claude-agent-wrap

A Docker-based wrapper for the Claude Code CLI that isolates the agent in containers, keeps API credentials out of the agent process (in the default provider), and lets each project customize its environment with a simple Dockerfile.agent.

It packages Claude Code into a reproducible container image and exposes a single bash function — agent — whose first argument is a verb (run, rebuild, create, stats, logs, update) that selects the operation. Volume mounts, credentials, and per-project image customization are handled automatically. Model traffic is routed through a provider plugin — all shipped providers use a LiteLLM sidecar. See Providers for available options.

Documentation

Category	Docs
Getting started	Requirements + Setup
Features	Docker Sandboxing · Telegram Notifications · WSLg Clipboard
Providers	Provider Setup
Configuration	Configuration
Reference	Volume Mounts · Shell Commands · Container Environment
Changelog	Release history

Quick Start

Source the wrapper in your shell (add it to ~/.bashrc to make it permanent):

source /path/to/claude-agent-wrap/agent-wrap.bashrc

Build the base image once:

agent rebuild

From any project directory, run:

agent run [claude-code-args...]

See the Getting Started guide for full setup instructions and Providers for your model-routing options.

Project Layout

.
├── .agent-launches/      # Project registry and pricing cache (git-ignored)
├── .claude_config/       # Global Claude config (git-ignored)
├── agent_wrap/           # Python orchestration (commands, providers, config)
├── docs/                 # Documentation (linked from this file)
├── logs_page/            # Static web viewer served by `agent logs`
├── ops/                  # Base image, validator, status line, hooks, clipboard shim
├── scripts/              # Repo tooling (e.g. markdown-link validator)
└── agent-wrap.bashrc     # Shell function: `agent <verb>`