oh-my-agentic-coder CLI
A Go CLI that runs a coding agent in an OS sandbox with skills and secrets behind a bridge and isolated files and network.
- 55 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The utility manages a sandbox and keeps skill secrets in the OS keystore.
- Wrong isolation setup would give the agent more access than expected.
Install
- Works in
Manual install
Read the quick start at tng.github.io/oh-my-agentic-coder, install dependencies for your OS, check `omac doctor` and start the agent with `omac start`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The utility manages a sandbox and keeps skill secrets in the OS keystore.
- Wrong isolation setup would give the agent more access than expected.
This is third-party code. Review the repository files before installing.
What it does
omac, or oh-my-agentic-coder, runs an agentic coding harness inside an operating system sandbox. OpenCode, Claude Code, Codex, Copilot, Pi and CodeWhale are supported. Isolation uses OS primitives: Seatbelt on macOS and bubblewrap with Landlock on Linux. The agent uses skills and secrets through a bridge and has no direct access to them, while the file system and network are restricted. The author states that omac is not a model, a marketplace or an agent: it isolates, bridges and audits, and skills and onboarding come from elsewhere. Skill secrets live in the OS keystore, and a confirmation dialog appears when the agent reaches an unknown host. The `omac doctor` command checks that the sandbox works.
Who it is for. Teams and developers who run coding agents and want to restrict files, network and secret access at the OS level.
Good fit when
- An agent works with a project but not with the rest of the home directory.
- Skill tokens live in the system keystore rather than in project files.
- A confirmation dialog is needed when the agent reaches a new host.
- Different agents, OpenCode and Claude Code, run under one isolation rule set.
Not a fit when
- Work happens in Windows without WSL2.
- You need the agent or a model itself rather than a wrapper.
- Bubblewrap or a secret store cannot be installed on the system.
Example request
Run omac doctor and show whether the sandbox is ready, then start a Claude Code session inside it.Limitations
It supports macOS 15 and Ubuntu 24.04, with Windows only through WSL2. Claude Code, OpenCode and Copilot are fully supported while Codex, Pi and CodeWhale are experimental. You need a sandbox, a secret store, a network confirmation dialog and `jq`: on Linux install bubblewrap, zenity, libsecret and jq, and Ubuntu 23.10 or newer may need an AppArmor exception for `bwrap`. Landlock network filtering in WSL2 needs Linux kernel 6.7 or newer. Keep WSL2 repositories on the native file system.
How to disable. Remove the `omac` binary and its configuration, and delete skill secrets from the OS keystore.
FAQ
What does omac do?
It runs an agentic harness inside an OS sandbox where file and network access is restricted and skills and secrets are reachable through a bridge.
Which agents are supported?
OpenCode (CLI and Desktop), Claude Code, Copilot, Codex, Pi and CodeWhale. The last three are experimental.
What does Linux need?
Install dependencies for Debian and Ubuntu with `sudo apt-get install -y bubblewrap zenity libnotify-bin libsecret-1-0 jq`. Swap `zenity` for `kdialog` on KDE.
Does it work on Windows?
Only through WSL2 with Ubuntu 24.04 and kernel 6.7 or newer for network filtering.
How do I check the sandbox?
Run `omac doctor`: on failure it says what is wrong, for example about AppArmor.
Discussion
No questions or reviews yet.
Related
CLI toolsThe safehouse command runs Claude Code, Codex and other agents in a macOS sandbox: files are closed by default and you open folders with flags
Claude Code, Codex, OpenCode and pi run in a hardened container: project files only, allowlisted network, keys held by a proxy.
cli sandbox for ai agents without containers or daemons: runs Claude Code, Codex, or OpenCode with restricted access to files, network, and credentials
An isolated box for coding agents on your own machine: access only to chosen projects and credentials, secured registry
A local dashboard for Claude Code, Codex CLI and OpenCode: what the agent loads, its token cost and an MCP server security scan
AgentGuard checks a coding agent's actions before they run and scans third-party skills for secrets, backdoors and injections.
README in short
The README briefly describes omac, the table of supported agents and OSs, documentation and bug reporting. Install and prerequisites are covered in the documentation.