IliasAlmerekov

Aegis Shellguard — AI skill for Claude Code

AI community

Aegis — the shell guard for AI coding agents.

How to install Aegis Shellguard

This entry records only its repository, not the path inside it, so there is no exact command to give. Open IliasAlmerekov/aegis and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Aegis Shellguard does

Aegis — the shell guard for AI coding agents. Destructive commands from Claude Code, Codex, or Cursor pause for your approval, with a snapshot taken first — so even a "yes" is recoverable.

Alternatives in AI

Paritok 4b V1 — Non-destructive compression gateway for AI coding agents 1.4k ★
Hol Guard — Open-source antivirus for AI agents: block risky tools, secret access, prompt injection, malicious packages, M 485 ★
MuseAI Skills — muse.ai (Muse AI) skills and runtime snapshot: 68 skills, workflow guides, connector manifests 261 ★

README

Aegis

**Make AI agents ask first. Keep a way back when they don't.** Your AI agent is one `rm -rf` away from ruining your week. Aegis proxies its shell: safe commands run instantly, destructive ones need your approval — and where a provider applies, a Snapshot captures the working state first, so a "yes" can be restored to what was captured. When no provider applies, Aegis says so instead of implying an undo it cannot perform.

[![version](https://img.shields.io/badge/version-0.6.5-60A5FA?style=flat-square)](CHANGELOG.md) [![platform](https://img.shields.io/badge/platform-Linux%20%7C%20macOS%20%7C%20WSL2-22C55E?style=flat-square)](#how-to-install) [![license](https://img.shields.io/badge/license-MIT-A855F7?style=flat-square)](LICENSE) [![built with](https://img.shields.io/badge/built%20with-Rust-F59E0B?style=flat-square)](Cargo.toml)

![How Aegis works: an AI agent's command is screened by Aegis — safe commands run instantly, dangerous ones wait for human approval](src/assets/aegis.gif)

What is Aegis?

Aegis is a Rust CLI that sits between an AI agent and your real shell. Claude Code, Codex, Cursor — any agent that runs shell commands goes through it. Every command is risk-scored before it executes:

Level	What happens
Safe	Runs immediately unless bounded effect-opaque Required recovery degrades
Warn	Pauses and asks for your approval
Danger	Asks first, then attempts configured Snapshots before execution
Block	Refused outright — no prompt

[!NOTE] Aegis decides by heuristics, not by understanding your command. It is not a privilege boundary: it does not stop code that already runs as you. See [`docs/threat-model.md`](docs/threat-model.md) for the full security model.

The **Sandbox** is a write/network guardrail. It is **not a confidentiality boundary** and **not a privilege boundary**: it does not promise