Tp2
- Category
- MCP Tool Poisoning
- Confidence
- 85% confidence
- Finding
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
- Content
Security audit
Security checks for vulnerabilities and agentic risk
ClawScan marks this skill suspicious because scanners report possible external data sharing, an unpinned package execution path, and other risky or unclear behavior; the evidence is incomplete because one referenced path was missing.
Review the external model data-sharing path and the package installation behavior before installing, and consider whether the skill’s testing guidance could touch operational knowledge data. The scan was partial because a referenced path was missing; treat the result as incomplete rather than as assurance that no other risks exist.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
The executable statement introduces an undeclared code-execution capability without validation or isolation. The surrounding comments acknowledge the vulnerability and recommend safer alternatives, but do not prevent execution. Although the file may be intended as security-review documentation, the supplied executable behavior does not implement the declared purpose.
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).
- ❌ `npm publish`
- ❌ 修改 `.sofagent/` 目录
- ❌ 硬编码密钥、令牌
- ❌ `rm -rf` / `git reset --hard`
## 📋 范围自检(每次提交前使用)
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).
| 检查项 | 命令 | 权限边界 | OpenClaw | WorkBuddy | Web | 结果标注 |
|---|---|---|:---:|:---:|:---:|---|
| 编排 | `command -v sofagent orchestrator` | 不可谎称编排可用 | ✅ | ⚠️ | ❌ | 编排=可用/手动 |
| bash | `command -v bash` | 不可 `rm -rf /`/删非项目文件/改系统配置/`curl\|bash` | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
| git | `command -v git` | 不可 `push --force` 到 main/master/改 `.git/config` | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
| jq/node | `command -v jq\|node` | — | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
| 文件写入 | shell 判定 | 不可覆盖 `.git/`/`~/.ssh/`/宪法文件 | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
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).
| 检查项 | 命令 | 权限边界 | OpenClaw | WorkBuddy | Web | 结果标注 |
|---|---|---|:---:|:---:|:---:|---|
| 编排 | `command -v sofagent orchestrator` | 不可谎称编排可用 | ✅ | ⚠️ | ❌ | 编排=可用/手动 |
| bash | `command -v bash` | 不可 `rm -rf /`/删非项目文件/改系统配置/`curl\|bash` | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
| git | `command -v git` | 不可 `push --force` 到 main/master/改 `.git/config` | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
| jq/node | `command -v jq\|node` | — | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
| 文件写入 | shell 判定 | 不可覆盖 `.git/`/`~/.ssh/`/宪法文件 | ✅ | ⚠️ | ❌ | ✅/❌/N/A |
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).
诊断表:
{ "step": 2, "status": "failed", "exitCode": 1, "stderrTail": "<最后 5 行>", "advice": "按 stderr 指路修复后重跑本步", "rollback": "rm -rf ~/.sofagent/<本版目录>(安装未完成时);已存在旧版则恢复备份" }
回滚:install.sh 迁移旧目录失败会自行中止(数据安全语义);Agent 侧只需确认 ~/.sofagent/ 状态与安装前一致(第 1 步检测时记的快照)。
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).
诊断表:
{ "step": 2, "status": "failed", "exitCode": 1, "stderrTail": "<最后 5 行>", "advice": "按 stderr 指路修复后重跑本步", "rollback": "rm -rf ~/.sofagent/<本版目录>(安装未完成时);已存在旧版则恢复备份" }
回滚:install.sh 迁移旧目录失败会自行中止(数据安全语义);Agent 侧只需确认 ~/.sofagent/ 状态与安装前一致(第 1 步检测时记的快照)。
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).
诊断表:
{ "step": 2, "status": "failed", "exitCode": 1, "stderrTail": "<最后 5 行>", "advice": "按 stderr 指路修复后重跑本步", "rollback": "rm -rf ~/.sofagent/<本版目录>(安装未完成时);已存在旧版则恢复备份" }
回滚:install.sh 迁移旧目录失败会自行中止(数据安全语义);Agent 侧只需确认 ~/.sofagent/ 状态与安装前一致(第 1 步检测时记的快照)。
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).
