AI Agent Assurance Suite

Problema

Agents need runtime guardrails and regression evidence before production.

Solution

Workspace Guard SDK (file confinement + command policy) + Red-Team Benchmark Lab (broken/fixed pairs).

Why it matters

Agent safety becomes testable, not just declared.

What you get

• Two standalone packages with their own tests

What is not included

• MIMARIO control-room app (crown-jewel)
• Any MASTER repo

Architecture

Guard SDK (policy → enforcement) + Benchmark Lab (fixtures → regressions → evidence).

File structure preview

packages/
packages/engineering/
packages/local-core/
src/
tests/

Exemplu input

{ command, policy }

Exemplu output

{ allow|deny, reason }

Integrare

import { guard } from './src/entry.mjs'

Runtime & version

Node.js 20+ · v1.0.0

Teste

Per-constituent test suites

Limitations

• In-process guardrails; not a sandbox for native code.

License

Proprietary non-exclusive

Delivery

GENESIS Direct: Active · private · automatic · USDC on Base

Canale disponibile

GitBuyer x402Git