中文 EN

扫描报告

扫描新文件

⚠ 低风险 — 风险评分 20/100
上次扫描：22 小时前 重新扫描
20 /100
lobster-keeper
自动执行日常维护：状态检查、自我复盘、记忆流动、任务记录。基于龙虾饲养员的六条经验。
A simple self-maintenance skill that reads agent state files and generates status reports, with minor doc-to-declaration mismatches for undeclared tool references.
技能名称lobster-keeper
分析耗时33.2s
引擎pi
✓
可以安装
No immediate action required. Consider adding explicit `allowed-tools` (Read, list_skills) and resource declarations (filesystem:READ) to the frontmatter to eliminate the documentation gap.

安全发现 2 项

严重性 安全发现 位置
低危
Undeclared tool references in documentation 文档欺骗
SKILL.md line 14 references 'write' and 'exec' tools without declaring them in the frontmatter. However, no actual code implements or uses these tools.
每一步都可以调用内置工具（如 `read`, `write`, `exec`, `list_skills` 等）来完成
→ Remove 'write' and 'exec' from the documentation example, or declare them in the frontmatter if they are genuinely needed.
 SKILL.md:14
低危
Missing resource permission declarations privile_escalation
The frontmatter lacks allowed-tools and resource declarations, making it unclear what permissions this skill requires. While the actual behavior is read-only, this creates an ambiguity gap.
No allowed-tools or resource declarations in frontmatter
→ Add explicit allowed-tools declaration: Read, list_skills (or equivalent), and resource: filesystem:READ.
 SKILL.md:1
资源类型声明权限推断权限状态证据
文件系统 NONE READ ✓ 一致 SKILL.md:27 — reads IDENTITY.md, SOUL.md, USER.md, MEMORY.md, memory/YYYY-MM-DD.…
技能调用 NONE READ ✓ 一致 SKILL.md:19 — calls list_skills to list installed skills
命令执行 NONE NONE — SKILL.md:14 — 'exec' tool is mentioned but not actually used or present in code

目录结构

1 文件 · 1.8 KB · 36 行
Markdown 1f · 36L
└─ 📝 SKILL.md Markdown 36L · 1.8 KB

安全亮点

✓ Single-file, self-contained skill with no external dependencies or scripts
✓ No network requests, credential access, or data exfiltration behavior
✓ No obfuscation, base64-encoded content, or suspicious patterns
✓ Functionality is limited to benign self-maintenance: reading state files and generating reports
✓ No sensitive paths (SSH, AWS, .env) are accessed
✓ No remote code execution, downloads, or supply chain risks