据 AlphaMoat 统计,截至 2026-10,Chrome 插件 AI Agent Guardrails 用户数为 30,环比 +3.4%,评分 5(1 人评价),在全部 10,012 款 AI 浏览器插件中排名第 6,874 位,在「通用 AI」分类(2,527 款)中排名第 1,696 位。
AI Agent Guardrails 是什么
AI Agent Guardrails adds a local safety layer that protects autonomous browser-based AI agents from malicious instructions hidden inside web pages. It scans page content in real time for instruction overrides, role manipulation, delimiter injection, encoding attacks and system prompt leak attempts, and when the detected risk crosses the configured threshold it shows a prominent warning or replaces the page with a hard block. An optional agent safety mode hides send and compose controls on supported webmail pages, custom site controls whitelist trusted or blocklist untrusted domains, and a local scan history keeps up to 100 recent results with severity scores. Scanning uses the bundled Vard detection library and runs entirely inside the browser with no accounts, telemetry or cloud processing.
- 01Real-time prompt injection detection
- 02Warning and hard block thresholds
- 03Email action protection mode
- 04Site whitelist and blocklist
- 05Local scan history, no telemetry
展开 Chrome 商店完整介绍▾
Prompt injection guardrails for OpenClaw, Hermes, Pi, Codex, Claude Code, and other autonomous AI agents. 🛡️ AI Agent Guardrails — Prompt Injection Protection for Autonomous AI Agents Browser-based AI agents can encounter malicious instructions hidden inside web pages. AI Agent Guardrails adds a local safety layer that detects suspicious content and warns or blocks your agent before it acts.
⚡ Real-Time Prompt Injection Detection Scans page content for instruction overrides, role manipulation, delimiter injection, encoding attacks, and system prompt leak attempts. When detected risk crosses your configured threshold, the extension displays a prominent warning or replaces the page with a hard block.
🚫 Optional Email Action Protection Agent Safety Mode hides common send and compose controls on supported webmail pages, reducing the risk of unintended email actions.
🌐 Custom Site Controls Whitelist trusted domains to skip scanning. Blocklist untrusted domains to prevent agents from reading or acting on their content.
📋 Local Scan History Review up to 100 recent scan results, including the page, time, severity score, detected threat types, and action taken. History stays inside your browser and can be cleared at any time.
🔒 Configurable Guardrails Choose which threat types to detect and configure separate thresholds for warnings and hard blocks.
Local by design - Scanning uses the bundled Vard detection library and runs entirely inside your browser. Page content and scan results are never sent to us. No accounts, telemetry, analytics, or cloud processing. Settings and history remain in local browser storage.
Designed for browser-based agent workflows involving OpenClaw, Hermes, Pi, Codex, Claude Code, and other autonomous AI agents.
插件档案
用户增长趋势
「通用 AI」分类下的热门插件
Read pages and carry out browser tasks with Claude, while retaining control over its actions
+ 对比
ChatGPT for Chrome lets the assistant control the browser, automate tasks, and answer questions on any page.
+ 对比
Use multiple AI models in a browser sidebar to read, write, translate, and work with files
+ 对比
ChatGPT search makes ChatGPT the default search engine in Chrome for fast, sourced answers.
+ 对比
Monica is an all-in-one AI assistant in the browser, combining chat, research, writing, translation, search and image creation behind one shortcut.
+ 对比
YouTube Summary with ChatGPT & Claude turns long videos, web articles and PDFs into short summaries so you can skip the parts you do not need.
+ 对比BAMBAM 的其他插件
A local, open-source chat UI for OpenCode Go that uses your own API key and saves nothing but your browser.
+ 对比数据月份:2026-10 · 用户数与评分来自 Chrome 应用商店公开数据,按月聚合更新。口径说明见数据方法。