AI Repo: tt-a1i/archify

Skyler: Hey everyone, welcome back to Git News A I, the show where we dig through the hottest A I repositories trending on GitHub so you don't have to. Big thanks to our sponsors today — Ada, helping teams integrate A I into their workflows seamlessly, Ago Consulting, that's "Ahh-Go," your go-to for silicon development from A I all the way down to SoC, and Zen Semiconductor, an umbrella company with a whole portfolio of silicon-industry companies, including Ada for deploying A I at scale in any business, plus their Vect appliance that runs local L L Ms and vector search on dedicated hardware. Love that ecosystem. Alright, today we've got a beautiful diagramming agent skill, a spicy offensive security library built for Claude, a wild repo of leaked A I system prompts, a fully autonomous penetration testing agent system, and a tool that turns any technical book into a Claude Code skill — let's get into it. Skyler: First up we have archify, from the tt-a1i organization on GitHub, sitting at nearly sixty thousand stars and climbing fast — this thing is blowing up. So archify is an agent skill designed to generate beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams. And the key word here is beautiful — this isn't your grandmother's box-and-arrow diagram generator. We're talking crisp exports, smooth motion, and the whole output is a self-contained H T M L file, which is genuinely clever because you don't need any special viewer or runtime to share it. Just open it in a browser, done. The "verifiable" aspect is really interesting too — the idea is that the diagrams aren't just pretty pictures, they're grounded in the actual structure of whatever system you're describing, so they hold up to scrutiny. Written in JavaScript, it's designed to plug into agentic workflows, so you can have your A I agent generate architecture docs automatically as part of a build or review process rather than doing it manually after the fact. Recent activity shows some solid maintenance work — they're making the packaging reproducible on Windows and tuning the CodeRabbit integration, which tells me this team is being thoughtful about developer experience. If you're building complex A I systems and you're tired of outdated architecture diagrams that nobody trusts, this is worth a serious look. M L engineers, platform teams, anyone doing system design reviews — absolutely check this out. Check the show notes for the link. Skyler: Okay, next one is going to raise some eyebrows, and I want to be upfront about that — we're covering Claude-Red from SnailSploit, currently sitting at around thirty-five hundred stars. So what is this? It's a curated library of offensive security skills specifically designed for the Claude skills system. Each skill is a structured SKILL dot M D file that primes Claude with expert-level methodology for a specific attack surface. We're talking SQL injection, shellcode, E D R evasion, exploit development — the works. Now, why are we covering this? Because from a purely technical standpoint, this represents a fascinating use of the Claude skills architecture. The idea that you can essentially give an L L M a structured knowledge module that shifts its reasoning toward deep domain expertise in a specific area — that's a powerful pattern, and the offensive security community has clearly figured out how to use it. The skills format itself is worth studying regardless of the domain, because the same approach could be used for defensive security, for incident response, for threat modeling. The repo is actively maintained, with recent commits improving skill depth and variety. That said, I want to be crystal clear — tools like this sit right on the ethical edge, and responsible use means authorized testing, legal engagements, and proper scope. If you're a red teamer, a penetration tester, or a security researcher operating within those boundaries, this is a genuinely interesting resource. If you're curious about the Claude skills architecture more broadly, the SKILL dot M D format here is educational on its own. Check the show notes for the link. Skyler: Alright, this next one is one of those repos that the A I community absolutely cannot stop talking about — system_prompts_leaks from asgeirtj, and wow, sixty-five thousand stars. That's a lot of people who are very curious about what's going on under the hood of their favorite A I products. So the premise here is straightforward — this is a collection of extracted system prompts from major A I providers. We're talking Anthropic's Claude models including Fable 5.1, Opus 5, Claude Design, and Claude Code. OpenAI's ChatGPT with something listed as G P T-6-Astra and Codex. Google's Gemini lineup. xAI's Grok. Cursor. Kimi. It's quite a list. And it's updated regularly, which based on the recent commits seems to be happening pretty actively. Now, I want to be clear about what this actually is — these are system prompts, not model weights, not training data. What it reveals is how these companies are instructing their models to behave, what guardrails they're setting, what personas they're crafting. From a developer perspective, this is honestly fascinating because it shows you the prompt engineering patterns that teams at the frontier labs consider production-ready. You can learn a ton about structuring instructions, handling edge cases, and setting tone just by reading through these. It also raises real questions about transparency and the gap between how these systems are marketed versus how they're actually configured. Whether you view this as a security research artifact, a prompt engineering study guide, or just satisfying curiosity — at sixty-five thousand stars, clearly the community has opinions. Check the show notes for the link. Skyler: Next we have pentagi from vxcontrol, and this one is genuinely impressive from a systems architecture standpoint — sitting at over twenty-three thousand stars. PentAGI is a fully autonomous A I agents system capable of performing complex penetration testing tasks end to end. Not just answering questions about pentesting, not just suggesting commands — actually executing them autonomously. It's built in Go, which is a really interesting choice for an agentic system, and it's leveraging LangChainGo under the hood, with a recent dependency upgrade to a newer release version. The OpenCode provider integration is also getting active development, which suggests the team is keeping pace with the rapidly evolving L L M provider landscape. What makes pentagi stand out architecturally is that it's designed for the full complexity of real penetration testing engagements — reconnaissance, exploitation, post-exploitation — not just one slice of the process. Building truly autonomous agents for security is hard because the decision trees are enormously complex and the consequences of wrong moves are real. The fact that this is getting serious traction at twenty-three thousand stars tells me the security research community is finding it genuinely useful rather than just theoretically interesting. If you're working in red team automation, autonomous security agents, or you're just fascinated by agentic systems applied to complex real-world tasks, pentagi is worth studying. Go developers in the A I space especially should look at how they've structured the agent orchestration here. Check the show notes for the link. Skyler: And our last repo today is book-to-skill from virgiliojr94, which has a deceptively simple premise and a huge following at over thirty thousand stars. The idea is this — you take any technical book in P D F format, and it converts it into a Claude Code skill that you can actually use while you're working. So instead of having a book sitting on your shelf or open in another tab, you have the knowledge from that book structured as a skill that Claude Code can draw on in real time. This is Python-based, and while the recent activity column is quiet, the star count tells the real story — developers clearly want this. Think about what this means practically. You've got a dense book on, say, distributed systems design or compiler theory or cryptography, and rather than context-switching constantly, you can have that expertise available contextually as you code. The conversion process essentially structures the knowledge in a way that fits the Claude skills format, making it both referenceable and applicable rather than just stored text. It's a really elegant intersection of document processing, knowledge structuring, and agentic tooling. The use cases are pretty broad — individual developers leveling up in a new domain, teams standardizing around shared technical references, anyone who learns by doing rather than reading. If you're already using Claude Code in your workflow and you have a stack of technical books you keep meaning to get through, this is a clever way to make that knowledge more actionable. Check the show notes for the link. Skyler: And that's a wrap on today's episode of Git News A I — five repos covering everything from diagram generation to autonomous security agents, and a whole lot of fascinating questions about how we're using and understanding these L L M systems. Huge thank you to our sponsors — Ada for helping teams actually ship A I in production, Ago Consulting for their expert silicon development work from A I down to SoC, and Zen Semiconductor for building the hardware foundation that makes local A I real with their Vect appliance. If you want to go deeper on A I developments and security topics, check out A I Special Report in the Ada podcast network — really solid companion show to this one. Make sure you subscribe so you never miss a trending repo, and as always, all the links we talked about today are in the show notes. See you next time.