ai-sentinel
由 amandiwakar 提供
Coding Agents & IDEs简介
Prompt injection detection and security scanning for OpenClaw agents.
描述
该技能的功能
Prompt injection firewall for OpenClaw agents. Hooks into gateway lifecycle events to scan inbound messages, tool call parameters, and tool results using heuristic pattern matching. Covers 7 threat categories including jailbreak, data exfiltration, and indirect injection.
Community mode runs entirely locally with no data leaving the machine, so teams get injection protection without transmitting message content to any external service.
何时使用
- Block prompt injection attempts in a public-facing customer support bot
- Audit tool call parameters for data exfiltration patterns before execution
- Monitor agent traffic for jailbreak attempts without blocking in monitor mode
- Enforce message filtering on an agent handling sensitive internal data
- Review threat detection analytics after a suspected attack via the Pro dashboard
示例工作流
User asks: Set up prompt injection protection on my OpenClaw agent
- 1
Installs ai-sentinel plugin via openclaw plugins install
- 2
Asks user to choose Community (local) or Pro (dashboard reporting) tier
- 3
Asks user to set detection mode (monitor or enforce) and confidence threshold
- 4
Generates plugin config under plugins.entries.ai-sentinel and writes it to OpenClaw configuration after user confirms
- 5
Tests detection by sending a known injection pattern through the gateway and verifying the log output
OpenClaw gateway scanning all lifecycle hooks for prompt injection, with detections logged or blocked based on chosen mode
前置要求
- AI_SENTINEL_API_KEY — Pro tier only; obtain from https://app.zetro.ai
- Zetro.ai account — Pro tier only, required for dashboard access and remote classification
安装
clawhub install amandiwakar/ai-sentinel运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。