A post I had to share from @Mišel Čupković (make sure you post here in AiTitus too :P you have great stuff) Who: Jamieson O’Reilly Articles on X: Part I: hacking clawdbot and eating lobster souls Part II: eating lobster souls Part II: the supply chain (aka - backdooring the #1 downloaded clawdhub skill) Part III: eating lobster souls Part III (the finale): Escape the Moltrix Be careful out there people! 🙏🏻

Fully autonomous AI hacker to find actual exploits in your web apps. Shannon has achieved a 96.15% success rate on the hint-free, source-aware XBOW Benchmark. ❤️ Shannon is open source (AGPL v3) 16.3K ⭐️

Paper Banana: AI Multi-Agent System for Publication-Quality Diagrams. Unlike single-model approaches, Paper Banana uses 5 specialized AI agents working together to create, critique, and refine technical diagrams with stunning accuracy. How It Works: PaperBanana implements a two-phase multi-agent pipeline with 5 specialized agents: - Phase 1 -- Linear Planning: - Retriever selects the most relevant reference examples from a curated set of 13 methodology diagrams spanning agent/reasoning, vision/perception, generative/learning, and science/applications domains -Planner generates a detailed textual description of the target diagram via in-context learning from the retrieved examples -Stylist refines the description for visual aesthetics using NeurIPS-style guidelines (color palette, layout, typography) - Phase 2 -- Iterative Refinement (3 rounds): Visualizer renders the description into an image (Gemini 3 Pro for diagrams, Matplotlib code for plots) Critic evaluates the generated image against the source context and provides a revised description addressing any issues - Steps 4-5 repeat for up to 3 iterations Open source ❤️ implementation and extension of Google Research’s PaperBanana ***Official code coming soon...

Built by a developer who refused to run code he couldn't audit, the project strips OpenClaw down to ~500 lines handling everything. Agents execute inside Apple Container sandboxes (or Docker on Linux), with filesystem isolation enforced at the OS level rather than through application permissions. Setup happens through Claude Code itself - no installation wizard, no configuration files to learn, just run /setup and the AI handles dependencies, authentication, container configuration, and service deployment. Highlights: 1. Container-Level Security - Agents run in lightweight Linux VMs where only mounted directories are visible. Bash access is safe because commands execute inside the sandbox, not on your host. Mount allowlist validation sits at ~/.config/nanoclaw/ where agents can't reach it. 2. Skills-Based Customization - Contributors write skills like /add-telegram that teach Claude Code how to transform your fork. No PRs adding features to core—users run skills and get clean code matching their exact needs without bloat. 3. Memory Hierarchy - Global memory in groups/CLAUDE.md gets read by all conversations but only written by main channel, while each group maintains isolated memory in its own folder. Claude Agent SDK automatically loads both levels when agents run. 4. Scheduled Task System - Groups schedule recurring or one-time tasks that run as full agents with complete tool access, optionally messaging results back. Main channel manages all tasks across groups, others manage only their own. Nanoclaw is open source ❤️ Clone the repo, run Claude Code, execute /setup, and you have a working assistant. *Requires macOS or Linux and Claude Code installed.

The new Claude Opus 4.6 improves on its predecessor’s coding skills. It plans more carefully, sustains agentic tasks for longer, can operate more reliably in larger codebases, and has better code review and debugging skills to catch its own mistakes. And, in a first for our Opus-class models, Opus 4.6 features a 1M token context window in beta.