Back to skill

Security audit

诗遇 · 唐诗宋词

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign with medium confidence: the scanners report mostly disclosed, purpose-related behavior, but flag a mismatch around external skill installation and optional cloud note saving, and the review is incomplete because some referenced paths were missing.

Consider whether you want a skill that may install other skills and optionally send study notes to cloud services. The verdict is benign, not risk-free, and confidence is medium: although all nine listed components were scanned, the review is marked partial because some referenced paths were not found, so related behavior could not be fully assessed.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

Installation can ordinarily be a supporting implementation detail. Here, however, the entire supplied chunk consists of external installation commands, which do not substantiate the advertised locally bundled, zero-dependency functionality and introduce network use beyond the explicitly stated exceptions. The installed skills' subsequent behavior cannot be determined from this chunk.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 153)May include surrounding context.

md
| 服务 | 端点 | 发出去的只有 | 用途 |
|------|------|-------------|------|
| 搜韵开放 API | `https://api.sou-yun.cn/open/poem?key=%2A%2A%2A<诗名或诗句>&scope=Title|Sentence&jsonType=true` | 诗名 / 作者 / 诗句 | 版本异文核对;模式 B 深读档的历代笺注素材(取 1-2 条译成人话,翻不出人话宁可不用) |
| wttr.in | `https://wttr.in/<城市拼音>?format=j1` | 城市名 | 模式 C 日签的天气关联层 |

**为什么是这两个**:中文历代笺注与免注册天气这两个用途上,没有更合适的公开替代。两者都是**只读、免 key** 的公开接口——不传任何凭证、token、账号或身份信息,也不需要读取环境变量。搜韵注明为非商业用途接口,wttr.in 为开源服务。

The workflow supports saving generated study notes to cloud services, although the manifest limits network use to optional poem verification and weather queries.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest at L004 and L006 describes network access as limited to poem verification and weather, with only poem identifiers, lines, or city names sent externally. L217 instead instructs the agent to offer saving generated study notes to ima or Tencent Docs, which entails transmitting note content to another service. User consent and the exception documented at L100–L108 make this opt-in, but do not resolve the manifest's narrower disclosure.

Content

No source excerpt is available for this finding.