Open Source Picks
Every repo covered on the DevPik channel, written up properly — what it actually does, how to install it, where it beats the alternatives, and when it doesn't. Each page links straight to the official repository.
gstack: Garry Tan's 57-Skill Claude Code Setup, Measured
Garry Tan's gstack turns Claude Code into a sprint of specialist skills, from /office-hours to /ship. The README still says 23 specialists; the repo now ships 57. We counted what they cost: about 1,200 tokens a session to keep installed, and about 107,000 to run the full sprint.
Scientific Agent Skills: Turn Your AI Agent Into an AI Scientist
Scientific Agent Skills gives any AI coding agent 177 validated research skills and 100+ scientific databases, covering drug discovery, genomics, proteomics, astronomy and lab automation. It has 49,900+ stars, an arXiv paper, and a one-command install.
Caveman for Claude Code: The Real Savings Are Not in the Grunting
Caveman started as a joke prompt that made Claude grunt. 111,000 stars later, the repo's own Honest Numbers page says the voice saves about 3% over "Answer concisely", and version 4.0 rebuilt the project around the tokens your agent reads instead.
Diagram Design: The Claude Code Skill That Writes Taste Down
The model already understood your architecture. What it was missing was taste. Diagram Design writes the taste down as a loadable reference, and the result is diagrams that do not look generated.
Matt Pocock's Skills: The Open-Source Agent Skills That Fix How Coding Agents Fail
Matt Pocock's skills repo is a collection of agent skills he uses every day for real engineering, not vibe coding. Built to fix the four ways coding agents fail, it gives Claude Code, Codex, GitHub Copilot and Gemini CLI composable workflows for grilling, TDD, debugging, code review and architecture, and it has 283,000+ stars.
Text-to-CAD: The Open-Source Plugin That Lets Your AI Agent Design Real CAD Parts
Text-to-CAD is an MIT-licensed plugin that gives Claude Code, Codex, Cursor, Gemini and Grok a full CAD workflow. Instead of generating a mesh, your agent writes a parametric Python model that produces exact STEP geometry, then exports STL or 3MF, draws dimensioned drawings, checks manufacturability and sends parts to a printer.
REA: Reverse Engineer Anything With Your AI Agent, From Electron Apps to x86 Binaries
REA (Reverse Engineer Anything) is an MIT-licensed MCP server and CLI that gives your AI coding agent 139 reverse engineering tools. It drives Ghidra, IDA Pro or Hopper for native binaries and also inspects Electron apps, .NET assemblies, Android APKs, firmware and websites. Every finding comes back with the evidence behind it.
Career-Ops: The Open-Source AI Job Search Agent That Tells You Where Not to Apply
Career-Ops is a free, open-source AI job search agent that runs inside Claude Code, Codex and other AI coding CLIs. It checks that a posting is real, scores it 1 to 5 against your actual CV, tells you to skip anything below 4.0, and drafts a tailored CV and cover letter for the rest. It never presses Submit.
Open Dots: A Self-Hosted, Open-Source Alternative to OpenAI Dots
Open Dots is a free, MIT-licensed AI agent workspace you run on your own machine: assistant personas, any Responses-compatible model, app connectors, web search and an optional computer runtime. Where OpenAI Dots screens actions with a second model, Open Dots stops every risky action and asks you first. It is an early prototype, and this page covers what works today.
Ponytail: The Open-Source Prompt That Makes AI Agents Write 54% Less Code
Ponytail is a single open-source prompt that stops AI coding agents from over-engineering. Before writing code, the agent climbs a seven-rung ladder: does it need to exist, is it already in the codebase, does the stdlib do it, and only then the minimum that works. The result, measured in real Claude Code sessions: about 54% less code, 20% lower cost, 27% faster.
Hypit: Clone Any Viral Video with AI Agents, One Command, 100 Variants
Hypit is a free, open-source system that gives AI agents like Claude Code and Codex a language and toolchain for video. Drop in any viral video and your agent clones it as a complete, editable workflow, then ships 100 variants from a single command.
