A secure multi‑agent execution engine for zero‑exposure automation. Store, execute, and delegate scripts or credentials locally with full encryption, sealed distribution, and argparse‑driven structured execution.
编程
Seal and License Workflows or Skill Packages (Anti‑Piracy, Node‑Authorized Delivery)
试用Seal skill packages and workflow folders so they execute only on authorized nodes — no read, no modify, no resell. 100% local, single‑node bound, zero‑exposure IP protection powered by MGC Blackbox 1.5.0+.
它能做什么
Seal skill packages and workflow folders so they execute only on authorized nodes — no read, no modify, no resell. 100% local, single‑node bound, zero‑exposure IP protection powered by MGC Blackbox 1.5.0+.
技能文档
1. Overview
Seal and License Workflows or Skill Packages is a teaching skill that explains how to use MGC Blackbox 1.5.0+ to ship a skill package or workflow folder to a target node — without giving away the source.
The target node can execute your work. It cannot read, modify, or resell the plaintext.
2. The Three Promises
Once your skill package is sealed and delivered to an authorized node:
- No read — the plaintext source never leaves your device in cleartext
- No modify — the capsule cannot be edited without breaking the AES signature
- No resell — the AES key is bound to the target node's RSA public key, so the capsule cannot be re‑sealed for a third party
3. Local‑First Guarantee
- All encryption happens locally on the author's device using MGC Blackbox.
- All decryption happens locally on the consumer's device.
- No data is uploaded to any cloud. The capsule can be hand‑delivered by email, USB, or any channel of your choice.
- MGC itself runs as a local service (
mgconhttp://127.0.0.1:57219); it has no external endpoint.
This makes the workflow suitable for air‑gapped, regulated, or privacy‑sensitive environments.
4. Prerequisites
- MGC Blackbox 1.5.0+ (required for
mgc_save_file,mgc_seal_package,mgc_package) - Python 3.10+ (Python 3.12 on macOS Apple Silicon is not supported)
- Install:
pip install mgc-blackbox - Start:
mgc(API athttp://127.0.0.1:57219, WebUI athttp://127.0.0.1:57218) - An MCP‑compatible agent runtime on both author and consumer side
5. Author Side — Seal a Skill Package or Workflow
Three steps to ship your work as an encrypted, single‑node‑bound capsule.
Step 1 — Prepare the folder
my_skill/
├── manifest.json ← workflow metadata (plaintext)
├── SKILL.md ← agent routing instructions (plaintext, see Rule 1)
├── run.py ← entry script
├── helpers/
│ ├── fetch.py
│ └── parse.py
└── config/
└── settings.json
Plaintext by design: only manifest.json and SKILL.md stay readable. Scripts and configs are encrypted.
Step 2 — Get the consumer node's public key
# On the consumer's MGC node
mgc_get(info_type="__NODE_PUB__", info_owner="__NODE_PUB__")
# → returns a PEM public key string
Hand this PEM string to the author through any secure channel (the PEM is not secret — it is a public key).
Step 3 — Save and seal
# On the author's MGC node
mgc_save_file(
path="./my_skill",
info_owner="my_skill",
diff_2="my_skill",
)
mgc_seal_package(
info_owner="my_skill",
diff_2="my_skill",
ext04="",
)
# → a sealed capsule (.mgc_file)
Hand the .mgc_file to the consumer through any channel (email, USB, cloud drive). The capsule is safe even if intercepted.
6. Consumer Side — Run a Sealed Package
Three steps to import and execute a sealed capsule you received.
Step 1 — Import
mgc_save_file(path="./received.mgc_file")
# MGC unpacks the capsule into local entries
Step 2 — Discover and read
mgc_find(info_owner="my_skill", diff_2="my_skill")
# → list of entries inside the package
mgc_get(info_type="file", info_owner="my_skill", diff_2="my_skill")
# → read the manifest (plaintext)
Read SKILL.md (plaintext) to understand which scripts to call in which order.
