Back to skill

Security audit

Neat Freak

Security checks across malware telemetry and agentic risk

Overview

This is a legitimate documentation and memory cleanup skill, but it asks for broad cross-project and global agent-memory access with automatic persistent edits that users should review carefully.

Install only if you want an agent to actively maintain project documentation and agent memory. Prefer invoking it with explicit commands like /sync or /neat, review planned edits before allowing deletions or global memory/config changes, and avoid running it in workspaces where unrelated projects or sensitive agent notes may be reachable.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Vague Triggers

High
Confidence
97% confidence
Finding
The trigger logic explicitly instructs the skill to activate on ambiguous milestone-style phrases and even bare terms like '整理'/'tidy' when prior dev context exists. This can cause the agent to launch a high-authority cleanup workflow unexpectedly, leading to unintended document edits, memory changes, or rule-audit actions without a clearly scoped user request.

Vague Triggers

Medium
Confidence
93% confidence
Finding
Several trigger phrases listed near the top are generic enough to match ordinary conversation rather than a deliberate request to run a destructive maintenance skill. Because this skill is authorized to inspect, edit, and delete docs and memory artifacts, accidental activation materially increases the risk of unintended state changes.

Agent Config Directory Access

High
Category
Agent Snooping
Content
- 新增数据库表 → CLAUDE.md + architecture 的 Data Model
- 新增大特性(跨多文件) → 以上全部 + architecture 新章节 + handoff 已完成清单
- 跨项目改动 → 上下游两边的 docs **都要对齐**(最常见的漏改场景)
- **退役 / 改名 / 下线** → `git show <删除 commit> --stat` 取被删的路由/导出符号/字段/枚举名,对每个跑 `grep -rn '<symbol>' docs/ <本 agent 记忆目录>`(Codex 还要 grep `~/.codex/AGENTS.md` + `~/.codex/memories/skills/`),**在同一次同步里清掉非载荷引用(示例代码/历史案例/枚举列举)**,别留到事后的「补漏」commit。死 skill 目录整个删。
- 记忆层面:相对时间→绝对日期、过期事实→改、重复→合并、已完成待办→删
- **过期开放项扫描**:grep 记忆里同时带「开放项标记(待办/未决/暂缓/搁置/待评估/仍未/观察期再评估/TODO)」**且**「绝对日期早于今天」的行(别裸扫日期——绝对日期满天飞、大多是正确历史;marker 同行才是信号)。每条强制处置:① 已落地→链接 commit 并删;② 没落地→从「计划」降级为「未决,未排期,触发条件=X」,别让它再冒充已排期承诺;③ 已放弃→删。**写「已完成」前先对照真实代码与产物核实**,别假设已上线。
Confidence
95% confidence
Finding
grep -rn '<symbol>' docs/ <本 agent 记忆目录>`(Codex 还要 grep `~/.codex/AGENTS.md` + `~/.codex/

Session Persistence

Medium
Category
Rogue Agent
Content
### 第四步: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
88% 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

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.