related failures in the raw tail.

airi

moeru-ai/airi · ★ 51k ▲225/7d · TypeScript

Project AIRI connects animated Live2D and VRM characters to language models and speech services so you can talk with a digital companion. Its English README describes browser, desktop and mobile routes, with local and hosted providers you configure separately.

Verdict Our AIRI source build failed with exit 129 after 68 seconds, so that checkout is an unverified starting point for source development. AIRI deserves a trial when the animated character is essential and you can configure the model and speech routes.

AstrBot

AstrBotDevs/AstrBot · ★ 45k ▲1.6k/7d · Python

AstrBot is a Python platform for putting one AI assistant into chat systems such as Telegram, Slack, Discord, QQ, Feishu, and DingTalk. It combines model providers, tools, knowledge bases, personas, MCP, and plugins behind a WebUI, with an English README and several translated editions.

Verdict Our AstrBot run installed 166 packages in 104 seconds and built in 9 seconds, but its test command stopped on a missing `pytest_asyncio` import before collection. Use it when official access to several chat networks, especially QQ, WeCom, Feishu, or DingTalk, is worth operating a broad Python service.

Codewhale

Hmbown/CodeWhale · ★ 41k ▲33/7d · Rust

Codewhale is a coding agent that reads a project, edits files, runs commands, and checks its work from a terminal or local browser. It can use hosted or local models, while saved sessions, goals, agent roles, skills, plugins, and MCP servers support longer jobs.

Verdict Our Codewhale run installed 47 packages in 28 seconds and node:test passed 67 of 67 cases, but the overall test step still exited 127. Use it when persistent sessions, model choice, MCP, and coordinated agents solve problems you already have, then reproduce the v0.10.0 MCP issue before rollout.

awesome-cursorrules

PatrickJS/awesome-cursorrules · ★ 41k ▲50/7d · JavaScript

Awesome Cursor Rules is a CC0 library of instruction files that tell Cursor how to work on particular stacks and tasks. You browse the catalog, copy a chosen `.mdc` file into `.cursor/rules/`, and edit it for your codebase; there is no app or service to run.

Verdict Our run installed 0 packages and passed all 84 tests in 8 seconds, so Awesome Cursor Rules is cheap to inspect but those checks do not prove that any copied framework advice fits your codebase. Use it as a source of one carefully chosen rule, then rewrite that file around your real commands and constraints.

oh-my-claudecode

Yeachan-Heo/oh-my-claudecode · ★ 40k ▲172/7d · TypeScript

Consider oh-my-claudecode if you already use Claude Code and need coordinated planning, execution and verification; start with a bounded trial before allowing unattended work. OMC is a Claude Code orchestration plugin with a separate npm CLI/runtime. Our run on September 28, 2026 tested commit 9fd35ec, associated with v5.5.0, in our sandbox. It checked installation, build and the test command, not an authenticated agent workflow or deployment.

Verdict OMC is worth a controlled trial when Claude Code is already your working environment and coordination is a recurring problem. Our September 28, 2026 run at commit 9fd35ec supports successful installation and build, but the test command timed out and the retained setup suite reports four failures.

ai-website-cloner-template

JCodesMore/ai-website-cloner-template · ★ 38k ▲704/7d · TypeScript

AI Website Cloner Template is a blank Next.js application paired with an agent skill for rebuilding the visible front end of a live website. The skill makes a coding agent inspect styles, behavior, responsive states, content, and assets before it writes isolated components and compares the result with screenshots.

Verdict Our run pulled 889 packages and 616 MB for a 46-file template, then built it in 25 seconds with 0 known audit findings, so this is a credible agent playbook rather than a push-button cloner. Use it for an authorized front-end rebuild when a developer will supervise the browser sweep and inspect the result.

pytorch-image-models

huggingface/pytorch-image-models · ★ 37k ▲17/7d · Python

PyTorch Image Models, usually called timm, is a Python collection of image models, training utilities and reference scripts. Its model factory lets PyTorch developers select pretrained models, replace classifiers and access features through documented methods.

Verdict Consider timm when comparing PyTorch image encoders will save integration work and your team can validate the selected checkpoints, transforms and outputs. Our commit c709c9a run installed in 52 seconds and used 5,439 MB, but tests exited 137 after 443 seconds without completing.

agentscope

agentscope-ai/agentscope · ★ 35k ▲963/7d · Python

AgentScope is a Python framework and FastAPI service for building agents that call models, use tools, remember context, and work in teams. It solves the engineering around an agent loop, including permissions, storage, scheduling, sandboxes, MCP connections, and a web interface.

Verdict Our AgentScope run installed 305 packages in 114 seconds, used 1,054 MB, and reported 2,531 passed with 0 failed, so the broad foundation held up in a clean container. Use it when you need an agent platform with permissions, storage, teams, and deployment pieces, not merely a ReAct loop.

x-algorithm

xai-org/x-algorithm · ★ 34k ▲168/7d · Rust

X Algorithm publishes core code for selecting, ranking and filtering posts in X's For You feed. Engineers can inspect recommendation and visibility decisions or assess individual components for adaptation. The published code does not establish a complete deployment package.

Verdict The bdsm/ component installed in 22 seconds and completed its test command successfully at commit 35650fb. That supports a component-level setup finding, not reproduction of the For You feed.

tinygrad

tinygrad/tinygrad · ★ 34k ▲17/7d · Python

tinygrad is a Python deep learning library that includes the compiler turning tensor calculations into device code. It lets developers write neural networks and inspect or change how their calculations run.

Verdict Our tinygrad build succeeded in 11 seconds, while the test command failed with 200 failures and 2 passes. The shown AMD emulator errors could not find clang, so our run leaves device execution unresolved.

context-mode

mksglu/context-mode · ★ 33k ▲694/7d · TypeScript

Context Mode is an MCP server and coding-agent plugin that processes bulky tool output before selected results enter an AI conversation. It uses SQLite and FTS5 for stored context and, on supported clients, captures session events for recovery after conversation compaction.

Verdict The Context Mode suite finished with 4,702 passed, 0 failed, and 40 skipped at commit e5fcca6. That supports trying the client integration, while leaving routing, retrieval completeness, and session recovery unverified.

gbrain

garrytan/gbrain · ★ 31k ▲149/7d · TypeScript

GBrain is a self-hosted memory layer that lets several AI agents recall the same sourced facts across sessions. It stores knowledge in Markdown and a database, then adds keyword search, optional semantic retrieval, a graph, corrections, withdrawal, and cited synthesis.

Verdict Our GBrain run installed 295 packages and built in 10 seconds, but its tests were still emitting 4-shard heartbeats when our 900-second limit ended. Use it when sourced, portable agent memory matters enough to justify explicit privacy and backup design.

composio

ComposioHQ/composio · ★ 31k ▲89/7d · TypeScript

Composio is a hosted action layer for AI agents, with TypeScript and Python SDKs that let an agent find tools, connect a user's account, and call outside services. It saves a product team from maintaining authentication and API wrappers for every app its agent uses.

Verdict Our Composio run installed 830 packages and used 1,931 MB, built successfully, then failed its test command because Bun was absent. Use it when managed authentication and a large connector catalog remove more work than a hosted control plane adds.

picoclaw

sipeed/picoclaw · ★ 30k · Go

PicoClaw is an open-source personal AI assistant initiated by Sipeed and written in Go. Its README documents terminal, browser and messaging interfaces, file operations, code execution, scheduled tasks and MCP connections to external tools. It needs a configured model provider or local model server.

