Agent Rbac — AI skill for Claude Code

agent-rbac

Role-based access control and memory isolation framework for multi-user AI agents.

[中文文档](#中文文档) | [English](#english-documentation)

---

Product Design Docs

• Design Index
• Product Architecture
• Policy Model Design
• Host Contract & Adapter Architecture
• Audit & Explainability
• Adaptive Policy Copilot
• Data, Storage & Privacy
• Implementation Roadmap

Runtime Usage

Production usage is no longer limited to importing the library. The package now includes:

• host contract types
• file-backed audit storage
• adaptive policy copilot storage and services
• a production runtime orchestrator
• first-party OpenClaw, Claude Code, and Codex adapters
• an installable OpenClaw plugin package
• a CLI for evaluation, audit review, suggestions, and familiarity queries

CLI Quick Start

npm install
npm run build
node dist/cli.js evaluate \
  --config ./examples/permissions.production.json \
  --state-dir ./.agent-rbac-state \
  --request ./examples/openclaw-request.json \
  --host openclaw

node dist/cli.js evaluate \
  --config ./examples/permissions.production.json \
  --state-dir ./.agent-rbac-state \
  --request ./examples/claude-code-request.json \
  --host claude-code

node dist/cli.js evaluate \
  --config ./examples/permissions.production.json \
  --state-dir ./.agent-rbac-state \
  --request ./examples/codex-request.json \
  --host codex

Useful commands:

node dist/cli.js audit timeline --state-dir ./.agent-rbac-state
node dist/cli.js suggestions --state-dir ./.agent-rbac-state --user alice
node dist/cli.js familiarity --state-dir ./.agent-rbac-state --user alice

OpenClaw Plugin Install

Build the package, then install it into a local OpenClaw runtime with on