zgsfsec/Full-Chain-Sentinel-System

What it solves

Full Chain Sentinel System (全链哨兵) is an authorized penetration-testing agent built on the pi agent harness. It lets security teams run methodical offensive security workflows—recon, enumeration, exploitation, post-exploitation, and reporting—in one place, with a web console and Docker-based isolation. It is strictly for targets you have written authorization to test, and the project emphasizes consent, audit trails, and minimal intrusion.

How it works

The system uses an AI agent (you provide the model endpoint) that executes commands inside per-project Kali containers. The web console (default Chinese, switchable to English) provides a chat interface, tool-call cards, session trees, and project management. It has three execution modes: approval (default, each bash command must be manually approved), auto-execute, and AI approval (a small gate model decides ALLOW/BLOCK; failures default to allow). The agent chains steps from reconnaissance to reporting, with system prompts restricting scope and requiring evidence. It also integrates asset-discovery platforms (FOFA, 360 Quake, Hunter), a web-search plugin (Tavily), and a custom MCP plugin system. It includes a package-replay and screenshot tool (replayd) for forensic evidence, plus Android and iOS device bridges for mobile testing.

Who it's for

Security professionals and penetration testers who have explicit written authorization to test specific systems. It suits teams working on web, mini-program, and mobile app pentests, especially those who want a unified platform with Docker isolation, audit trails, and controlled AI-driven execution. Users need to be comfortable with Docker, Linux, and consent workflows.

Highlights

• Methodology-driven flow: reconnaissance to report, with emphasis on authorization and evidence.
• Three execution modes: human approval (default), auto-execute, and AI approval with fail-open behavior.
• Per-project Kali containers with host or bridge networking, plus Android/iOS device plugins.
• Asset mapping via FOFA, Quake, Hunter; web research via Tavily; plugin center with MCP servers.
• Repacket-replay and screenshot tool (replayd) for proof-of-exploit evidence, with all actions logged.
• Session-as-audit: every command, output, and model decision recorded in JSONL session trees.