Verdict PicoClaw is a personal-lab candidate for developers who want a Go assistant and can operate its model and tool connections. The harness reported build success in 134 seconds, but clean compilation remains unverified because the supplied accounts of the same run conflict.

mastra

mastra-ai/mastra · ★ 29k ▲147/7d · TypeScript

Mastra is a TypeScript framework for building agents, tool-using assistants, and graph workflows. It puts model routing, memory, retrieval, human approval, evaluation, observability, and Model Context Protocol support in one system for JavaScript teams.

Verdict Our Mastra build took 693 seconds and its test command was still running at the 900-second cap, so full-repository work carries a real CI cost. Use it when a TypeScript team wants agents, explicit workflows, storage, Studio, and MCP under one roof.

repomix

yamadashy/repomix · ★ 29k ▲70/7d · TypeScript

Repomix combines selected repository files into a structured document for an AI assistant. Its README documents include and ignore rules, token counting, compression, MCP integration and Claude Code plugins.

Verdict Repomix is worth a trial when repository context needs repeatable file selection. At commit 8d64291, our October 6, 2026 run passed all 1,843 tests in 27 seconds, while npm audit reported 8 known vulnerabilities, including 1 critical.

mlx

ml-explore/mlx · ★ 29k ▲41/7d · C++

MLX is an array framework for machine learning, designed by Apple researchers around Apple silicon. It gives Python developers familiar array and neural-network interfaces, with lazy computation and shared CPU/GPU memory. The current README also documents Linux CPU and CUDA packages.

Verdict Our MLX run installed in 28 seconds and built in 4 seconds, but its test command stopped with 34 collection/setup errors, leaving model correctness unverified. MLX is worth a focused evaluation for researchers whose target hardware and model code suit its array framework.

qwen-code

QwenLM/qwen-code · ★ 29k ▲125/7d · TypeScript

Qwen Code is an open-source coding agent for the terminal, with documented editor, desktop, browser and chat interfaces. Its README lists OpenAI, Anthropic, Gemini and Qwen APIs, plus local endpoints. These are documented capabilities; our recorded source-checkout run did not verify model compatibility or coding quality.

Verdict Qwen Code merits a trial when provider choice matters. The recorded source setup and build completed, while the test command reached its time limit without a completed suite result.

planning-with-files

OthmanAdi/planning-with-files · ★ 28k ▲134/7d · Python

Planning with Files keeps an AI coding task's plan, findings and progress in three Markdown files on disk. It is a Claude Code plugin and a skill for other coding agents. Its documented recovery behavior depends on the host and installation route.

Verdict Planning with Files is worth evaluating when context resets repeatedly leave your agent without a usable task record. Our repository checks did not establish host recovery, and the test command ended with a yaml import error.

onlook

onlook-dev/onlook · ★ 27k ▲46/7d · TypeScript

Onlook is an open-source visual editor for Next.js and Tailwind applications that connects changes in a page preview to source code. Its AI chat can also edit the project, while the newer hosted product is separately described as early access.

Verdict Our Onlook build failed in 14 seconds on missing Supabase configuration after installation succeeded in 104 seconds. The repository is worth a focused trial for Next.js and Tailwind teams that can own its service setup.

shap

shap/shap · ★ 26k ▲8/7d · Jupyter Notebook

Choose SHAP when you need feature attributions for an existing machine learning model and can validate the explainer against your own model and data. SHAP explains model output using Shapley values and related methods. Our run on October 11, 2026, at commit 6514e36 installed and built successfully, but its test command stopped while loading conftest because matplotlib was missing. We did not test explanation quality, model compatibility, runtime speed or GPU execution.

Verdict SHAP is a reasonable choice for feature attribution across several model families when you can validate the relevant explainer yourself. Our October 11, 2026 run at commit 6514e36 supports a narrower conclusion: installation and build succeeded, while the test command stopped on a missing matplotlib import.

flash-attention

Dao-AILab/flash-attention · ★ 25k ▲11/7d · Python

FlashAttention is the official implementation of an exact attention algorithm that reduces data movement between GPU memory and faster on-chip memory. It supplies specialized training and inference kernels for supported NVIDIA and AMD GPUs, so model builders can replace a costly attention operation without changing its mathematical result.

Verdict Our FlashAttention checkout installed and built in 30 seconds, but pytest then stopped with 122 collection or setup errors and no executed test result. Use it when a supported GPU, an exact attention workload, and measured model-level gains justify owning a compiled extension.

GenAI_Agents

NirDiamant/GenAI_Agents · ★ 25k ▲56/7d · Jupyter Notebook

GenAI Agents is a collection of Jupyter notebooks that teaches agent design through working examples, from a plain model-and-tools loop to LangGraph workflows and multi-agent systems. It helps you compare patterns before committing to a framework. You do not get one agent application to deploy.

Verdict Our GenAI Agents run installed 35 packages, built in 7 seconds, and returned 38 passing tests with 0 failures, making it a credible learning repo for Python developers. Use it to study one agent pattern and compare implementations.

json-render

vercel-labs/json-render · ★ 19k ▲68/7d · TypeScript

json-render is a TypeScript framework for turning JSON descriptions into working interfaces. An AI model can choose only the components, properties, and actions that your application places in a catalog, which is safer and easier to validate than letting the model write arbitrary interface code.

Verdict Our json-render run passed all 1,212 tests in 67 seconds, but its 2,502-package install used 2,986 MB and the full build ended with exit 137 after 42 of 53 tasks. Adopt an individual renderer package when you need AI to arrange a controlled set of real application components.

rowboat

rowboatlabs/rowboat · ★ 18k ▲13/7d · TypeScript

Rowboat is a desktop work assistant that keeps each person's email, meetings, notes, code context, and model credentials on their own machine. Its Spaces feature gives a team shared chat, files, whiteboards, and an MCP server through Harbor, while each teammate's local Rowboat answers with that person's private context.

Verdict Our Harbor run installed 195 packages and passed 381 of 382 Vitest checks, with 1 skipped and none failed. Rowboat is a serious choice for a team that wants each member's work memory kept local while agents collaborate through an explicit shared Space.

docker-android

budtmo/docker-android · ★ 16k ▲38/7d · Python

Docker-Android runs Android emulators in Docker containers so you can see their screens in a browser and control them remotely. It gives developers a reusable Android workspace for app debugging and automation, with separate MCP and sponsor-only AI agent options.

Verdict Our Docker-Android CLI setup installed 43 dependencies in 16 seconds, but the skipped test step leaves emulator operation unverified. Use it for a browser-visible Android workspace when you already have a KVM-capable host and can verify your chosen image.

Paradev

ParadevAI/Paradev · ★ 16k · TypeScript

Paradev is a desktop workspace for CLI coding agents, terminals, code review and browser previews. It gives tasks separate Git worktrees. Its source is available under Elastic License 2.0, which restricts providing substantial functionality as a hosted or managed service to third parties.

