Back to skill

Security audit

元盾 yotta-guardian

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed safety checker for risky agent actions; the alarming command and credential strings are used as examples, rules, and tests rather than as actions it performs.

Install with a specific target such as --agent or --dir, avoid broad global installation unless you intend to enable it across multiple agents, and configure external verifiers only if you trust the command or service receiving command and content previews. Audit and report files persist on disk when requested.

Vulnerability Patterns
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (88)

Tool Parameter Abuse

High
Category
Tool Misuse
Content
YottaMeta 自有实现首版(护栏/拦截方向参考开源社区 safe-guardian 类技能思路,已完全重写,零依赖、无上游代码):

- **零依赖自研引擎**(scripts/yotta_guardian.py,Python 3.8+ 标准库):确定性规则引擎 + 可插拔意图验证,对 exec / write / edit / read / run / shell 工具调用做安全评估。
- **四层规则**:文本模式(下载即执行 / 编码执行 / 反向 shell / 系统文件追加)+ argv 级动词/目标分析(rm / dd / mkfs / chmod / chown / 提权 / 防火墙 / 服务 / 持久化 / 反向 shell)+ 敏感路径(/etc 核心文件、/boot、/dev、SSH 授权、Windows 系统目录与 hosts、注册表启动项等)+ 写入内容(私钥 / 密钥令牌)。
- **三档策略**:default(拒绝 high+)/ strict(拒绝 medium+)/ loose(仅拒绝 critical);放行规则(--allow / --allow-path)可降级 high,critical 不可覆盖。
- **可插拔意图验证(不绑模型)**:默认零依赖;--heuristic 内置本地启发式;--verifier / 配置文件外接任意验证器(stdin/stdout JSON 协议)。
- **审计**:JSONL 审计日志 + audit 子命令(--tail / --denied / --since / --tool / --json)。
Confidence
85% confidence
Finding
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Tool Parameter Abuse

High
Category
Tool Misuse
Content
YottaMeta 自有实现首版(护栏/拦截方向参考开源社区 safe-guardian 类技能思路,已完全重写,零依赖、无上游代码):

- **零依赖自研引擎**(scripts/yotta_guardian.py,Python 3.8+ 标准库):确定性规则引擎 + 可插拔意图验证,对 exec / write / edit / read / run / shell 工具调用做安全评估。
- **四层规则**:文本模式(下载即执行 / 编码执行 / 反向 shell / 系统文件追加)+ argv 级动词/目标分析(rm / dd / mkfs / chmod / chown / 提权 / 防火墙 / 服务 / 持久化 / 反向 shell)+ 敏感路径(/etc 核心文件、/boot、/dev、SSH 授权、Windows 系统目录与 hosts、注册表启动项等)+ 写入内容(私钥 / 密钥令牌)。
- **三档策略**:default(拒绝 high+)/ strict(拒绝 medium+)/ loose(仅拒绝 critical);放行规则(--allow / --allow-path)可降级 high,critical 不可覆盖。
- **可插拔意图验证(不绑模型)**:默认零依赖;--heuristic 内置本地启发式;--verifier / 配置文件外接任意验证器(stdin/stdout JSON 协议)。
- **审计**:JSONL 审计日志 + audit 子命令(--tail / --denied / --since / --tool / --json)。
Confidence
85% confidence
Finding
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Tool Parameter Abuse

High
Category
Tool Misuse
Content
YottaMeta 自有实现首版(护栏/拦截方向参考开源社区 safe-guardian 类技能思路,已完全重写,零依赖、无上游代码):

- **零依赖自研引擎**(scripts/yotta_guardian.py,Python 3.8+ 标准库):确定性规则引擎 + 可插拔意图验证,对 exec / write / edit / read / run / shell 工具调用做安全评估。
- **四层规则**:文本模式(下载即执行 / 编码执行 / 反向 shell / 系统文件追加)+ argv 级动词/目标分析(rm / dd / mkfs / chmod / chown / 提权 / 防火墙 / 服务 / 持久化 / 反向 shell)+ 敏感路径(/etc 核心文件、/boot、/dev、SSH 授权、Windows 系统目录与 hosts、注册表启动项等)+ 写入内容(私钥 / 密钥令牌)。
- **三档策略**:default(拒绝 high+)/ strict(拒绝 medium+)/ loose(仅拒绝 critical);放行规则(--allow / --allow-path)可降级 high,critical 不可覆盖。
- **可插拔意图验证(不绑模型)**:默认零依赖;--heuristic 内置本地启发式;--verifier / 配置文件外接任意验证器(stdin/stdout JSON 协议)。
- **审计**:JSONL 审计日志 + audit 子命令(--tail / --denied / --since / --tool / --json)。
Confidence
85% confidence
Finding
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding
A second independently reported behavior mismatch indicates the same core issue: the advertised safety-only role conflicts with installer-like behavior that writes to disk, copies files, removes `.git`, and performs bulk/global installation into agent-specific directories. Such hidden or under-disclosed side effects increase the risk of privilege overreach, persistence, and unreviewed spread across multiple agent toolchains.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding
A second independently reported behavior mismatch indicates the same core issue: the advertised safety-only role conflicts with installer-like behavior that writes to disk, copies files, removes `.git`, and performs bulk/global installation into agent-specific directories. Such hidden or under-disclosed side effects increase the risk of privilege overreach, persistence, and unreviewed spread across multiple agent toolchains.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Credential Access

High
Category
Privilege Escalation
Content
## 何时使用

- 代理要执行高风险命令:递归删除系统路径、磁盘格式化(mkfs / fdisk / dd 写设备)、提权(chmod 全权限 / chown 系统路径 / 账户管理)、防火墙改动(iptables 清空 / ufw disable / netsh 关闭)、反向 shell(netcat 执行 / bash /dev/tcp)、下载即执行(curl / wget 管道交给 shell)等;
- 代理要写入系统敏感路径(/etc/passwd、/etc/sudoers、SSH 授权文件、Windows hosts / 系统目录、注册表启动项)或修改系统配置;
- 需要在执行危险操作前做一次确定性安全检查(gate),并留下审计记录。

**Do NOT trigger**:
Confidence
60% confidence
Finding
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Ae1

High
Category
analysis-evasion
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Ae1

High
Category
analysis-evasion
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Ae1

High
Category
analysis-evasion
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Ae1

High
Category
analysis-evasion
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Ae1

High
Category
analysis-evasion
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Ae1

High
Category
analysis-evasion
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Tool Parameter Abuse

High
Category
Tool Misuse
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"

# 检查危险命令(默认拒绝,退出码 3)
python3 scripts/yotta_guardian.py check exec --cmd "rm -rf /"

# 检查写操作
python3 scripts/yotta_guardian.py check write --path /etc/passwd --content "..."
Confidence
85% confidence
Finding
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Tool Parameter Abuse

High
Category
Tool Misuse
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"

# 检查危险命令(默认拒绝,退出码 3)
python3 scripts/yotta_guardian.py check exec --cmd "rm -rf /"

# 检查写操作
python3 scripts/yotta_guardian.py check write --path /etc/passwd --content "..."
Confidence
85% confidence
Finding
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Tool Parameter Abuse

High
Category
Tool Misuse
Content
python3 scripts/yotta_guardian.py check exec --cmd "git status"

# 检查危险命令(默认拒绝,退出码 3)
python3 scripts/yotta_guardian.py check exec --cmd "rm -rf /"

# 检查写操作
python3 scripts/yotta_guardian.py check write --path /etc/passwd --content "..."
Confidence
85% confidence
Finding
Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Static analysis

No suspicious patterns detected.