yean-sec/strikeagent_atkbrain-flash
Browser automationThis page was translated automatically. Technical details may differ from the original. View the original
The AI penetration testing platform developed by the Night An team covers Red Team Placement, SRC, and CTF, and is especially outstanding in the Red Team field
Overview
This project was developed by the Ye An team to explore AI penetration capabilities, aiming to create something with its own ideas rather than AI being randomly made The project as a whole seems simple, but it has undergone multiple rounds of real-world testing, with the team pouring a lot of effort. The details are countless: just the fact that "not missing but gaining a larger attack surface" alone made the team painstakingly (Flash version). Additionally, to prevent AI-generated vulnerabilities from being exaggerated or falsely positive, the team specially designed red team second-rate and secondary verification, making the vulnerability "what you see is what you get" and "ready to use" — these small details. The team's proprietary Pro version has already been implemented in dozens of projects and over a thousand real licensing environments on external networks, with all efforts focused solely on external network operations. Focusing on one point to achieve the pinnacle Console scheduling for Hunter Face; Attack map driving self-loop; Overseeing only coming up with solutions at turn boundaries; Distilling transferable techniques into memory and feeding them back into the next round. Ranked 3rd on the official Cybench certification chart (deepseek-v4-flash, 84.13 / 100). Give the entire section below along with the source code to any AI capable of running native commands. It should be deployed as originally written, without changing skill binding methods or writing paths to other people's machines. It is recommended to run it on (the execution layer will call the existing local penetration tools). Other Debian/Ubuntu consoles can also be built, but the tools may not be fully available. The model key can be set with environment variables. The installation script will snap to backend/data/atkbrain-claude.env (600 access, do not submit to git):
Overview (machine translated)
StrikeAgent × Night Safety Team SEC
StrikeAgent_AtkBrain-Flash
This project was developed by the Ye An team to explore AI penetration capabilities, aiming to create something with its own ideas rather than AI being randomly made
The project as a whole seems simple, but it has undergone multiple rounds of real-world testing, with the team pouring a lot of effort. The details are countless: just the fact that “not missing but gaining a larger attack surface” alone made the team painstakingly (Flash version). Additionally, to prevent AI-generated vulnerabilities from being exaggerated or falsely positive, the team specially designed red team second-rate and secondary verification, making the vulnerability “what you see is what you get” and “ready to use” — these small details.
The team’s proprietary Pro version has already been implemented in dozens of projects and over a thousand real licensing environments on external networks, with all efforts focused solely on external network operations. Focusing on one point to achieve the pinnacle
Architecture
Console scheduling of Hunter’s Face; Attack chart driver self-looping. Servants only speak after the full round is asked by the Master, only speaking when stuck or in cycle; Manual commands in the dialog box immediately interrupt the current round. Finish and distill the technique into memory. Runtime is Pi; which model to use is determined in the JSON in the settings. Agent tests default to direct direct connection to real IPs; Only when the proxy pool imported in settings is opened does the proxy process proceed.
Product Page Display
Product Overview:
Home and New Tasks: Agent testing and CTF entry points. Objectives can be entered individually or batch created by importing txt/csv files with additional information. CTF can also be selected as Web, PWN, Inverse, Miscellaneous, or Cryptography, and you can upload problem attachments.
Project console: attack chart, timeline, vulnerabilities, dialogue.
Vulnerability Database:
Delivery Report
Report template
Open HTML
Rankings
Tsecbench v1 No.1 (StrikeAgent_AtkBrain-Flash, 97.89 / 100). Tencent Security Yunding Lab’s Intelligent Offense and Defense Agent review.
Efforts for Platform Security
Unauthorized access does not yield login pages, interfaces, or plaintext passwords.
- Random entry: 8-bit path writes only to the local machine
backend/data/security_entry(0600); Webpage/API not visible.python -m atkbrain.panelPrint URL. - No entry, no subdirectory: Bare
:2334/is the introduction page;/login,/api,/.envetc. are all 404. - Anti-capture packets / Replay / Bursting: Browser RSA-OAEP encrypts before POST; One-time ticket (120s); Locked by username/IP.
- Password: Database Argon2id; This machine
admin_bootstrap.secret(0600) is for people with shells,panelcan be printed. - Session: Cookie
HttpOnly; None?token=; Swagger is closed. Default HTTPS:2334.
Environment and Installation
This page was translated automatically. Technical details may differ from the original.
Recommended Tools
Try a different keyword or remove a filter.
Install
npx skillfish add yean-sec/strikeagent_atkbrain-flash