Verdict Paradev is worth evaluating on macOS if you already use CLI coding agents and want their patches and previews in one workspace. The recorded source build and tests failed, so the evidence supports an evaluation recommendation rather than a verified setup endorsement.

speech-to-speech

huggingface/speech-to-speech · ★ 14k ▲46/7d · Python

Speech to Speech is a Python voice-agent pipeline that listens, turns speech into text, gets a language-model response, and speaks the answer. Each stage can be swapped, while applications connect through a documented subset of the OpenAI Realtime protocol over WebSocket or WebRTC.

Verdict Our speech-to-speech run produced 43 collection/setup errors and only 6 passing tests, even though installation and build both succeeded, so version 1.0.0 did not clear a basic source-checkout gate in our sandbox. It is still worth evaluating for experienced voice teams that need swappable local components and a Realtime-compatible client surface.

FluidVoice

altic-dev/FluidVoice · ★ 13k ▲165/7d · Swift

FluidVoice is a macOS dictation app that turns speech into text locally and inserts it into almost any app. It solves the usual privacy and subscription concerns of cloud dictation, while adding optional voice commands and AI cleanup.

Verdict FluidVoice is one of the strongest free Mac dictation trials because installation is simple, model choice is broad, and the core speech path can remain local. The private Fluid Intelligence runtime weakens the open-source pitch, and recent microphone and insertion bugs make it risky for work where a missed sentence is costly.

codex-security

openai/codex-security · ★ 11k ▲64/7d · TypeScript

Codex Security is a command-line scanner and TypeScript SDK that uses coding agents to find, validate, and help fix vulnerabilities in source repositories. It can turn a repository, path, Git diff, or imported alert backlog into findings with evidence, coverage records, patches, and machine-readable exports.

Verdict Our repository run added 0 packages and had no build or test target, so it proves almost nothing about the scanner's security judgment. Codex Security is worth a controlled trial when your AppSec team wants agent-led investigation tied to evidence, coverage, and reviewed patches.

tokenizers

huggingface/tokenizers · ★ 11k ▲15/7d · Rust

Hugging Face Tokenizers converts text into the integer IDs a language model consumes, and converts those IDs back into text. Its Rust core has official Python and Node.js bindings, while the current v1 release candidate splits inference, serialization, and conversion into smaller pieces and is still restoring training.

Verdict Our Tokenizers build succeeded after a 41-second, 102-package install, but the 56-second cargo test step discovered 0 tests. Use the established v0.23.2 line when you need the broad Python authoring surface today, and evaluate v1.0.0-rc.2 separately for its inference-focused design.

RapidRAW

CyberTimon/RapidRAW · ★ 11k ▲143/7d · TypeScript

RapidRAW is a desktop photo editor for developing RAW files without changing the originals. It combines a photo library, GPU-based adjustments, masks, batch export, and optional local AI tools in one app for Windows, macOS, Linux, and Android.

Verdict Our RapidRAW run installed 283 packages in 9 seconds and built in 22 seconds, but it offered no automated test target, so the app is easy to try and harder to trust without your own photo checks. Use it if you want a modern, free RAW workflow and can validate exports from your cameras before moving a library.

drawio-skill

Agents365-ai/drawio-skill · ★ 10k ▲115/7d · Python

Drawio Skill is an agent skill that turns descriptions, code, and infrastructure definitions into editable draw.io diagrams. It helps a coding assistant maintain the diagram after the first draft, including source synchronization and different views of the same architecture model.

Verdict Our Drawio Skill run installed 35 dependencies in 4 seconds and completed its build and test steps, making the Python tooling inexpensive to try while native rendering remains unverified. Use it when editable diagrams and source synchronization will save work on later revisions.

feynman

advaitpaliwal/feynman · ★ 10k ▲26/7d · TypeScript

Feynman is a terminal AI research agent for finding papers, comparing sources and drafting research briefs. Its literature-review and deep-research workflows save provenance files so you can inspect the sources behind the output.

Verdict Our Feynman sandbox build completed in 1 second, but its test command failed after 127 seconds on a path involving archive reproducibility. Its saved provenance makes it worth a trial for researchers who will verify citations, while the documented small-local-model limitations deserve attention.

rf-detr

roboflow/rf-detr · ★ 9.9k ▲48/7d · Python

RF-DETR is a Python model family for finding objects, drawing instance masks, and, in preview form, locating human keypoints in images. It gives computer-vision teams one API for using pretrained models, fine-tuning them on COCO-format data, and exporting the result to several deployment runtimes.

Verdict Our RF-DETR install took 160 seconds and 5,778 MB, then its tests stopped on an unrecognized pytest option, so adopting the source tree requires more care than the short inference example suggests. It deserves a trial if detection and segmentation under one API fit your work and you can compare models on your own GPU.

monty

pydantic/monty · ★ 8.8k ▲31/7d · Rust

Monty is a small Python 3.14 interpreter written in Rust for running code generated by AI. It gives that code no filesystem, environment, network, subprocess, or native-extension access unless the host application supplies a narrow callback or mount.

Verdict Our Monty install took 28 seconds and 37 MB, but the build exited 1 after 6 seconds and the log tail did not reveal why, so source builders should reproduce that step before adoption. The packaged interpreter is a strong fit for agents that need a small, deliberately limited Python 3.14 language sandbox and only a few audited host calls.

vibe

thewh1teagle/vibe · ★ 8.0k ▲63/7d · TypeScript

Vibe is a desktop app that turns audio and video into transcripts or subtitle files on your computer. It supports local speech models and batch processing, with optional transcript summaries through the Claude API or local analysis through Ollama.

Verdict Our Vibe run installed 3 recorded dependencies in 4 seconds, but skipped build and tests, leaving desktop transcription unverified. Try a packaged release for local interview or caption work if you can check a representative recording through export.

koharu

koharu-rs/koharu · ★ 6.5k ▲48/7d · Rust

Koharu is a desktop manga translation app. Its documentation describes text detection, OCR, translation, cleanup, proofreading, and an editable page canvas, with PNG and layered PSD export. Our sandbox run did not reach a working application.

Verdict Koharu's documented correction and layered-export workflow warrants a trial for manga translators who expect to edit machine output. At commit 2c78c0c, our dependency-install step completed, the build failed on missing GTK4 metadata, and Vitest reported 102 passing tests.

koharu

mayocream/koharu · ★ 6.5k ▲48/7d · Rust

Koharu is a desktop manga translation editor with text detection, OCR, translation, cleanup and page composition. Its English README documents proofreading, a WebGPU canvas and layered PSD export. The requested mayocream/koharu repository resolves to koharu-rs/koharu.

Verdict At commit 988a6cd, our source build failed when pkg-config could not find GTK4; Vitest passed 102 cases. Try the official release if you want to evaluate Koharu's documented correction and lettering workflow.

open-duet

Anil-matcha/open-duet · ★ 6.4k ▲198/7d · Python

Open Duet is a workspace for named AI agents with their own instructions, conversations, memory and assigned folders. It lets one operator organize delegated work and recurring prompts while bringing a separately configured model engine.

Verdict Our Open Duet backend run installed 34 dependencies in 12 seconds and used 38 MB, then built successfully in 1 second. Tests stopped on a missing pytest_asyncio import, so that small environment gives us no evidence of working agent tasks.

zcf

UfoMiao/zcf · ★ 6.1k · TypeScript

ZCF is an interactive setup and configuration manager for Claude Code and Codex. It installs the command-line tools, writes their global settings, adds workflows and MCP services, and lets you switch among official logins, custom endpoints, and provider presets.

Verdict Our ZCF run installed 899 packages, built successfully, and passed all 2,552 tests, so its codebase earns more trust than the phrase "one-click setup" usually does. Use it if you actively manage both Claude Code and Codex, especially with several provider profiles.

ccg-workflow

fengshao1227/ccg-workflow · ★ 6.0k ▲9/7d · TypeScript

CCG Workflow is a Claude Code extension that turns one coding request into a persistent, approval-gated workflow and can route parts of the job to other model CLIs. Its documentation is available in English and Simplified Chinese, and it is aimed at developers who want Claude to coordinate Codex, Gemini, and other supported backends without hand-managing each call.

Verdict Our CCG Workflow run installed 617 packages and passed all 206 tests in 43 seconds across install, build, and test, making the codebase easy to trial for a tool with this much reach. Use it when you already pay attention to several coding models and want Claude Code to hold one approved plan and a review trail.

see-through

shitagaki-lab/see-through · ★ 5.2k · Python

See-through is a research pipeline that separates a single anime character illustration into editable layers and exports a layered PSD. It uses layer generation and depth estimation to help recover hidden parts and infer drawing order. The output is a starting point for an artist's work, with rigging explicitly outside the project's scope.

Verdict Our See-through environment occupied 6,395 MB and its audit reported 96 known vulnerabilities, while image decomposition and PSD output remained unverified. It merits a contained trial for technical artists who want help preparing layers and can inspect the result before rigging.

json_repair

mangiucugna/json_repair · ★ 5.1k ▲5/7d · Python

json_repair is a Python parser that turns malformed JSON from language models, APIs, logs, or people into a valid JSON string or Python object. It fixes common syntax damage such as missing quotes, commas, and closing brackets, and it can use a JSON Schema or Pydantic model to guide a repair. Treat the result as recovered input that still needs validation, because valid syntax does not prove that a guessed value is correct.

Verdict Our run passed 238 of 244 executed tests after a 29-second install, while all 6 failures were schema-aware cases that said `jsonschema` was required. json_repair is an easy recommendation for Python teams that need a local fallback for malformed model or API output, provided every repaired object still goes through application validation.

agent-sprite-forge

0x0funky/agent-sprite-forge · ★ 4.7k · Python

Agent Sprite Forge is a set of skills for Claude Code, Codex and Grok that turns an approved character image into animated sprite assets. It coordinates image and video generation, then uses Python tools to remove backgrounds, align frames and package actions for game projects. The documentation requires visual review and says editor imports are unverified.

Verdict Agent Sprite Forge installed in 7 seconds and built in 3 seconds in our sandbox, but its tests timed out after 900 seconds without a final result. Evaluate it for a supervised sprite experiment if you can inspect each action and verify the export in your game editor.

desktop-cc-gui

zhukunpenglinyutong/desktop-cc-gui · ★ 4.5k ▲27/7d · TypeScript

Desktop CC GUI is a bilingual, English-and-Chinese desktop client for AI coding tools. It puts Claude Code, Codex CLI, OpenCode, DeepSeek Harness, and 6 other engines in one Tauri interface, with chat, files, Git, a terminal, sessions, and model controls.

Verdict Our desktop-cc-gui run installed 581 packages, built in 55 seconds, and passed all 1,231 tests in 110 seconds. That is the cleanest lab result in this batch and earns a trial for developers who regularly switch between Claude Code, Codex, and other CLIs.

Hyper-Extract

yifanfeng97/Hyper-Extract · ★ 4.2k ▲8/7d · Python

Hyper-Extract turns documents into searchable, structured knowledge that can be updated when the source changes. It can produce lists, graphs, hypergraphs, and time- or location-aware records, then trace or remove facts contributed by a particular document.

Verdict Our Hyper-Extract run installed 164 packages in 44 seconds and completed its test command in 36 seconds with 715 passed, 0 failed, and 15 skipped. That makes it a credible trial for a Python team that needs document updates, provenance, and several knowledge shapes behind one interface.

SoL-Pi

NVlabs/SoL-Pi · ★ 3.6k ▲40/7d · TypeScript

SoL-Pi is an extension for the Pi coding agent that tries to cut repeated tool calls, replayed output, long diagnostic logs, and stale task context. It adds four opt-in mechanisms without patching Pi itself.

Verdict Our SoL-Pi run passed 195 of 195 tests in 25 seconds after a 179-package install, making it unusually easy to inspect for a new agent extension. Use it as a measured experiment on Pi 0.85.1, starting with Action Fusion or ObservationPack before enabling model-backed reduction or automatic compaction.

agent-of-empires

agent-of-empires/agent-of-empires · ★ 3.4k ▲13/7d · Rust

Agent of Empires is a session manager for running several coding agents from one terminal or web dashboard. It wraps installed tools such as Claude Code, Codex, OpenCode, Gemini CLI, and Oh My Pi in persistent tmux sessions, with optional worktrees and containers to keep parallel jobs apart.

Verdict Our Agent of Empires build succeeded, and 4,656 of 4,660 tests passed before four failures, including one tied directly to missing tmux, so the codebase is credible but the stated prerequisite is real. Use AoE when supervising multiple Claude Code, Codex, or OpenCode jobs has become harder than the jobs themselves.

mjlab

mujocolab/mjlab · ★ 3.3k ▲34/7d · Python

mjlab is a Python framework for training robot behaviors across many MuJoCo simulations on NVIDIA GPUs. It brings Isaac Lab's manager-style environment building to MuJoCo Warp, so researchers can assemble scenes, observations, rewards, and commands without running Isaac Sim.

Verdict Our mjlab sandbox installed 149 packages using 5,765 MB, then its 135-second test step reported 1,067 passes and 0 failures. Use mjlab when your team has NVIDIA hardware, knows Isaac Lab's manager pattern, and wants MuJoCo Warp without the Isaac Sim platform.

ainovel-cli

voocel/ainovel-cli · ★ 2.3k ▲23/7d · Go

The supplied documentation is primarily Chinese, and the provided material does not establish whether a full English guide exists. ainovel-cli is a Go command-line application that turns a short story brief into a long novel by coordinating planning, drafting, editing, and continuity tracking without requiring the author to drive every chapter.

Verdict Our build finished in 31 seconds, and all 52 reported tests passed, which makes ainovel-cli unusually credible for a fast-moving autonomous writing project. Use it if you want a structured, restartable novel factory and are comfortable reading Chinese documentation and paying for model calls.

webbrain

webbrain-one/webbrain · ★ 1.6k ▲44/7d · JavaScript

WebBrain is an open-source browser extension that lets an AI model read pages, answer questions, and carry out browser tasks in the tabs you already use. It can use its managed model, a cloud API, or a local OpenAI-compatible server, and it also lets coding agents delegate browser work through MCP.

Verdict Our WebBrain checkout installed 4 packages in 11 seconds and passed its available tests in 74 seconds, with 0 npm audit findings. Use it when an agent must work in the signed-in browser session you already have, especially through a local model or MCP client.

CozyClay

NomaDamas/CozyClay · ★ 1.3k ▲217/7d · JavaScript

CozyClay is a browser-based studio for arranging a 3D scene, posing characters and setting camera shots before handing references to an AI video model. It exports frames, reference clips and prompts, and lets an AI assistant control the scene through an in-app panel, MCP or a terminal.