Step 3 — Execute
mgc_run(info_owner="my_skill", diff_2="my_skill")
# → starts the entry script
The agent never sees the script source. Output flows back to the consumer's MGC runtime.
7. Three Critical Rules
These rules are non‑negotiable. Breaking them defeats sealing.
Rule 1 — Scripts must call each other via the MGC API, never via filesystem paths
If your scripts have a chain (A calls B calls C), they must invoke each other through the MGC REST API by info_type + info_owner + diff_*. They must not rely on local paths, because the local path on the consumer's machine may not match the author's layout.
# inside run.py — call another stored script by MGC location
def call_script(info_owner, ext02=None, diff_1=""):
payload = {
"info_type": "script",
"info_owner": info_owner,
"diff_1": diff_1,
"action": "run",
"ext01": "python",
}
if ext02:
payload["ext02"] = ext02
r = requests.post(
"http://127.0.0.1:57219/api/mgc/sensitive/get",
headers={"X-MGC-Token": MGC_TOKEN},
json=payload,
)
return r.json()
Rule 2 — The author's SKILL.md stays plaintext
The author's SKILL.md (the one shipped inside the package) is plaintext by design — it tells the consumer's agent which scripts to call and when. Without it, the agent has no routing instructions.
This is fine: the SKILL.md is the public API contract, not the implementation.
Rule 3 — Core logic lives in scripts, not in the prompt
Prompts and SKILL.md are plaintext — anyone can read them. Therefore:
- Do put core algorithms, business logic, and proprietary workflows inside scripts (which get sealed)
- Do not put them in the SKILL.md prompt (which stays plaintext)
- The SKILL.md should be a thin routing layer: "call script X with args Y when condition Z"
8. Security Mechanism (Quick View)
| Layer | What it does |
|---|---|
| AES‑256 | Encrypts each script body and config file in the package |
| RSA‑2048 | Wraps the AES key with the target node's public key (from __NODE_PUB__) |
| Single‑node binding | The wrapped AES key can only be unwrapped by the target node's private key |
| No re‑seal | The consumer node cannot re‑wrap the AES key for another node |
| Local execution | Decryption and execution both happen inside the consumer's MGC runtime; the plaintext is never written to disk in cleartext |
For full details, see the MGC_META_SKILL skill.
9. When NOT to Use
This skill is not the right tool if:
- ❌ You want to publish an open‑source skill (use plain
mgc_saveinstead) - ❌ You need to distribute to many nodes simultaneously (one capsule = one node; for multi‑node, seal once per node)
- ❌ Your workflow logic is so simple it can fit in a SKILL.md prompt (just write it directly)
- ❌ Your scripts cannot be refactored to call each other via the MGC API (chains that depend on local paths will break)
10. Related Skills
- MGC Blackbox (META_SKILL) — full reference for all MGC tools, fields, and security model. Read this first if you are new to MGC.
- MGC Script Execution Auth (Cross‑Device) — sealing model for single scripts (lighter‑weight than this folder‑level skill).
11. Coming Next
We are preparing a batch chain authorization feature. Send your node_pub to mirgincipher@outlook.com to claim a free first-batch trial.
相关技能
Zero‑exposure script execution using MGC Blackbox. Store scripts encrypted, execute in blackbox, AI sees no plaintext. Supports MCP (mgc_run) / API / WebUI execution, internal credential calls, and script sealing. Includes MGC 1.4.9 temporary workaround guidelines. This skill only executes scripts w
Run a security review workflow for Codex or ClawHub skills before activation or publication. Use when the user needs to inspect skill files, scripts, permiss...
Scan skills in a project directory for security issues and generate a markdown table report, then install skills from a local registry. Combines static analysis of code and markdown files with supply chain checks. Use when auditing a skills directory, generating a security summary table, or installi
Audit skill packages before publishing, then generate concrete fixes and a shareable proof card
A secure data analysis workflow suite based on MGC Blackbox, providing credential protection, zero-exposure script application, script sealing collaboration, and knowledge management. Also includes a Data Analyst Agent system prompt template.