Sandy — DevOps skill for Claude Code
Sandboxed TypeScript runtime for AI coding agents to query AWS — full SDK access with in-sandbox aggregation, credentials flow via IMDS and never enter the VM.
How to install Sandy
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open jamestelfer/sandy and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Sandy does
Sandboxed TypeScript runtime for AI coding agents to query AWS — full SDK access with in-sandbox aggregation, credentials flow via IMDS and never enter the VM. Ships as a Claude Code plugin (MCP) and a standalone CLI, backed by Shuru microVMs or Docker.
Alternatives in DevOps
- Skypilot — Run, manage, and scale AI workloads on any AI infrastructure 9.7k ★
- Azure Functions TypeScript — Azure Functions for TypeScript 1.8k ★
- GPT Telegram Worker — A multi-model AI Telegram bot powered by Cloudflare Workers, supporting various APIs including OpenAI, Claude 268 ★
README
Sandy
**Sandy runs TypeScript AWS queries inside a disposable sandbox, giving AI coding agents full SDK access and cross-account aggregation with no host credentials exposed.**
Sandy is for AI coding agents — Claude Code and peers — running against multi-account AWS estates, and the humans who drive them. Agents describe an investigation in natural language. Sandy executes the generated TypeScript in a fresh microVM or container. The script uses the full AWS SDK to gather and collate results, then returns only what the agent needs. The sandbox and IMDS flow exist because agents should not hold host-level credentials or reach AWS directly.
flowchart LR
subgraph AgentBox["Agent sandbox"]
A["Coding agent"]
end
subgraph Host["Host (outside agent sandbox)"]
direction TB
CLI["sandy run
(CLI)"]
MCP["sandy mcp
(MCP server)"]
IMDS["imds-broker
(MCP)"]
end
subgraph Sandbox["Ephemeral sandbox
(Shuru microVM or Docker container)"]
N["Node.js +
AWS SDK v3"]
end
AWS((("AWS APIs")))
A -- "Bash: sandy run" --> CLI
A -- "MCP: sandy_run" --> MCP
A -- "MCP: start_server" --> IMDS
CLI -- "spawn, mount script" --> N
MCP -- "spawn, mount script" --> N
N -- "IMDS: GET credentials" --> IMDS
N -- "HTTPS to *.amazonaws.com" --> AWS
classDef consumer fill:#e6f3ff,stroke:#2b6cb0,color:#1a365d
classDef core fill:#fefcbf,stroke:#b7791f,color:#5f370e
classDef sandboxed fill:#e9f7ef,stroke:#276749,color:#22543d
classDef external fill:#fce4ec,stroke:#b83280,color:#702459
class A consumer
class CLI,MCP,IMDS core
class N sandboxed
class AWS external
How to use it
- As an MCP server —
sandy mcp, registered automatically by thesandy-mcpClaude Code plugin. Exposes thesandy_image,sandy_check,sandy_run,sandy_create_session,sandy_resume_session, andprimetools, plus embedded `sandy://skills/mcp/.
Related Skills
Porta
Sandboxed runtime for AI agents. WASM isolation + OS-level sandbox. Run Claude Code, Python, or any command wi
Spec Kit Plus
A practical fork of github/spec-kit with patterns & templates for building scalable multi-agent AI systems. Sh
Sandbox Shell
macOS Seatbelt sandbox CLI for developers. Protect credentials (SSH, AWS, GPG) from malicious npm packages, su
Multi Modal Agent TS
TypeScript multimodal AI agent: GPT-4o / Claude / Gemini + Whisper + Ollama (LLaVA). REST API, streaming, Dock
Claude In The Box
Use Anthropic Claude Agent SDK in a Cloudflare Sandbox
Antigravity Sandbox
Lightweight CLI manager for Docker SBX and AI coding agents (Omp, Agy, Opencode, Prime, Codex, Claude, etc). B
Related Agents
AWS Credential Exposure
Assess AWS root credentials, IAM user MFA state and long-lived access keys against what each one reaches.
Tsk - AI Agent Task Manager And Sandbox
by dtormoen - A Rust CLI tool that lets you delegate development tasks to AI agents running in sandboxed Docke
Query Reviewer
Reviews Active Record queries for N+1, missing indexes, unscoped reads, aggregation done in Ruby instead of th