Verdict Our CozyClay build passed in 11 seconds, but its test command failed after 137 seconds without a failure diagnostic in the supplied tail. Try it for preparing scene and camera references if you can verify the exported shot before paying for video generation.

no-negative-echo

LB623/no-negative-echo · ★ 954 ▲14/7d · Python

No Negative Echo is primarily documented in Chinese, with a complete English README available. It is an Agent Skills-format instruction set and scanner that asks coding agents to write final titles, commits, pull requests, and handoffs from the accepted result instead of repeating ideas that were rejected during the conversation.

Verdict Our no-negative-echo run installed 35 packages and passed 105 collected tests in 15 seconds, making it cheap to inspect and try. Use it when rejected ideas routinely leak into commit messages or handoffs, especially after long agent sessions.

Mimi

yuxino/Mimi · ★ 763 ▲23/7d · Rust

Mimi's primary README is Chinese, and maintained English setup and usage guides are available. It is a desktop and Android app that turns computer or microphone speech into floating original-language or translated subtitles, using cloud speech providers or supported Apple on-device services.

Verdict Our Mimi run built in 17 seconds and passed all 2,019 frontend tests in 79 seconds, the cleanest verification result in this batch. Use it when you want one floating subtitle window across desktop apps and accept the privacy, billing, and latency tradeoffs of your chosen provider.

hermes-agent

NousResearch/hermes-agent · ★ 259k ▲1.8k/7d · Python

Consider Hermes Agent when you want a persistent assistant across terminal sessions and messaging services and can manage its access to commands, files, and credentials. Hermes Agent documents persistent memory, reusable skills, scheduled jobs, and a messaging gateway. Our run on September 28, 2026, at commit 9a0a162 established that the root npm installation completed; it did not verify a working assistant deployment.

Verdict Try Hermes Agent in a restricted environment if persistent memory, skills, scheduling, and messaging match your needs. Our September 28, 2026 run at commit 9a0a162 completed root npm installation, but application behavior remains unverified.

opencode

sst/opencode · ★ 223k ▲1.8k/7d · TypeScript

OpenCode is an open-source coding agent that works in a terminal and, in beta, as a desktop app. It gives developers a read-only planning mode and a full-access build mode, so the same tool can inspect a codebase or change it under different permission rules.

Verdict Our OpenCode install reached the 900-second cap without finishing, so contributors should prove the Bun workspace on their own machine before committing to it. The product is still a strong candidate for developers who want open source, model choice, and a clear split between planning and editing.

skills

anthropics/skills · ★ 185k ▲1.6k/7d · Python

Anthropic Skills is a public collection of instruction folders, scripts, and examples that teach Claude how to handle repeatable jobs. It gives developers working patterns for document creation, design, testing, API work, and building their own skills without starting from an empty file.

Verdict Our sandbox installed 39 packages in 23 seconds and built successfully, yet pytest collected 0 tests, so Anthropic Skills is useful source material rather than a pre-validated toolkit. Use it to study or adopt one skill whose scripts, permissions, and license you have checked.

awesome-design-md

VoltAgent/awesome-design-md · ★ 125k ▲1.2k/7d

awesome-design-md is a curated library of DESIGN.md files derived from the visual language of real technology websites. You copy one into a project so an AI coding or design agent has concrete guidance on typography, color, spacing, components, and overall feel instead of improvising from a vague style prompt. It is a reference collection, not a UI framework or executable application.

Verdict Our run produced no install, build, or test result because commit 8147538 has neither a supported language ecosystem nor a Dockerfile. Use awesome-design-md as a rich prompt library and learning resource, especially when an AI-generated page keeps drifting stylistically.

taste-skill

leonxlnx/taste-skill · ★ 106k ▲3.9k/7d · JavaScript

Taste Skill is a collection of structured prompts, or "skills," designed to improve the design quality of AI-generated frontends. It solves the problem of AI agents producing generic, boilerplate user interfaces by giving them expert-level guidance on layout, typography, motion, and spacing.

Verdict Taste Skill is an essential tool for anyone serious about using AI for frontend development. It brilliantly addresses the most common weakness of AI-generated code, its lack of design taste, by packaging expert guidance into a simple, effective format.

ragflow

infiniflow/ragflow · ★ 93k ▲367/7d · Go

RAGFlow is a self-hosted application for turning documents and other data into searchable knowledge bases, cited answers, and agent workflows. It handles parsing, chunking, retrieval, model connections, and a web interface, which saves teams from assembling those parts as separate services.

Verdict Our RAGFlow run installed 34 packages in 58 seconds and built in 11 seconds, but the test command exited after 10 seconds because a Zhipu AI key was missing. Use it when document parsing, inspected chunks, citations, and agentic retrieval justify operating a full data service.

servers

modelcontextprotocol/servers · ★ 92k ▲145/7d · TypeScript

This repository holds seven reference servers maintained by the Model Context Protocol steering group. They show developers how MCP tools, resources, prompts, and transports work, but the maintainers explicitly describe them as educational examples rather than production services.

Verdict Our run built all 156 files in 19 seconds, but it had no test target and npm audit found 9 known vulnerabilities, so this is useful teaching code with a clear production boundary. Use it to learn MCP shapes, test a client, or borrow a small pattern.

career-ops

career-ops-hq/career-ops · ★ 79k ▲1.6k/7d · JavaScript

Career Ops is a local job-search workflow that runs through an AI coding CLI, including Claude Code. It compares job postings with your CV, prepares tailored documents, and tracks applications while leaving review and submission to you.

Verdict Our Career Ops run installed 8 packages in 39 seconds and used 23 MB, but build and tests were skipped, leaving its job-search workflow unverified by that run. Try it if you already use an AI CLI and will review every tailored CV before submitting.

headroom

headroomlabs-ai/headroom · ★ 78k ▲570/7d · Python

Headroom compresses logs, tool results, retrieved text, files, and conversation history before an AI model reads them. It runs as a Python or TypeScript library, a local proxy, an agent wrapper, or an MCP server, and it can retain the original material for later retrieval.

Verdict Our Headroom run installed 222 packages into 6,807 MB and reached only 47% of the test suite before the 900-second cap, so the full setup is a serious local dependency. Trial it for long, tool-heavy agent sessions where you can measure answer quality and provider cost with and without compression.

awesome-claude-skills

ComposioHQ/awesome-claude-skills · ★ 78k ▲421/7d · Python

Awesome Claude Skills is a catalog of instruction packs for Claude Code, Codex, Cursor, and other agents. It combines links to outside projects with skills stored in the repository, so it solves discovery rather than giving every entry one shared installer or quality bar.

Verdict Our Slack GIF Creator run installed 39 packages in 28 seconds and built in 9 seconds, but the repository supplied no test target, so Awesome Claude Skills is a useful catalog rather than evidence that every entry is safe to trust. Use it to find one promising workflow, then read that skill's scripts, permissions, dependencies, and license before giving it credentials.

models

tensorflow/models · ★ 78k · Python

TensorFlow Model Garden collects model implementations and training components for tasks such as recognizing images and understanding text. It gives TensorFlow developers starting recipes and pretrained checkpoints so they can adapt existing work to their own data.

