What is this?
Full Chain Sentinel is an authorized penetration-testing agent system, forked from the 'pi' coding-agent harness, that adds a Chinese-language web console and a per-project containerized job environment to run the full recon-to-report attack workflow.
Full Chain Sentinel is an authorized penetration-testing agent system, forked from the 'pi' coding-agent harness, that adds a Chinese-language web console and a per-project containerized job environment to run the full recon-to-report attack workflow.Is this a library I import, or a standalone app I run?It is a standalone self-hosted application — a Node.js monorepo fork of the 'pi' harness with its own web console and per-project Kali containers — not a library you import into other code.
It is a standalone self-hosted application — a Node.js monorepo fork of the 'pi' harness with its own web console and per-project Kali containers — not a library you import into other code.How do I get it running?Install Node.js >= 22.19.0, run 'npm run build' for the backend packages, and 'npm run build:web' for the console. On first access you set a console password, accept the authorized-use consent, then configure your model endpoint.
Install Node.js >= 22.19.0, run 'npm run build' for the backend packages, and 'npm run build:web' for the console. On first access you set a console password, accept the authorized-use consent, then configure your model endpoint.Do I need Docker to run this?Yes — Docker is a hard dependency; every project gets its own Kali container (sentinel:latest) in which the agent's bash commands are executed via docker exec.
Yes — Docker is a hard dependency; every project gets its own Kali container (sentinel:latest) in which the agent's bash commands are executed via docker exec.Is it safe to just point this at any target?No — this is an offensive tool for targets you already have written authorization to test; unauthorized scanning or exploitation is illegal in China and most jurisdictions. In auto-execute mode the model can run any command, and even AI-gatekeeper mode fails open (allows the command) if its judgment fails or times out.
No — this is an offensive tool for targets you already have written authorization to test; unauthorized scanning or exploitation is illegal in China and most jurisdictions. In auto-execute mode the model can run any command, and even AI-gatekeeper mode fails open (allows the command) if its judgment fails or times out.Is the Docker container a full sandbox?No. Only bash and Android-shell commands actually go through the container; file read/edit/write, MCP tools, frida, packet capture, and APK analysis call the host directly, and the container itself runs as root with no resource limits or seccomp.
No. Only bash and Android-shell commands actually go through the container; file read/edit/write, MCP tools, frida, packet capture, and APK analysis call the host directly, and the container itself runs as root with no resource limits or seccomp.What does the AI gatekeeper (守门员) actually check?A separate, smaller model (e.g. Qwen) reviews each bash command and decides ALLOW or BLOCK before it runs, meant to catch the primary model's risky actions — but it fails open (allows the command) if it can't parse a verdict or times out.
A separate, smaller model (e.g. Qwen) reviews each bash command and decides ALLOW or BLOCK before it runs, meant to catch the primary model's risky actions — but it fails open (allows the command) if it can't parse a verdict or times out.Can I feed it my own vulnerability feed?Yes — the POC subscription feature lets you point it at a zip of Nuclei POCs hosted on your own server (or sync the official Nuclei feed) for testing against known n-day vulnerabilities.
Yes — the POC subscription feature lets you point it at a zip of Nuclei POCs hosted on your own server (or sync the official Nuclei feed) for testing against known n-day vulnerabilities.What does the Android plugin give me?It pairs with a self-made DroidBridge APK installed on a rooted test device to provide a reverse-connected shell, file transfer, screenshots, logcat, tcpdump packet capture, frida dynamic instrumentation, and static APK analysis from the console.
It pairs with a self-made DroidBridge APK installed on a rooted test device to provide a reverse-connected shell, file transfer, screenshots, logcat, tcpdump packet capture, frida dynamic instrumentation, and static APK analysis from the console.What are similar projects?Decepticon, ARTEX, redamon, Pentest-Swarm-AI, ARTEX, hexstrike-ai, strix, dsh-pentest
Decepticon, ARTEX, redamon, Pentest-Swarm-AI, ARTEX, hexstrike-ai, strix, dsh-pentestWhich topics is it in?kali-linux, vulnerability-scanner, offensive-security, red-team, pentesting, penetration-testing, hacking, harness, openai-compatible, cybersecurity, security-tools, china
kali-linux, vulnerability-scanner, offensive-security, red-team, pentesting, penetration-testing, hacking, harness, openai-compatible, cybersecurity, security-tools, china