Agentic Sandbox — AI skill for Claude Code
Run AI agents(Claude Code, Codex, OpenClaw, etc.) with Full Power in Isolated Local Sandbox.
How to install Agentic Sandbox
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open peter741/Agentic-sandbox and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Agentic Sandbox does
Run AI agents(Claude Code, Codex, OpenClaw, etc.) with Full Power in Isolated Local Sandbox. Zero cost, Zero risk to your host.
Alternatives in AI
- Manifest — Real-time cost observability for OpenClaw agents — track tokens, costs, messages, and model usage 4.1k ★
- Rb Setup — First-time setup 1.3k ★
- Power BI Agentic Development — Power BI AI skills and Power BI agents for Claude Code and GitHub Copilot: a plugin marketplace of Power BI sk 879 ★
README
Agentic Sandbox
An independent, early-stage open-source project maintained at [peter741/Agentic-sandbox](https://github.com/peter741/Agentic-sandbox), derived from Agents Sandbox, for running coding agents in local Docker sandboxes.
**Upstream provides the sandbox engine and agent integrations. This derivative adds setup diagnostics through `agbox doctor` and a metadata-only workspace exposure review through `agbox preflight`.** See [provenance](docs/provenance.md) and [CHANGELOG.md](CHANGELOG.md).
Diagnose setup before launching an agent
Check missing Docker, an unreachable engine, and daemon setup in one command:
agbox doctor
agbox doctor --workspace ./my-project
agbox doctor --json --timeout 3s
Doctor checks Linux/macOS host support, Docker CLI availability, Docker Engine connectivity and Linux-container mode, daemon Ping, version differences, and optionally the workspace path. It does not start services, create containers, or run an agent. See [doctor documentation](docs/doctor.md).
Review a workspace before giving it to an agent
agbox preflight --workspace ./my-project
agbox preflight --workspace ./my-project --json
Preflight flags credential-looking filenames, private-key names, credential folders, and symlinks without reading file contents or changing files. It works without Docker. Incomplete scans and findings exit 1 for review. This is an advisory filename scan, not a security guarantee or an automatic launch gate. See [preflight documentation](docs/preflight.md).
Build this derivative
Requirements: Go 1.25 or newer. Docker and a configured agboxd daemon are needed for live sandbox use, but not for help or unit/mock tests.
git clone https://github.com/peter741/Agentic-sandbox.git
cd Agentic-sandbox
mkdir -p .build
go build -buildvcs=false -o .build/agbox ./cmd/agbox
go build -buildvcs=false -o .build/agboxd ./cmd/agboxd
.build/agbox doctor --help
.build/agbox doctor --json
Without Docker and the d
Related Skills
Agents Sandbox
Run AI agents(Claude Code, Codex, OpenClaw, etc.) with Full Power in Isolated Local Sandbox. Zero cost, Zero r
Enclave
Sandbox for running AI coding agents autonomously: isolated, network-restricted, host-safe
Ollama Cc Plugin
Use local/cloud ollama models from Claude Code as a second engine: read-only consult/review, agentic worktree-
Xiaoyaoclaw Usage Report
OpenClaw usage & performance reporting - parse session JSONL for task duration, tool/skill/model usage, token
Local LLM Setup
Runbook and PowerShell automation to host two isolated local LLMs (KYC analyst and KB-chatbot reasoning) on on
Axocoatl
Run coding agents unattended. Agents write the code. Axocoatl writes the receipt. Claude Code, Codex or a loca
Related Agents
Omhc Tester
Hands-on tester for omhc. Use after a change is implemented (and ideally reviewed) to exercise it for real in
Delforge Fullstack Engineer
Use for general full-stack software development tasks — turning requirements into working, tested applications
Sandbox Runner
Runs untrusted, unfamiliar or agent-written code inside a sealed kosa8 microVM and reports the results, so not