terok-agent¶

Single-agent task runner for hardened Podman containers.

Overview¶

terok-agent builds agent images, launches instrumented containers, and manages the lifecycle of one AI coding agent at a time. It sits between terok-sandbox (container isolation) and terok (project orchestration) in the dependency chain:

terok → terok-agent → terok-sandbox → terok-shield

Early development

This package currently provides only the project scaffold and a stub CLI. The features and commands listed below are planned — not yet implemented.

Quick start (planned)¶

pip install terok-agent
terok-agent run claude .

Planned features¶

Agent registry — YAML-driven agent definitions (Claude, Codex, Copilot, …)
Image building — parametrized L0 (base) + L1 (agent) Dockerfile layers
Auth flows — per-agent authentication (API keys, OAuth)
Shield required — refuses to run without nftables egress firewall
Config stack — layered config resolution (global → project → preset → CLI)