中文 EN

扫描报告

扫描新文件

✓ 可信 — 风险评分 5/100
上次扫描：2 天前 重新扫描
5 /100
AI Socializer | AI 社交者
Interact with AI social networks and communities. 在 AI 社交平台上互动。
Documentation-only skill with clear security boundaries, no implementation code, and well-defined operational rules for read-only social platform interaction.
技能名称AI Socializer | AI 社交者
分析耗时20.1s
引擎pi
✓
可以安装
Skill is safe to use. Monitor any future script additions to this skill before installation.

安全发现 2 项

严重性 安全发现 位置
低危
Platform flexibility through environment variables
The skill accepts any API URL from users without validation. While documented as intentional design for flexibility, users should be cautious of social engineering attacks where a malicious platform is introduced.
API Base URL | AI_SOCIAL_API_URL | 平台 API 地址
→ Verify the legitimacy of the platform before providing API credentials.
 SKILL.md:11
低危
Patrol logs written to filesystem
The heartbeat/patrol feature appends to ~/.openclaw/workspace/projects/ai-social/{platform}/patrol-log.md. This is filesystem:WRITE but is declared, scoped, and append-only (not overwriting).
projects/ai-social/{platform}/patrol-log.md
→ This is legitimate scoped logging. No action needed.
 SKILL.md:88
资源类型声明权限推断权限状态证据
文件系统 NONE NONE — No file operations in any scripts (none exist)
网络访问 READ READ ✓ 一致 API calls only to user-configured platform URL
命令执行 NONE NONE — No shell execution in any scripts (none exist)
环境变量 READ READ ✓ 一致 Reads AI_SOCIAL_API_URL and AI_SOCIAL_API_KEY only
技能调用 NONE NONE — No cross-skill invocation declared or implemented
剪贴板 NONE NONE — No clipboard access found
浏览器 NONE NONE — No browser automation declared or found
数据库 NONE NONE — No database access found
1 项发现
🔗
中危 外部 URL 外部 URL
https://www.example.com/api/v1
SKILL.md:16

目录结构

2 文件 · 7.0 KB · 190 行
Markdown 1f · 181L JSON 1f · 9L
├─ 📋 _meta.json JSON 9L · 300 B
└─ 📝 SKILL.md Markdown 181L · 6.7 KB

安全亮点

✓ Comprehensive security rules clearly separating data from commands
✓ No execution of social platform instructions — strictly read-only with user consent for any actions
✓ Prompt injection defense explicitly documented with examples
✓ API credentials scoped to user-configured platform domain only
✓ De-identification checklist prevents inadvertent location/system disclosure
✓ Patrol logging isolated to dedicated project files, protecting MEMORY.md from injection contamination
✓ All publishing operations require explicit human consent — no autonomous posting
✓ Comment replies require human approval before any response
✓ No external script downloads or remote code execution
✓ No credential exfiltration or external data transmission beyond declared API usage