诊断表:
{ "step": 2, "status": "failed", "exitCode": 1, "stderrTail": "<最后 5 行>", "advice": "按 stderr 指路修复后重跑本步", "rollback": "rm -rf ~/.sofagent/<本版目录>(安装未完成时);已存在旧版则恢复备份" }
回滚:install.sh 迁移旧目录失败会自行中止(数据安全语义);Agent 侧只需确认 ~/.sofagent/ 状态与安装前一致(第 1 步检测时记的快照)。
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).
🟢继续 / 🟡调整(子任务间→改下个子任务;60%→全量重编排剩余)/ 🔴暂停等确认。历史优先:有 Loop 记录时加倍检查最容易出问题的节点。
> ⛔ 重大操作前自检(约束回响):执行 rm / git reset / DROP / 外部 API 等不可逆操作前,先自问——当前生效的铁律有哪些?最近一条反思区记录是什么?如果答不上来,说明加载链已失效,暂停操作等用户确认。
> ⛔ **防雪崩**:共享状态(文件系统、git working tree、环境变量)被多个子任务并发修改,会导致 Agent「自己修自己改、越修越乱」。规则:
> 1. 子任务间必须显式标注共享状态变更——在 task/logs 中记录「修改了 X 文件/变量 Y」
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).
| 🟡 | 多步清楚/3-5文件 | 「收到。中途需要确认的问你。」 | ❌ |
| 🔴 | 模糊/多模块/高风险/犹豫 | 「涉及[简述],拆解一下?」→ 确认后澄清 | ✅ |
**硬信号强制 🔴**:文件删除/重构≥5个 | 破坏性操作(rm -rf/git reset/DROP)| 首次用新工具 | 用户提供Spec。硬信号>软判断,即使用户说「直接执行」。⛔ 🔴 后严禁自主扫描文件——仅输出 1.3 响应,等确认。
## 1.4 两轮澄清(仅 🔴 + 用户同意)
The workflow directs the agent to incorporate local node definitions into model prompts and recommends direct use of the external DeepSeek API without a data-classification or approval gate. In an enterprise deployment, these definitions can contain confidential workflows or customer information, creating an external disclosure path even though model-based decomposition is a legitimate function.
This is a TypeScript execution command, not an MCP server reference, so the detector's categorization is inaccurate. Nevertheless, npx ts-node can fetch and execute an unpinned registry package when no local installation is available, introducing a conditional supply-chain execution risk during an otherwise routine audit.
The response table prescribes Chinese replies, and line 4 requires users to receive only the corresponding final response. No instruction permits translating these replies to the user's language, offers a language choice, or justifies a Chinese-only locale constraint.
These terms can occur in routine requests unrelated to the code-change auditing role described on line 10. The activation instruction does not explicitly require that narrower context, so it could load the audit workflow for unrelated checking tasks.
Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
| 层 | 是什么 | 模板 |
|---|---|---|
| 📄 文档层 | 节点详细说明(人读 + 编排模块读) | `FDE/templates/nodes/node-template.md` |
| 🧠 Skill 层 | Agent 行为指令(节点的大脑) | `FDE/templates/skills/skill-template/SKILL.md` |
| 🔴 运行层 | 真实环境配置 | 写进部署清单 |
2. **黄金顺序(铁律)**:先运行层(跑通真实任务)→ Skill 层(抽象为模板)→ 文档层(补人读文档)。禁止直接写 Skill
The skill requires audit results to use the supplied Chinese-language templates and explicitly prohibits rewording them. This constrains user-facing output to Chinese without offering language selection or documenting a justification for the language restriction.
The adversarial-testing guidance directs writes of injection-bearing pages and changes to knowledge trust and sensitivity metadata without requiring isolated fixtures or restoration. In this deployed-enterprise context, following those directions on operational data could contaminate retrieval results or cause legitimate knowledge to be excluded, even though the testing purpose is defensive.
Suspicious Unicode normalization or mixed-script content