Verdict Our TensorFlow Model Garden run installed 141 packages in 33 seconds, used 2,903 MB, and passed the build, giving TensorFlow teams a workable starting environment without a verified test result. Use it for a specific official model you intend to adapt and maintain.

oh-my-openagent

code-yeongyu/oh-my-openagent · ★ 71k ▲174/7d · TypeScript

OmO is a coding and research agent that can divide one job among many model-backed workers, preserve memory, and continue long tasks. It now runs as its own `omo` command, while older OpenCode and Codex integrations remain part of the repository and migration path.

Verdict Our OmO run installed 476 packages and used 2,178 MB, built in 58 seconds, then remained unfinished when the test suite hit 900 seconds. Try it if multi-model delegation is valuable enough that you will cap concurrency, cost, permissions, and wall time outside the agent itself.

free-claude-code

Alishahryar1/free-claude-code · ★ 65k ▲1.8k/7d · Python

Free Claude Code is a local proxy and control panel that connects Claude Code, Codex, and other coding agents to cloud or local model providers. It gives you one model catalog, fallback routing, and client launchers, but it does not turn paid models into free services or remove provider quotas.

Verdict Our Free Claude Code run installed 116 packages and built in 8 seconds, but its test command failed before collection because `pytest_asyncio` was missing. Use it if keeping familiar coding clients while switching providers is worth running a local control plane and checking each provider yourself.

yolov5

ultralytics/yolov5 · ★ 58k ▲25/7d · Python

YOLOv5 is a PyTorch codebase for object detection, instance segmentation, and image classification. It trains custom models, runs pretrained checkpoints, and exports them for several runtimes. Its maintainer now positions this repository for legacy projects and reproducibility, while directing new Ultralytics work to `ultralytics/ultralytics`.

Verdict Our run of YOLOv5 installed 89 packages using 6,001 MB, then stopped test collection because `libGL.so.1` was missing, so a fresh project should start with the main Ultralytics repository instead. Keep YOLOv5 for an existing model, a pinned checkpoint, or a reproduction that depends on its scripts.

CLI-Anything

HKUDS/CLI-Anything · ★ 58k ▲716/7d · Python

Use CLI-Anything when you need a reusable command interface for an application and can review and maintain its wrapper. CLI-Anything combines CLI-Hub, a catalog and package manager for existing CLIs, with an agent-driven generator for missing interfaces. Our run on September 27, 2026, at commit 34f5195 tested the cli-hub subproject in our sandbox; it did not establish that a generated wrapper could complete your application workflow.

Verdict Use CLI-Anything if you can own and verify the wrapper for a specific application task. Our run on September 27, 2026, at commit 34f5195 recorded a 26-second installation.

multica

multica-ai/multica · ★ 54k ▲612/7d · Go

Multica is a source-available workspace where people assign issues to coding agents, watch their runs, and review the result on a shared board. It coordinates existing agent CLIs such as Claude Code and Codex across machines, projects, schedules, and chat channels; it does not provide the models or agent subscriptions.

Verdict Our Multica run passed all 5,961 tests but failed its 14-second build in `@multica/desktop`, so the codebase has strong test coverage and a real source-build barrier at the measured commit. Try the cloud or packaged desktop path if you need one control room for several coding agents; self-host only after your team accepts the daemon, database, telemetry choice, and custom license.

new-api

QuantumNous/new-api · ★ 51k ▲383/7d · Go

New API is a self-hosted gateway that puts OpenAI, Anthropic, Gemini, and other model providers behind one managed service. Its default README and linked documentation are available in English, while much of the issue activity is in Chinese. It solves the practical mess of routing requests, issuing client credentials, and tracking usage across several upstream accounts.

Verdict Our New API run built in 6 seconds, but 5 of 89 Go test packages failed, so adopt it only if its broad gateway console is worth your own release gate. It is a strong fit for an operator managing several providers, users, quotas, and protocol shapes from one service.

langgraph

langchain-ai/langgraph · ★ 44k ▲301/7d · Python

LangGraph is a low-level Python framework for agents and workflows that must keep state, pause, resume, and survive failures. You describe work as a graph of steps and transitions, then add persistence or human interrupts where a long-running process needs control.

Verdict Our LangGraph CLI run installed 66 packages in 41 seconds and built in 6 seconds, but pytest ran 0 of the 510 collected tests because it rejected `asyncio_mode`. Use LangGraph when resuming durable, inspectable state is central to the product and your team wants control over each transition.

agno

agno-agi/agno · ★ 43k ▲93/7d · Python

Agno is a Python framework and runtime for building agents, teams, workflows, and the services around them. It starts with small model-and-tool scripts, then extends into an API, storage, traces, access control, and a web interface for running an agent platform.

Verdict Our Agno run installed 137 packages in 55 seconds and built in 9 seconds, but it offered no test target and pip-audit found 9 known vulnerabilities. Use it when you want a Python agent to grow into an owned API and control plane without changing frameworks.

chatbox

chatboxai/chatbox · ★ 42k ▲58/7d · TypeScript

Chatbox Community Edition is a desktop client for talking to hosted AI providers and local models from one interface. It keeps conversation data on your device and runs on Windows, macOS, and Linux, while web and mobile versions cover lighter use away from the desktop.

Verdict Our Chatbox run built in 171 seconds and passed 4,506 of 4,596 tests, but 2 attachment-ownership tests failed. Use the packaged app when you want local conversation storage and one polished desktop client for several providers.

OpenMAIC

THU-MAIC/OpenMAIC · ★ 42k ▲509/7d · TypeScript

OpenMAIC turns a topic or uploaded material into an editable browser classroom with slides, quizzes, simulations, project work, narration, and AI-led discussion. It solves the gap between getting an answer from a chatbot and assembling a lesson that a learner can play through, revise, and export.

Verdict Our OpenMAIC run installed 2,546 packages and built successfully, but 1 test timed out after 8,848 had passed, so adoption should start with a pinned sandbox and your own release gate. Use it when an editable, interactive classroom is worth operating PostgreSQL and several provider choices.

PageIndex

VectifyAI/PageIndex · ★ 40k ▲220/7d · Python

PageIndex helps a language model find information in long documents by organizing each document as a tree of sections. Instead of chopping text into fixed chunks and searching a vector database, it asks a model to navigate that tree and read the relevant pages.

Verdict Our PageIndex run installed 106 packages and completed the build plus pytest in 154 seconds, with 703 reported passes, 0 failures, 100 skips, and 1 audit finding. Trial it when section structure matters more than word similarity and your corpus is mostly text PDFs.

CopilotKit

CopilotKit/CopilotKit · ★ 38k ▲174/7d · TypeScript

CopilotKit is a TypeScript toolkit for putting AI agents inside apps, with chat, agent-rendered interfaces, shared app state, and approval steps. It connects a user interface to an agent through AG-UI, an event protocol, and provides a runtime that sits in your server. The open source SDK handles the interaction layer, while persistent threads, memories, analytics, and learning sit in the CopilotKit Intelligence product.

