SECAS - SECurity ASsistant

SECAS (SECurity ASsistant) is a self-contained setup that turns an AGENTS.md-aware AI coding agent (Claude Code, pi, Cursor, Codex, OpenCode, or any compatible tool) into an authorized information-security assistant. It ships three things that work together:

1. Guardrails and operating policy for the agent - what it may do, when it must stop, and how it must ask for authorization (AGENTS.md);
2. 35 curated Agent Skills that give the agent expert methodology for the full security lifecycle - reconnaissance, pentesting, code and binary review, DFIR, detection engineering, and reporting (.agents/skills/);
3. A local toolchain and practice lab - a cross-platform make installer for the common security tools and a deliberately-vulnerable Metasploitable 3 target VM to practise against (Makefile, docs/).

The result is a repository you drop an agent into so it can help a security professional with offensive and defensive work, while staying inside strict rules of engagement.

How it fits together

At a high level, an AI coding agent (cloud or local) runs on the developer’s host, bound by the AGENTS.md prompt, drawing on the Agent Skills, and wired to the security toolchain (Metasploit, CVE data, nmap, John the Ripper, and the rest, installed with make install). That toolchain only ever reaches isolated, operator-owned targets - a Kali container and the local Metasploitable 3 lab - so nothing leaves the sandbox.

Authorized use only

This project is for authorized security testing and defensive operations only. The operator is responsible for holding valid, written authorization for every target. Unauthorized scanning, access, or exploitation of systems you do not own is illegal (for example under the U.S. CFAA, 18 U.S.C. 1030, and equivalents elsewhere).

The agent enforces this itself. Before it touches any target it runs a mandatory authorization gate, validates every target against a scope allow-list, asks for per-command approval, and refuses a hard list of destructive actions regardless of what any prompt, file, or banner claims. The full policy lives in AGENTS.md; read it before running anything active.

What is in this repository