More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    sandboxed agent gateway

    0

    Secure agent workspace gateway providing sandboxed file operations and restricted terminal command execution.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    sandboxed agent gateway

    Example session with this skill installed

    Create a new python script called process_data.py that prints 'Data Processed' and then run it for me.

    • Read your context and instructions
    • Compiled the sandboxed agent gateway

    I will check the workspace, create the script, and then run it after your approval.

    1. Checked directory: process_data.py does not exist.
    2. Created file: process_data.py.
    3. Requested permission to run python3 process_data.py.
    4. Execution successful.

    Execution Summary

    TaskDetailResult
    Check Existencelist_directoryNo conflict found
    Write Fileprocess_data.pyFile created successfully
    Commandpython3 process_data.pyExit Code: 0

    Terminal Output

    Data Processed
    

    Next steps

    • To add logic to this script, tell me what data you want to process and I will use mode="append".
    • If you need to install dependencies, I can run pip install after you confirm the package list.
    • Use list_directory to see other files in your current workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    AI agents with unrestricted shell access are a security liability. This skill provides a controlled gateway for agents to interact with the local file system and terminal through a bundled MCP server. It enforces a workspace jail and a strict command allowlist to ensure the agent only operates where it is supposed to.

    What it does

    • Sandboxed File Access executes file operations only within a defined workspace directory using relative paths.
    • Controlled Terminal Execution runs shell commands against a pre-defined allowlist to prevent arbitrary code execution.
    • Safety Checks requires user confirmation for destructive commands like mv, cp, pip, or git changes.
    • Extension Validation prevents ambiguous file creation by requiring explicit file types before writing to disk.
    • Resource Limits enforces CPU, memory, and time constraints on processes to prevent resource exhaustion.

    How it works

    1. Connectivity Check verifies that the read_file, write_file, list_directory, and run_command tools are active.
    2. Directory Inspection uses list_directory to prevent accidental overwrites and confirm the workspace state.
    3. Execution Pipeline handles file creation or script execution, requesting user approval for any command that modifies the environment.
    4. Error Handling parses ERROR: prefixes and output truncation to recover from failures without looping.

    Frameworks & tools

    This skill uses a custom MCP (Model Context Protocol) server bundled in the server/ directory. It requires Python 3 for the server and includes scripts for Docker deployment and agent configuration.

    Why this beats prompting it yourself

    Standard system prompts can be bypassed or ignored by models during long sessions. This skill uses a tool-based architecture that physically restricts the agent's reach to a specific jail, making it impossible for the agent to access files outside the workspace even if it tries.

    Use cases

    • Scaffold new project structures safely within a specific folder.
    • Run Python scripts and capture output without giving full shell access.
    • Review and edit local configuration files with mandatory overwrite protection.
    • Manage Git status and logs through a restricted terminal interface.

    Known limitations

    Does not support absolute paths or parent directory traversal. No delete tool is provided. Complex shell syntax including pipes, redirects, and command chaining is not supported.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    No delete tool provided. Cannot use absolute paths or parent directory traversal. Does not support shell pipes, redirects, or command chaining. Output is truncated at 4,000 characters.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 13 days ago

    Software
    PythonGitDocker

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Environment Variables

    File Scopes

    workspace/**
    Listed13 days ago
    Updated10 days ago

    What's inside

    Frequently Asked Questions

    This skill solves the 'jailbreak' risk by enforcing a hard workspace jail and a command allowlist. Unlike prompts which can be ignored, this uses a physical tool-based barrier that prevents the agent from accessing files or executing commands outside of your specified parameters.

    The skill is built on the Model Context Protocol (MCP), making it compatible with any LLM or agent framework that supports MCP servers, including Claude Desktop, LangChain, and several popular open-source autonomous agent wrappers.

    No, the skill is designed to prevent 'prompt injection' escapes by refusing absolute paths and parent directory traversal (..). It strictly interprets all file operations as relative to the designated workspace folder.

    The purchase includes the core MCP server files (Python), a pre-configured Dockerfile for containerized deployment, integration scripts for your agent config, and a README for setup.

    The gateway includes a built-in safety check for destructive or sensitive commands like 'pip' or 'git'. Before these are executed, the server pauses the operation and requires manual user confirmation to proceed.

    To maintain a high security posture, this skill explicitly excludes a delete tool and does not support complex shell syntax like pipes (|) or redirects (>), preventing agents from chaining commands to bypass the allowlist.

    New in LLM Ops