ClawHub
Back to skill

Security audit

Neat

Security checks across malware telemetry and agentic risk

Overview

The skill has an understandable documentation-sync purpose, but it gives agents broad authority to edit, delete, and persist project or agent-memory state without enough visible scoping or confirmation.

Install only if you explicitly want an agent to modify project documentation, CLAUDE.md/AGENTS.md, and persistent memory. Use it on a clean branch, review a proposed file list before changes, require confirmation before deletion or memory/global config updates, and avoid broad rollback commands like git restore . unless you have backed up or stashed work.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Missing User Warnings

Low
Confidence
82% confidence
Finding
The file recommends `git restore .` or `git checkout -- <file>` as a rollback step after a bad run, but it does not explicitly warn that these commands discard uncommitted local changes. In a skill that performs automated doc and memory reconciliation, that omission increases the chance an agent or user could destroy unrelated in-progress work while trying to recover.

Session Persistence

Medium
Category
Rogue Agent
Content
### 第三步：实际修改（用工具，不只是描述）

你必须**真的用 Edit 修改现有文件、用 Write 创建新文件、用删除命令清理废弃文件**。"我会怎么改"的描述不算完成。

**顺序建议**：先改 docs/（改错影响外部）→ 再改 CLAUDE.md/AGENTS.md → 最后理记忆。先动外部优先级最高的，即使中途被打断，读者看到的也是对齐的最新状态。
Confidence
89% confidence
Finding
Write 创建新文件、用删除命令清理废弃文件**。"我会怎么改"的描述不算完成。 **顺序建议**：先改 docs/（改错影响外部）→ 再改 CLAUDE.md/AGENTS.md → 最后理记忆。先动外部优先级最高的，即使中途被打断，读者看到的也是对齐的最新状态。 **编辑原则**： - **减优于加**（最重要）：每次同步动作结束后，CLAUDE.md / AGENTS.md 净涨幅

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal