Secure Script Runner (Zero‑Exposure Sandbox)

Overview

Secure Script Runner is a documentation skill that teaches how to execute scripts with zero plaintext exposure using MGC Blackbox.

This skill enables:

• Store scripts encrypted in local MGC
• Execute scripts locally (AI sees results only)
• Three execution modes: MCP, API, WebUI
• Scripts can call internal MGC credentials
• Script sealing for cross‑node delegation

This skill contains no executable code and is safe for automatic approval.

What This Skill Enables

After reading this documentation, an AI agent will understand how to:

• Store scripts securely in MGC Blackbox
• Execute scripts via MCP (AI), API (script), or WebUI (human)
• Call MGC internal credentials from scripts
• Seal scripts using node public key
• Build zero‑exposure workflows

Prerequisites

1. Install MGC Blackbox: pip install mgc-blackbox
2. Start MGC service: mgc (runs at http://127.0.0.1:57219)
3. MCP tools available: Use mgc_save, mgc_get, mgc_seal
4. Token file: ~/.mgc/database/mgc_black_box/.mgc_token

Important: For AI agents, use MCP tools. CLI may have port conflicts in some environments.

Zero‑Exposure Execution

Core Concept

Script (plaintext) → MGC Encryption → Encrypted Storage
                                              ↓
                              Local Execution (MGC)
                                              ↓
                              AI receives result only

AI executes but never sees script plaintext.

Three Execution Modes

Storing Scripts

Step 1: Prepare Script

Store a script with execution metadata:

# Via MCP tool
mgc_save(
    info_type="script",
    info_owner="my_script",
    ext01="python",           # Startup command
    ext02="script.py arg1",  # Default runtime args
    content="print('Hello from zero‑exposure!')"
)

Parameters

Executing Scripts

Mode 1: Via MCP (AI)

# Execute via MCP tool
result = mgc_get(
    info_type="script",
    info_owner="my_script",
    action="run"
)
# AI receives execution result only

Mode 2: Via REST API (Script)

curl -X POST http://127.0.0.1:57219/api/mgc/sensitive/get \
  -H "Content-Type: application/json" \
  -H "X-MGC-Token: $(cat ~/.mgc/database/mgc_black_box/.mgc_token)" \
  -d '{
    "info_type": "script",
    "info_owner": "my_script",
    "action": "run"
  }'

Mode 3: Via WebUI (Human)

1. Open WebUI: http://127.0.0.1:57218
2. Navigate to Get page
3. Find your script
4. Click "Run" button

Calling MGC Credentials

Scripts can call MGC internal credentials using the internal API:

# Example: Call MGC credential from script
import urllib.request
import json

def get_mgc_credential(info_type, info_owner):
    data = json.dumps({
        "info_type": info_type,
        "info_owner": info_owner
    }).encode("utf-8")

    req = urllib.request.Request(
        "http://127.0.0.1:57219/api/mgc/sensitive/get",
        data=data,
        headers={
            "Content-Type": "application/json",
            "X-MGC-Token": open("/path/to/token").read()
        },
        method="POST"
    )

    with urllib.request.urlopen(req) as resp:
        return json.loads(resp.read().decode())["data"]

Note: Credentials are retrieved locally, script executes locally, AI never sees plaintext.

Script Sealing (Advanced)

For cross‑node delegation, scripts can be sealed using the node's public key:

Step 1: Get Node Public Key

# Via MCP tool
node_pub = mgc_get(
    info_type="__NODE_PUB__",
    info_owner="__NODE_PUB__"
)

Step 2: Seal Script

# Via MCP tool
sealed = mgc_seal(
    info_type="script",
    info_owner="my_script",
    ext04=node_pub  # Target node public key
)

Step 3: Store Sealed Script

# Store sealed version
mgc_save(
    info_type="script",
    info_owner="my_script_sealed",
    ext01="python",
    content=sealed
)

Sealed scripts are encrypted and can only be executed by the target node.

MCP Tools Reference

mgc_save

Arguments:

{
  "info_type": "script",
  "info_owner": "unique identifier",
  "ext01": "startup command (python, node, etc.)",
  "ext02": "default runtime arguments",
  "content": "script plaintext"
}

mgc_get

Arguments:

{
  "info_type": "script",
  "info_owner": "script identifier",
  "action": "get | run"
}

Returns: Script content or execution result

mgc_seal

Arguments:

{
  "info_type": "script",
  "info_owner": "script identifier",
  "ext04": "target node RSA public key"
}

Returns: Sealed script (encrypted with target node key)

Security Notes

1. Zero‑exposure: Script executes locally, AI receives result only
2. Encrypted storage: All scripts encrypted at rest
3. No plaintext leakage: AI never sees script content
4. Script sealing: Cross‑node scripts stay encrypted

Links

• Main Repository: https://github.com/zkeviny/MGC-Blackbox
• Issues: https://github.com/zkeviny/MGC-Blackbox/issues
• Contact: mirgincipher@outlook.com