We believe that the BitGN Sandbox has successfully passed the resilience test.
Engineers connect to Harness, launch AI agents within the environment, and already receive initial assessments of their actions. Afterwards, they ask questions about potential traps and pitfalls that may be encountered during the competition on April 11.
Sandbox is a testing platform with a fairly simple environment—similar to a folder like Obsidian Vault, containing a small collection of markdown files and the ability to create typed notes, such as TODOs or contacts. Currently, there are only seven tasks, and no password or authentication is required for access.
Regarding the competition itself—creating personal agents—I will prepare a more comprehensive runtime that will simulate working with a larger set of tools. The plan includes something like chat emulation, mailboxes, interaction with remote servers, and even executing destructive commands—in short, a setup that makes it interesting to break virtual systems in this way.
Yours, @llm_under_hood 🤗
P.S. Soon, I will enable leaderboards, participant profiles, access keys, debug mode, and other features from previous competitions.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby