app icon
Sandbox Agent
0.0.1

ReAct agent strategy with remote sandbox execution. Runs shell commands inside an isolated E2B or Daytona sandbox.

ki3nd/sandbox-agent77 installs

Sandbox Agent

Author: ki3nd
Type: Agent Strategy
Github Repo: https://github.com/ki3nd/sandbox-agent-dify-plugin
Github Issues: https://github.com/ki3nd/sandbox-agent-dify-plugin/issues

A Dify plugin that provides a ReAct agent strategy with remote sandbox execution capabilities.

Features

  • Remote Sandbox Execution: Execute shell commands in isolated sandbox environments (E2B, Daytona)
  • Skills System: Load and use agent skills from configurable paths
  • Conversation History: Persistent conversation memory with configurable turn limits
  • Function Calling: Support for parallel tool execution with LLM function calling
  • Built-in Tools: for shell execution (more coming: , )

Configuration

Parameters

ParameterTypeRequiredDefaultDescription
model-selectorYes-LLM model for reasoning
array[tools]No[]Additional Dify tools
stringYes-User query / input
stringYes-System prompt / instruction for the agent
stringYes-ID of the sandbox to connect
stringNo-YAML configuration for sandbox
secretYes-API key for sandbox provider
numberNo20Max ReAct loop iterations
numberNo10Number of conversation turns to remember

Sandbox Configuration (YAML)

Supported Sandbox Providers

E2B

Daytona

Skills System

Skills are stored in directories with a file containing YAML frontmatter:

Example :

Usage

  1. Create a sandbox in your provider (E2B or Daytona)
  2. Configure the plugin with your sandbox ID and API key
  3. Add the Sandbox Agent strategy to your Dify workflow
  4. The agent will execute commands in the sandbox and reason about results

License

MIT License

CATEGORY
Agent Strategy
VERSION
0.0.1
ki3nd·05/11/2026 03:02 AM
REQUIREMENTS
LLM invocation
Tool invocation
Maximum memory
256MB
Maximum storage
1MB