Browser Automation Skills 安全扫描报告 — 可信 | ClawSafe

5 /100

Browser Automation Skills

Browser automation skills for AI models — navigate, screenshot, interact, scrape, debug, test, and record browser sessions via Chrome DevTools Protocol

Legitimate browser automation skill using Playwright/CDP. All declared tools map cleanly to documented functionality with no hidden behavior.

技能名称Browser Automation Skills

分析耗时34.4s

引擎pi

✓

可以安装

No action needed. The skill is a standard browser automation package with well-documented capabilities.

资源类型	声明权限	推断权限	状态	证据
命令执行	`WRITE`	`WRITE`	✓ 一致	allowed-tools: Bash in all SKILL.md files
文件系统	`READ+WRITE`	`READ+WRITE`	✓ 一致	allowed-tools: Read + Write in all SKILL.md files
网络访问	`READ`	`READ`	✓ 一致	browser.py navigates URLs via CDP; inherent to browser automation purpose
浏览器	`READ+WRITE`	`READ+WRITE`	✓ 一致	Full CDP control via Playwright — navigate, click, type, screenshot, DOM read, c…
环境变量	`NONE`	`NONE`	—	No os.environ iteration; only reads BROWSER_CDP_ENDPOINT from env (line 55)
剪贴板	`NONE`	`NONE`	—	No clipboard access observed
技能调用	`NONE`	`NONE`	—	No skill-to-skill invocation

目录结构

12 文件 · 47.0 KB · 1341 行

Markdown 11f · 875L Python 1f · 466L

├─ ▾ 📁 docs

│ └─ 📝 api-reference.md Markdown 193L · 7.2 KB

├─ ▾ 📁 scripts

│ └─ 🐍 browser.py Python 466L · 16.7 KB

├─ ▾ 📁 skills

│ ├─ ▾ 📁 browser-context

│ │ └─ 📝 SKILL.md Markdown 66L · 3.1 KB

│ ├─ ▾ 📁 debug

│ │ └─ 📝 SKILL.md Markdown 64L · 1.9 KB

│ ├─ ▾ 📁 interact

│ │ └─ 📝 SKILL.md Markdown 53L · 1.8 KB

│ ├─ ▾ 📁 navigate

│ │ └─ 📝 SKILL.md Markdown 29L · 1.1 KB

│ ├─ ▾ 📁 record

│ │ └─ 📝 SKILL.md Markdown 38L · 1.3 KB

│ ├─ ▾ 📁 scrape

│ │ └─ 📝 SKILL.md Markdown 55L · 1.6 KB

│ ├─ ▾ 📁 screenshot

│ │ └─ 📝 SKILL.md Markdown 42L · 1.1 KB

│ └─ ▾ 📁 test

│ └─ 📝 SKILL.md Markdown 51L · 1.5 KB

├─ 📝 README.md Markdown 238L · 7.7 KB

└─ 📝 SKILL.md Markdown 46L · 2.0 KB

依赖分析 1 项

包名	版本	来源	已知漏洞	备注
`playwright`	`unpinned`	pip	否	Version not pinned; recommend pinning to a specific version

安全亮点

✓ All allowed-tools (Bash, Read, Write) are fully documented and aligned with stated capabilities

✓ No base64-encoded execution, eval(), or obfuscation patterns found

✓ No credential harvesting — no iteration through os.environ for secrets

✓ No C2 communication, reverse shells, or data exfiltration endpoints

✓ No remote script execution (curl|bash, wget|sh)

✓ No sensitive path access (~/.ssh, ~/.aws, .env files)

✓ Playwright CDP approach is well-documented and legitimate

✓ Lock overlay mechanism is declared in both SKILL.md and browser.py docstring

✓ Visual overlay provides user-facing stop button with clear UI feedback

✓ MIT-0 license, open-source transparency