Agentacl — AI skill for Claude Code
Access control and sandboxing for AI coding agents on macOS.
How to install Agentacl
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open chaitanya-sistla/agentacl and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Agentacl does
Access control and sandboxing for AI coding agents on macOS. Stop Claude Code, Codex, Gemini CLI, Copilot CLI and OpenCode from reading your secrets, with a kernel sandbox, an egress firewall, live approvals and an audit log.
Alternatives in AI
- System — You are YouClaw, a helpful AI assistant 684 ★
- TokenEater — 179 Native macOS menu bar app for monitoring Claude AI usage limits and watching coding sessions live 196 ★
- Easel — Easel is a Codex and Claude macOS workspace for AI-assisted product design and frontend iteration 165 ★
README
AgentACL: access control and sandboxing for AI coding agents on macOS
**Stop Claude Code, Codex, Gemini CLI, Copilot CLI and OpenCode from reading your secrets. AgentACL enforces it in the macOS kernel, not in the prompt.**
[](https://github.com/chaitanya-sistla/agentacl/actions/workflows/ci.yml) [](https://github.com/chaitanya-sistla/agentacl/releases/latest) [](LICENSE) 
AI coding agents run shell commands **as you**. Anything you can read, they can read: SSH keys, `.env` files, AWS and GCP credentials, Terraform state, browser cookies. They can also send it anywhere on the internet.
AgentACL is an open-source **security layer for AI coding agents**. It runs each agent inside a macOS kernel sandbox (Seatbelt), with a network egress firewall and an audit log. The agent keeps full access to your project and loses access to everything it shouldn't touch. You decide what that is, in a local web console or a YAML policy.

Why AgentACL
| Without AgentACL | With AgentACL | |
|---|---|---|
Agent reads ~/.ssh, ~/.aws, .env, browser cookies |
✅ succeeds | ⛔ Operation not permitted (kernel) |
| Agent uploads data to an unknown site | ✅ succeeds | ⛔ refused by the egress proxy, or held until you approve |
Agent plants a git hook or edits ~/.zshrc |
✅ succeeds | ⛔ blocked |
Subprocesses (sh, python, curl) |
inherit everything | inherit the same sandbox |
| You know what the agent tried | no record | every attempt logged: agent, process chain, rule |
| An agent runs unprotected | invisible | flagged in the console |
Ins
Related Skills
Secure Agent
🔐 Egress inspection & secret-leak firewall for local AI agents — see what your agents send, catch secrets bef
Agentic Access Control
Read-only, local-first dashboard showing what your AI agents (Claude Code, Gemini CLI, MCP servers) are actual
Skilljail
Manifest-declared, kernel-enforced least privilege for AI agent skills (Claude Code SKILL.md). Seatbelt/Landlo
AI Agent Sandbox
A secure bubblewrap-based sandboxing solution for running Claude Code with strict filesystem isolation.
Apohara Agentguard
Anti-bypass command-safety hook, local seccomp+Landlock sandbox, and deterministic prompt-injection firewall f
Agentwall
Zero-Dependency Runtime Firewall, Secret DLP & Process Sandbox for AI Agents and MCP Servers
Related Agents
Developer Overview
code-container (container) creates isolated Docker environments for AI coding harnesses (Claude Code, OpenCode
Consult Agent
Execute cross-tool AI consultations via Task spawning. Use when agents or workflows need a second opinion from
Qp Security Reviewer
Read-only reviewer for the quality-pass security & data-integrity dimension — authz gaps, tenant/owner scoping