A pentest agent with a conscience (and an approval queue)

What it does

Full Chain Sentinel System (全链哨兵) is an authorized-pentest agent built as a fork of the MIT-licensed pi coding agent. It gives you a Chinese/English web console that manages Web, mini-program, and mobile testing: streaming chat, session forking, sub-agents, batch task orchestration, asset and vulnerability libraries, and Markdown reporting. Each project spins up a Kali work container behind a persistent gateway, so switching proxies doesn’t mean rebuilding your environment.

The interesting bit

The boring part is the value: it treats approval and audit as first-class features — human approval, auto-execution, or AI approval modes, with full call logs and session records. The agent can run arbitrary commands in auto mode, so the paper trail isn’t decoration. Mobile testing is handled via device bridges (DroidBridge for Android, SentinelBridge for iOS) supporting device reverse-connection, traffic capture, and dynamic analysis.

Key highlights

• Kali work containers with persistent gateways; SOCKS5/HTTP, WireGuard, OpenVPN, and hot-swappable exits
• Asset library, vulnerability hub with CVSS, evidence retention, and Markdown reports
• OpenAPI business modeling plus Nuclei POC subscription, search, and citation
• Intelligence integrations: FOFA, Quake, Hunter, Tavily, MCP, and local/remote plugin hosts
• Request replay, history search, screenshot evidence, and packaged file downloads

Caveats

• Explicitly for targets with written authorization only; the README repeats this and points to a security-boundary doc
• It’s a fork of pi, not the upstream project — worth knowing if you track the upstream codebase
• One bundled DOOM example extension is GPLv2, an odd license island in an otherwise MIT repo

Verdict

Worth a look if you deliver authorized pentest or app security work and want an agent that keeps receipts. Not for anyone hoping to point it at targets they don’t own — the project is blunt about that.