Verdict Our CopilotKit run installed 4,778 packages and used 4,450 MB, then returned exit 1 from the test command despite 29 passing tests and 0 failed assertions. Use it for a serious React product that needs shared state, rendered tools, and human approval around an existing agent, especially if AG-UI matches your architecture.

mediapipe

google-ai-edge/mediapipe · ★ 37k ▲75/7d · C++

MediaPipe adds features such as hand tracking, pose detection, and audio classification to applications that process data on the device. It supplies trained models and task APIs, plus a lower-level framework for developers who need to assemble their own media-processing pipeline.

Verdict Our MediaPipe pnpm install failed after 6 seconds because protobufjs@7.5.6 build scripts were ignored, leaving that checkout path unvalidated. Trial it when an existing on-device task matches your feature and you can manage model assets and platform-specific behavior.

oh-my-pi

can1357/oh-my-pi · ★ 37k ▲645/7d · TypeScript

Oh My Pi, or OMP, is a terminal coding agent that connects language models to editing, search, language servers, debuggers, subagents, browser control, and desktop automation. It aims to replace a stack of separate coding-agent add-ons with one configurable tool.

Verdict Our OMP run installed 421 packages and used 1,542 MB, then its build and test commands both exited with code 1, so adopt the packaged CLI before committing to source development. It suits an experienced terminal user who will use its language-server, debugger, subagent, and provider controls enough to justify the surface area.

Vibe-Trading

HKUDS/Vibe-Trading · ★ 37k ▲552/7d · Python

Vibe-Trading is a local research workspace that turns plain-English finance questions into data pulls, backtests, reports, and reusable research records. It also connects to brokers for read-only account work, paper trading, and bounded live trading, but the maintainers describe broker execution as experimental.

Verdict Our Vibe-Trading run installed 209 packages, used 800 MB, and reached only 62% of the tests before the 900-second cap, so it fits teams willing to audit a large research system rather than depend on a quick green check. Its strongest case is the connection between agent research, deterministic backtests, saved evidence, and several client surfaces.

next-ai-draw-io

DayuanJiang/next-ai-draw-io · ★ 37k ▲99/7d · TypeScript

Next AI Draw.io turns plain-language requests, documents, and images into editable draw.io diagrams. You can revise the result in the familiar draw.io editor or ask the model to change the underlying diagram XML.

Verdict Our run passed all 174 tests, but npm audit reported 51 known vulnerabilities, so Next AI Draw.io fits a trusted local or tightly controlled deployment better than an exposed shared service. Use it when editable draw.io output is worth more than a flat generated image, especially from Claude Code or another MCP client.

laya

NandhaKishorM/laya · ★ 35k ▲408/7d · Python

Laya is a local decision model that turns text into typed choices, scores, or yes/no probabilities in one model pass. Its router selects among English, multilingual, and specialized checkpoints, giving applications a smaller decision layer for triage, guardrails, and routing.

Verdict Our Laya run installed 80 packages into 5,472 MB, built successfully, and then hit 2 pytest collection/setup errors, so adoption needs more than a quick pip install. Try it when typed local decisions can replace repeated generative calls and you have labelled data for evaluation or fine-tuning.

nacos

alibaba/nacos · ★ 33k ▲11/7d · Java

Nacos is a server for service discovery, health checks, shared configuration, and service metadata. It gives microservices a common place to find one another and receive configuration changes, while recent releases also manage AI prompts, MCP servers, and tool imports.

Verdict Our Nacos console run installed 1,400 packages, built in 52 seconds, and exposed 94 known vulnerabilities, so a passing build is only the start of the adoption decision. Use Nacos when service discovery and live configuration already belong to your platform team and its AI registry features reduce another control plane.

cua

trycua/cua · ★ 31k ▲415/7d · Rust

Choose Cua if your agent needs native desktop controls or computers to run on, and you can validate the specific component on your target platform. Cua Driver provides tools to inspect and operate native desktop apps and browsers on macOS, Windows, and Linux. Our run covered a root install, not desktop automation or VM operation.

Verdict Cua deserves a component-level trial when an agent needs desktop controls, local VMs, or computer-use evaluation tools. Our successful root install does not establish that those products work on your target machine.

FastGPT

labring/FastGPT · ★ 30k ▲47/7d · TypeScript

FastGPT's primary README and current release notes are written in Chinese, although the project supplies an English README and English deployment docs. It is a visual system for building AI agents, retrieval-backed chat, knowledge bases, and tool workflows, with hosted and self-hosted editions.

Verdict Our FastGPT run pulled 2,598 packages and occupied 2,480 MB, then its test command failed because Turborepo could not find `@fastgpt/admin`. Trial it when the visual agent, RAG, MCP, and operations console can replace several tools and you have people to run its stateful services.

page-agent

alibaba/page-agent · ★ 30k ▲51/7d · TypeScript

Page Agent is an English-documented TypeScript library that puts a natural-language operator inside a web page. Chinese documentation is also available. It turns the page DOM into text for a language model, then clicks, types, selects, or scrolls without taking screenshots or starting a separate headless browser.

Verdict Our Page Agent run installed 742 packages and used 460 MB, then passed its build and all 4 tests, while npm audit still found 4 high-severity advisories. Use it for a copilot inside a web product you control, especially when the interface has clean semantic HTML.

Chat2DB

OtterMind/Chat2DB · ★ 28k ▲13/7d · Java

Chat2DB Community is a local-first database client and SQL workspace for more than 40 databases, with desktop and browser interfaces. Its assistant uses a model you connect yourself to generate, explain, and optimize SQL, while the same app handles schema browsing, data editing, imports, exports, charts, and ER diagrams.

Verdict Our Chat2DB install failed after 239 seconds because Yarn rejected `marked@16.4.2` under the detected Node 18.20.8, so source users should check the resolved dependency tree before trusting the documented Node 18.17 floor. The packaged desktop app is still worth trying for one technical user who wants broad database coverage and optional bring-your-own-model SQL help.

pm-skills

phuryn/pm-skills · ★ 28k ▲177/7d

PM Skills is a collection of instructions and guided workflows that help an AI assistant with product discovery, planning, and launch decisions. It gives product managers a repeatable way to question assumptions and draft working documents inside tools such as Claude Code and Cowork.

Verdict Our PM Skills run installed 35 packages in 5 seconds and passed all 15 tests, making it an inexpensive instruction pack to try with an existing assistant. Use it to structure discovery and working documents when you can check the evidence yourself.

baoyu-skills

JimLiu/baoyu-skills · ★ 27k ▲140/7d · TypeScript

Baoyu Skills is an English-documented collection of more than 20 agent skills for creating images, diagrams, slides, formatted articles, translations, and social posts. It gives Claude Code, Codex, and other file-capable agents reusable publishing procedures instead of asking you to rebuild the same long prompt for every job. A Chinese README is also available.

Verdict Our Baoyu Skills run installed 326 packages and passed 236 of 238 tests, but npm audit found 11 vulnerabilities, including 8 high-severity findings. Use it as a shelf of selected publishing tools, especially for visual explainers or WeChat work, and review each chosen skill before it reaches credentials or a public account.

MiniCPM-V

OpenBMB/MiniCPM-V · ★ 27k ▲65/7d · Python

MiniCPM-V is a family of models for asking questions about images, documents, and video on local devices or servers. The repository also contains MiniCPM-o for speech interaction, and provides English and Chinese documentation.

Verdict Our MiniCPM-V run installed 35 packages in 9 seconds and built in 1 second, but tests were skipped because no runnable target was found, leaving model behavior unverified by our lab. Try it when local vision processing matters and you can evaluate an exact model and runtime on your own documents.

