Cap creator launches Sandchest, an AI hacker that scans repos with abliterated models
JosephJacks_ · x · 2026-11-04
richiemcilroy, creator of Cap, launched Sandchest, a tool that embeds an AI hacker in your team. It runs abliterated (safety-stripped) open models — the same kind attackers use — against your GitHub repos to find vulnerabilities before exploitation.
Workflow: install the GitHub app, pick repos, and it maps them for free (routes, auth checks, data stores). The first full deep scan is free; each finding lands on the exact code line with a step-by-step attack path, a runnable PoC, what the fix must and mustn't break, plus a copy-paste fix prompt with regression test for your coding agent. Findings are delivered as reviewed PR comments. The author says it has already found real vulnerabilities in multiple projects.
More from coding & agent
- Corebrum OS: open-source Physical AI OS now installable on NVIDIA Jetson and any Ubuntu machine — chrismatthieu · 2026-11-04
- Reddit user open-sources full MiniMax H3 ComfyUI workflows: TTS, image editing, 3MP upscale — Support_Marmoset · 2026-11-04
- TLAPS-Bench: 265 real-world TLA+ proof targets; top model GPT-6.1 Sol solves only 44.3% — tianyin_xu · 2026-11-04
- Fireworks' Nexus Auto (Sol) first model to hit 60% on Vals Router Index at $1.25/test — sophiamyang · 2026-11-04
- One 421 error blocked a full day of posts: a retry playbook for WhatsApp agent send failures — petrusenko_max · 2026-11-04
- HeyGen's HyperFrames skill turns text into editable MP4 videos in Perplexity — toolstelegraph · 2026-11-04