OpenVid: Studio-Quality Demo Videos and 3D Mockups, Right in Your Browser
OpenVid is a free, open-source web app that turns a screen recording or an uploaded clip into a studio-quality demo: device mockups, 3D transforms, cinematic zooms with a virtual camera, more than a hundred backgrounds, multi-track audio and export up to 4K MP4, all running in your browser.
Logo Design Skill: Turn Claude and AI Agents Into a Disciplined Logo Designer
Logo Design Skill is a free, MIT-licensed agent skill that turns Claude or any Agent Skills agent into a logo designer: a brief-to-checkpoint workflow, 1,400+ visually classified reference logos, dependency-free Python audit and export tools, and production-ready SVG output.
Anti-Slop: The Rulebook That Stops AI Coding Agents Shipping Generic UI, Copy and Code
Anti-Slop is a free, MIT-licensed rulebook for AI coding agents: 38 rules in three tiers, six skills and a pass/fail Delivery Gate that strip generic UI, filler copy and slop code out of what your agent ships.
Claude for Financial Services: Ten Analyst Agents, and What They Will Not Do
Anthropic published its reference agents for investment banking, equity research, private equity and fund admin. They draft analyst work product for human sign-off, and the prompts are free while the data behind them is emphatically not.
OpenStock: A Self-Hosted Market Dashboard, and What Free Data Really Costs
A 17,000-star Next.js market dashboard you can host yourself: watchlists, TradingView charts, and email digests. Here is what it genuinely gives you, and why the word real-time needs an asterisk on the free tier.
Agent-Native: Stop Making Your Agent Click Through the UI
Most tools teach an agent to operate your interface. Agent-Native argues the interface was never the API. Define a capability once and the agent, React, HTTP, MCP, A2A and the CLI all call the same function.
Octop: A Self-Hosted Multi-Agent AI Assistant That Runs as One Process
Most self-hosted AI stacks are a docker-compose file with five services in it. Octop is one process and a SQLite file that rebuilds its entire state on boot. Here is the architecture, what you actually get, and two things worth knowing before you pipe the installer into bash.
OpenResearch: Give Every Experiment Its Own Agent and Its Own Worktree
A local-first workspace that gives each research direction its own agent session and its own git worktree, then tracks every run in a git-native experiment tree. How it differs from Karpathy's autoresearch, and the shared-cluster gotcha its own README admits to.
BrowserSkill: Let AI Agents Drive the Browser You Are Already Logged Into
Most browser agents open a clean browser that is signed into nothing. BrowserSkill connects your agent to the profile you already use, in a separate window, borrowing tabs only when it asks. Here is the architecture, the safety switches, and the question you should ask before installing it.
Cloudflare's Security Audit Skill: An AI Auditor Built to Disprove Itself
A coding-agent skill that turns Claude Code into a security auditor, where the agent that checks a finding is never the agent that found it. How the six phases work, what the three verdicts mean, and the sandbox requirement most people will skip.
VoiceStudio: Local Voice Cloning and Dubbing in 646 Languages
A desktop app that runs 16 text-to-speech engines and 11 transcription engines on your own hardware, with no account and no usage meter. Here is what the 646-language claim actually means, and the licensing detail most write-ups leave out.
Colibri: Self-Host a 744B LLM on Hardware You Already Own
A single C file that runs frontier Mixture-of-Experts models on a normal desktop by streaming experts from disk. Here is how it works, what speeds it actually reaches, and what "runs" really means.
What Is a System Prompt? 430 Real Ones, Captured From Every Major AI
Every AI chatbot reads a hidden instruction set before your first message. One repo has captured 430 of them. Here is what a system prompt actually is, and what these files reveal.
DeskcommCRM: An Open Source WhatsApp CRM With AI Agents Built In
A self-hosted, MIT-licensed CRM that runs AI agents inside WhatsApp to qualify and close deals. Here is what it actually does, what running it really costs, and how it compares to Chatwoot and Kommo.