Ryan

vibe coded a fuzzing ai agent last month and let it run for a week using my $200 claude max. it then found 21 high/critical vulnerabilities in Chrome.

For anyone dealing with RASP protected apps, frida-strace is now your first step. Trace the syscalls, find what the app checks, hook those specific functions, bypass. No more guessing. Frida 17.8.0+, kernel 6.1+ required. #Frida #MobileSecurity #AppSec

A nice feature of evaluating on zero-days is that you can be pretty sure the exploits aren't in the training corpus 😁

Every AI agent sandbox project

Syscall Tracer🔥🔥 Sometimes it’s useful to observe the system calls happening inside a given target process. Especially if the target includes some kind of Frida detection, root detection, or any other kind of Runtime Application Self-Protection (RASP). frida.re/news/2026/03/0…

the full preprint is available here: arxiv.org/pdf/2603.08225 source code will be made available soon! also thank you to my co-authors @xorpse @binarly_io @mlsec @bifoldberlin

this last week, most of my posts on here about LLMs probably sound like "so i checked out fire. seems warm. i think we can use this. last night i got close to it and i wasn't that cold anymore"

This is a very interesting piece of research by the team on algorithmic methods for effective complex type recovery. Practical type inference is a challenging problem the industry has been grappling with for years. paper: arxiv.org/abs/2603.08225 blog: binarly.io/blog/type-infe…

Next features in mind: - MCP Server - Git log <struct/field>: track how stable a structure is across 28 years of Windows - Binary hardening timeline with checksec analysis Happy to hear your ideas ! 💡

One weird trick for getting Codex to run for a very long time is to ask it to boot a kernel in QEMU for a machine that QEMU doesn't support. It's been at it for like 8 hours, making steady forward progress and getting further into the boot process

If you find Claude Code with local models to be 90% slower, it's because CC prepends some attribution headers, and this changes per message causing it to invalidate the entire prompt cache / KV cache. So generation becomes O(N^2) not O(N) for LLMs.

1/ 🧵 If you’re building in ZK, your code is only as strong as your math. I’m sharing a gem from Carnegie Mellon University (CMU): "Everything You Always Wanted To Know About Mathematics." A must-read for systems engineers. 🦀✨

Love seeing these talks at @REverseConf all using AI or agentic approaches. BSim uses cosine similarity search and the Apple patch diffing talk uses MCP agents. The students of my training class this week will understand what’s going on under the hood of these systems!

powered by sem github.com/Ataraxy-Labs/s…

This is very much true. The crash analysis subagents that I wrote that made it into raptor take exactly that approach.

I cant believe this guy just made a permanent solution to context bloat and open sourced it all! when we tested this tool (Context+) for solving an issue on the OpenCode repository, the agent using this tool used ~6.5k fewer tokens, found the code and fixed it in half the time! the results were surprising: 6 to 10k tokens saved per prompt, completed task in ~2 minutes while the agent running without the tool took ~4 mins for the same and got stuck in loops bro built an entire beast by using all the modern tools that we could think of: undo trees, semantic search by meaning (by haskellforall), advanced refactoring, blast radius, advanced file context trees, restore points... i can keep going on semantic code search and context trees are the future of agentic coding and this tool proves it the feature i loved the most is semantic search and how it gets things done 2x faster with least possible tokens it makes an agent that actually knows what it’s doing and not just guessing, it makes meaning from your code similar to RAG. if you aren't optimizing your context, you are just burning money the developer says this tool is still under development, it can have unexpected behavior and the docs need updates but the video shows the reality of how fast it can be github: github.com/ForLoopCodes/c… get here: contextplus.vercel.app

POV: you just started a new AI side project from scratch and are enjoying the dopamine rush before the existential dread appears and you send it to your dead idea graveyard

I replaced Claude Code’s default --worktree command with a custom one built around Ghostty, Lazygit, and Yazi. By default, Claude Code creates worktrees inside the .claude/worktrees folder within the same project. That means if you spin up 3 worktrees, you end up with 3 complete copies of your project nested inside your main project. It makes the project structure messy and difficult to manage. So I built a hook that: - Overrides the default --worktree command - Creates each branch in a sibling directory: ../worktrees/branch-name - Automatically opens Ghostty panes - Launches Lazygit and Yazi already positioned in the correct branch directory Install it with: npx claude-code-templates@latest --hook=development-tools/worktree-ghostty --yes Now each branch lives where it should, outside the main project, and your terminal environment is ready instantly.

Recently I gave a talk on LLMs for Math Research (mostly to an audience of pure and applied mathematicians) I tried to compile the latest progress in one presentation pdf and video recording: drive.google.com/drive/folders/…

If you didn't know, the wonderful folks at @OpenAI have an open source evals package you can build with and learn from github.com/openai/evals It also has fantastic examples to learn from and take baby steps towards building your own evals.