ClawHub
Back to skill

Security audit

Seedance 提示词生成器

Security checks for vulnerabilities and agentic risk

Overview

This skill is a prompt-writing helper for Seedance video generation and does not show hidden execution, persistence, credential access, or destructive behavior.

Installers should expect this skill to steer Seedance prompts toward Chinese wording unless English is requested, and to use local reference files plus an optional prompt linter. Review that language behavior if your workflow must stay English-only.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding
The declared description promises a comprehensive prompt construction, rewriting, auditing, and troubleshooting tool for many Seedance 2.0 use cases. The supplied code does only one small part of that: static linting of prompt text with a handful of heuristics. While some checks loosely relate to the declared auditing/troubleshooting theme—camera motion control, subject definitions, subtitle/watermark constraints, and edit/extension wording—the primary behavior is much narrower than advertised. There is no prompt generation, rewriting, multimodal planning, speech/audio analysis, or remediation of the many listed failure categories. This is therefore a material description/behavior mismatch due to substantial overstatement of capabilities and a different practical scope.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Content
Create stable, production-ready prompts for Doubao Seedance 2.0 video generation and editing. Treat Seedance as a multimodal director: separate what is in the frame from how events unfold over time, then produce prompt instructions that are explicit, concise, and operational.

Use the official guide in `references/official-guide.md` as the primary source when task-specific details are needed. Use the compact references below for fast routing and output rules.

## Core workflow
Confidence
85% confidence
Finding
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Lp3

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding
Without declared permissions the skill's intent is opaque and cannot be validated.

Natural-Language Policy Violations

Medium
Confidence
95% confidence
Finding
The file explicitly instructs the agent to switch to Chinese output based on inferred user language rather than an explicit user request. This can override user expectations or system behavior, creating a prompt-quality and consent issue even though it is not a classic security exploit. In this skill context, the risk is somewhat limited because it affects output language preference rather than data access or code execution.

Natural-Language Policy Violations

Medium
Confidence
93% confidence
Finding
Line L0264 states that if dialogue is in a 'small language' it must be marked as 'non-Chinese/non-English', which explicitly treats Chinese and English as the default allowed languages. This is a natural-language locale policy constraint and the document does not provide user opt-in or a clear region-specific justification for privileging those languages.

Natural-Language Policy Violations

Medium
Confidence
94% confidence
Finding
The instruction to add no-subtitle and no-text constraints specifically in Chinese imposes a language choice in the skill's guidance. This is a natural-language locale policy issue because the file does not indicate user opt-in or explain why Chinese is required here.

Natural-Language Policy Violations

Medium
Confidence
94% confidence
Finding
The instruction to add no-logo and no-watermark constraints in Chinese requires a specific language with no opt-in or documented necessity. That creates a locale-policy concern under the natural-language policy rules.

Natural-Language Policy Violations

Medium
Confidence
92% confidence
Finding
The linter’s detection logic is built around Chinese string literals such as 参考视频, 严格编辑, 镜头, 字幕, and watermark-related terms, which effectively enforces a specific language/locale for successful use. There is no user-facing indication in the file that the tool is Chinese-only or that users may choose another language, which can violate language/locale policy requirements.

Natural-Language Policy Violations

Low
Confidence
72% confidence
Finding
Line L0782 includes Korean dialogue content in a prescriptive example, while earlier language guidance sets Chinese/English as the normative baseline. This inconsistency reflects unclear language policy handling and could cause skills built from the guide to enforce or assume language constraints without explicit user choice.

Natural-Language Policy Violations

Low
Confidence
88% confidence
Finding
The markdown instructs users to add Chinese no-text constraints and references Chinese-style usage for on-screen text formatting. This can be a locale/language policy concern because it nudges output toward a specific language convention without explicitly requiring user choice or opt-in.

Static analysis

No suspicious patterns detected.