Back to skill

Security audit

Seedance 2.0 Prompt Writing

Security checks for vulnerabilities and agentic risk

Overview

This is a prompt-writing guide with no hidden code or credential behavior, though users should prefer manual installation and review the violent-fiction examples.

Install manually rather than using the unpinned npx command unless you verify the package version and publisher. Review the example templates before use, especially the fictional violence examples, and keep normal platform and safety rules in place when generating video prompts.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Warning
Location
README.md:22
Finding
Unpinned Third-Party Package Execution Through npx## Vulnerability Details **File Location**: `README.md`, lines 22–26 **Vulnerability Type**: Unpinned third-party dependency execution **Risk Level**: Medium The documented optional installation method executes a remotely resolved npm package without pinning its version or verifying its integrity. ```bash ### Option B: Via skills CLI ```bash npx skills add dexhunter/seedance2-skill ``` ``` ### Technical Analysis The command `npx skills add dexhunter/seedance2-skill` resolves and executes the `skills` package through the npm ecosystem. No exact package version, lockfile, integrity digest, or verified binary source is specified. Because package resolution occurs at installation time, the code executed by users can differ from the code available when this project was audited. If the package, publisher account, registry resolution path, or a transitive dependency is compromised, malicious package code or lifecycle scripts could execute on the user's system. This is a supply-chain weakness rather than evidence that the currently resolved package is malicious. The project also provides a manual-copy installation method that does not introduce this execution path. ### Attack Path 1. An attacker compromises the npm package, its publisher account, a dependency, or another relevant package-resolution component. 2. The attacker publishes a malicious version containing executable package logic or lifecycle behavior. 3. A user follows the optional installation instructions and runs: ```bash npx skills add dexhunter/seedance2-skill ``` 4. `npx` resolves and downloads the unpinned package version available at that time. 5. The malicious package code executes with the privileges of the user running the command. 6. Subject to those privileges and local security controls, the payload could access user-readable files, modify user-owned configuration, invoke network services, or install additional components. ### ...[truncated 602 chars]
Remediation
## Remediation Suggestions 1. Pin the CLI to an exact, reviewed version rather than allowing dynamic resolution: ```bash npx --yes skills@<reviewed-exact-version> add dexhunter/seedance2-skill ``` 2. Verify the selected package version and its transitive dependency tree before recommending it. 3. Record and validate an integrity digest where the installation workflow supports it. 4. Document the expected package publisher, official registry location, and source repository so users can detect name confusion or publisher changes. 5. Disable or avoid unnecessary package lifecycle scripts where feasible. 6. Retain the manual-copy method as the recommended default because it does not require executing a remotely resolved installer. 7. Advise users not to run the installation command with administrator or root privileges. 8. Periodically re-audit the pinned package before updating the documented version.
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (7)

Skill Enumeration

Medium
Category
Agent Snooping
Content
Clone or download this repository, then copy the skill file(s) to your Claude skills directory:

```bash
mkdir -p ~/.claude/skills

# English
cp SKILL.md ~/.claude/skills/seedance-prompt-en.md
Confidence
85% confidence
Finding
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.

Session Persistence

Medium
Category
Rogue Agent
Content
Clone or download this repository, then copy the skill file(s) to your Claude skills directory:

```bash
mkdir -p ~/.claude/skills

# English
cp SKILL.md ~/.claude/skills/seedance-prompt-en.md
Confidence
60% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding
npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Natural-Language Policy Violations

Medium
Confidence
92% confidence
Finding
The file is explicitly scoped to English via the skill name `seedance-prompt-en`, and the guidance positions the assistant as operating in English without any opt-in or alternative locale support. This is a natural-language policy concern because it imposes a language constraint on users rather than offering a language choice or documenting a justified region-specific limitation.

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The skill includes a concrete example prompt for violent interpersonal harm and coercive dramatic abuse, which goes beyond neutral formatting guidance and actively helps users author harmful content. In a prompt-writing skill, such examples normalize and facilitate generation of abusive violent scenarios, increasing misuse risk even if the skill is not directly executing code or exfiltrating data.

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The skill includes a detailed example prompt depicting intimate-partner violence and attempted murder without any safety framing, restriction, or guidance to refuse harmful real-world abuse content. In a prompt-writing skill, such examples normalize and operationalize generation of abusive violent scenes, which can enable creation of harmful media or harassment content more easily.

Vague Triggers

Low
Confidence
82% confidence
Finding
The README says, "Then ask your AI agent to help you write a Seedance 2.0 video prompt," but it does not specify exact trigger phrases, boundaries, or negative examples for when this skill should activate. In a markdown skill description, this can be considered an ambiguous activation description because it relies on a general request pattern rather than a narrowly defined invocation scope.

Static analysis

No suspicious patterns detected.