A2A

a2aproject/A2A · ★ 26k ▲66/7d · Shell

A2A is an open protocol for one software agent to discover and hand work to another across frameworks or organizations. It standardizes machine-readable Agent Cards, messages, long-running tasks, streaming updates, and push notifications over HTTP while each agent keeps its model and internal tools private. This repository holds the protocol and documentation, while working clients and servers live in separate SDK repositories.

Verdict Our A2A checkout installed 35 packages in 7 seconds and built in 1 second, but the repository exposed no test target. Use A2A v1.0 when agents cross a real team, vendor, or service boundary and long-running task state matters.

faster-whisper

SYSTRAN/faster-whisper · ★ 26k ▲207/7d · Python

Faster-whisper is a Python library that turns audio into text using Whisper models on your own machine. It uses the CTranslate2 inference engine and gives an application control over CPU or NVIDIA GPU execution, batching, and timestamps.

Verdict Our faster-whisper run passed all 19 tests, but its 62-package environment carried 6 known vulnerabilities in pip-audit. Use it for local transcription inside a Python application after investigating those findings and proving your chosen runtime.

gallery

google-ai-edge/gallery · ★ 25k ▲54/7d · Kotlin

Google AI Edge Gallery is a mobile and macOS app for downloading, running, and comparing open models on your own hardware. It puts chat, image questions, audio transcription, prompt controls, model benchmarks, skills, and an experimental MCP client in one interface. It solves the awkward first step of finding out whether an on-device model works on a particular phone or computer.

Verdict Our 3-CPU, 8 GB sandbox produced no install, build, or test result for Gallery because the Kotlin repo fell outside the harness and supplied no Dockerfile. Use the packaged app when you need to compare on-device models on real hardware before committing to a runtime.

prime-agent

PrimeIntellect-ai/prime-agent · ★ 22k ▲135/7d · Rust

Prime Agent is an open-source terminal agent for coding, research, and jobs that may outlast one chat session. It keeps a Python workspace, child agents, goals, and scheduled work alive behind a background service, solving the problem of an agent losing its working state when the terminal disconnects.

Verdict Our Prime Agent run built in 268 seconds and reached 775 passing tests with no failures, but the full command was still running when the 900-second cap stopped it. That makes it a serious option for developers who need persistent, multi-agent work and can budget for a large Rust codebase plus a slow release gate.

skills

google/skills · ★ 22k ▲126/7d · Python

Google's `skills` repository is a collection of pre-built capabilities for AI agents, designed to automate complex tasks within the Google Cloud Platform. Instead of writing low-level API calls, this library provides high-level "skills", like deploying an AI model on Kubernetes or setting up a database, that an agent can use directly. It solves the problem of bridging the gap between a general-purpose AI agent and the specialized, intricate world of cloud management.

Verdict If your organization lives and breathes Google Cloud, `google/skills` is an essential toolkit for building the next generation of infrastructure automation. It packages Google's own expert knowledge into executable skills for AI agents, saving countless hours of development.

onnx

onnx/onnx · ★ 22k ▲14/7d · Python

ONNX is an open file format and specification for moving machine-learning models between frameworks, tools, and hardware. It gives teams a shared representation for model graphs, operators, and data types, reducing the need to tie training and inference to the same vendor stack.

Verdict Our installation did not finish before the 900-second cap. ONNX is still a strong default interchange contract when models must cross framework or hardware boundaries, backed by a current codebase and unusually detailed specifications.

FunASR

modelscope/FunASR · ★ 21k ▲74/7d · Python

FunASR is a Python toolkit for offline and streaming speech recognition, voice activity detection, punctuation, speaker pipelines, emotion tags, and speech serving. Its main README is English and links Simplified Chinese, Japanese, and Korean versions, but each checkpoint and runtime supports a different slice of those jobs.

Verdict Our FunASR run used 6,040 MB and built in 7 seconds, but pytest executed 0 tests because collection/setup ended with `SystemExit: 2`. Trial it when its Chinese and multilingual checkpoints, streaming path, or edge runtimes solve a requirement that a narrower Whisper stack does not.

voyager

voyager-crew/voyager · ★ 21k ▲38/7d · TypeScript

Voyager is a browser extension that adds folders, reusable prompts, message timelines, and other controls to AI chat websites. It helps you find and reuse work inside Gemini, AI Studio, ChatGPT, Claude, and DeepSeek, with different features available on each site.

Verdict Our Voyager run passed all 12,084 tests, but its build failed after 3 seconds because catalog freshness requires full Git history. Try the browser release if folders and reusable prompts would improve your existing chat workflow, and check feature coverage on your exact site.

nuclear

nukeop/nuclear · ★ 19k ▲54/7d · TypeScript

Nuclear is a free desktop music player that searches multiple sources, streams tracks, manages a library, and avoids ads or tracking. Its plugin system supplies music sources and metadata, while an optional MCP server lets AI clients control playback.

Verdict Our Nuclear build completed 8 of 9 tasks, then failed after 67 seconds because `cargo metadata` was unavailable, while all 5 Vitest tests passed. Download the packaged player if you want ad-free desktop streaming and are comfortable selecting source plugins.

SkillOpt

microsoft/SkillOpt · ★ 19k ▲108/7d · Python

SkillOpt improves the Markdown instruction files that guide an AI agent while leaving the model itself unchanged. It runs tasks, studies the results, proposes bounded edits, and keeps a candidate only when it passes validation; a separate preview called SkillOpt-Sleep can learn from supported coding-agent session histories.

Verdict Our SkillOpt run passed 1,499 tests after a 116-package, 652 MB install, so the research engine earns a serious trial. Use it when you can define held-out tasks and score the behavior you want; that validation discipline is the product.

Auto-claude-code-research-in-sleep

wanshuiyin/Auto-claude-code-research-in-sleep · ★ 18k ▲284/7d · Python

ARIS is an English-first collection of agent skills and a standalone CLI for machine-learning research, with separate Chinese documentation available. It gives Claude Code, Codex, and other coding agents repeatable steps for finding papers, planning experiments, reviewing results, and preparing a paper.

Verdict Our Feishu bridge run installed 43 packages in 19 seconds and built successfully, but pytest found 0 tests and exited 5. ARIS is a credible choice for an ML researcher who wants structured overnight work and will audit the morning output.

xiaohongshu-mcp

xpzouying/xiaohongshu-mcp · ★ 17k ▲118/7d · Go

The primary documentation is in Chinese, with a separate English README available. xiaohongshu-mcp gives an MCP client logged-in access to Xiaohongshu for searching, reading posts, publishing media, and handling interactions through browser automation.

Verdict Our run built xiaohongshu-mcp in 40 seconds and passed all 16 tests, so the code is easier to verify than the browser account it controls. Use it for a supervised Xiaohongshu workflow where search, publishing, and replies belong in one MCP service.

trigger.dev

triggerdotdev/trigger.dev · ★ 17k ▲48/7d · TypeScript

Trigger.dev runs long-lived TypeScript jobs and AI agent workflows with queues, retries, schedules, waits, logs, and deployment history. It gives application developers a place for work that outlives an HTTP request, either on Trigger.dev Cloud or on infrastructure they operate.