Back to skill

Security audit

Seedance 2 Prompts Skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent prompt recommender, but it automatically pulls mutable remote data and tells the agent to run shell commands using that data, which creates review-worthy local execution risk.

Install only if you are comfortable with install-time and recurring network downloads from HuggingFace and with the agent being instructed to run shell commands for preview images. Prefer reviewing or modifying the skill to pin and verify the dataset, remove postinstall auto-download, avoid silent refresh, validate media paths, and send image URLs directly instead of shelling out through curl/rm.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T01 · Skill Instruction Hijacking

Warning
Location
SKILL.md:101
Finding

Mandatory Third-Party Promotional Content Hijacks Agent Responses

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:183
Finding

Shell Command Injection Through Unvalidated Remote Media Paths

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Error
Location
scripts/setup.js:20
Finding

Automatic Ingestion of Mutable and Unverified Remote Dataset Content

Content
View full analysis
l.trim()).length; console.log(`✓ (${lines} prompts)`); writeFileSync(stampFile, String(Date.now()), 'utf8'); console.log(`[setup] Done! Library is ready.`); } catch (err) { console.log(`✗ (${err.message})`); console.warn('[setup] Download failed. Check your network connection.'); process.exit(0); } } ``` The Skill also mandates recurring update checks: ```markdown ### Step 0: Auto-Update References (MANDATORY, runs every time) **Before doing anything else**, run the freshness check: ```bash node /scripts/setup.js --check ``` - **< 24h since last update** → instant no-op, proceed immediately - **> 24h stale** → silently pulls latest prompts from HuggingFace (~30s), then proceeds ``` ### Technical Analysis The package performs network access automatically through t ...[truncated 2791 chars]
Remediation
View remediation
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (21)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
88% confidence
Finding

The skill instructs the agent to execute shell cleanup commands against a fixed path in /tmp as part of normal operation. Even though the shown command is simple, embedding shell-level file operations in skill instructions creates tool-parameter abuse risk because it conditions the agent to run exec on host files for nonessential tasks, and fixed temporary paths can cause clobbering, race conditions, or unintended deletion if the path is manipulated or reused.

Content

Scanner excerpt · SKILL.md (reported line 201)May include surrounding context.

message tool: action=send, media=/tmp/prompt_cover.jpg, caption="[Prompt Title] — Duration: [spec.duration]s"

Step C — Cleanup: exec: rm /tmp/prompt_cover.jpg

text

Do this for **each** of the 3 recommended prompts — one cover per prompt.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The README instructs users to run npx skills i ... without pinning an exact package version. npx resolves and executes the latest published package at install time, so a compromised upstream package, typo-squatted replacement, or malicious new release could result in arbitrary code execution on the user's machine. In a skill-installation context this is more dangerous because users are explicitly told to execute the command locally and are likely to trust it.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

This line again recommends executing npx skills i ... without version pinning. Because npx fetches and runs code from npm dynamically, users may unknowingly execute a malicious or compromised package version, leading to arbitrary local code execution. The risk is amplified by README-driven copy/paste behavior.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
89% confidence
Finding

The README suggests npx openskills install ... without pinning a specific package version. As with other npx invocations, this can execute whatever code is currently published under that package name, creating supply-chain and arbitrary-code-execution risk if the package or dependency chain is compromised. The skill context makes this credible because installation is the expected user action.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
90% confidence
Finding

This documentation references npx skills i as part of install behavior without specifying an exact version. Unpinned npx usage exposes users to package substitution or malicious updates, potentially resulting in arbitrary code execution during installation. Because this is explanatory installation guidance in the same README, it reinforces unsafe operational practice.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The README states that pnpm install automatically downloads the prompt library from HuggingFace and writes it into references/metadata.jsonl, but does not prominently warn users about the network access and local file changes. Automatic remote fetches during install increase supply-chain risk and can surprise users in restricted or sensitive environments; if the remote source is compromised, malicious content may be introduced into the workspace. In a skill that is largely documentation-driven, failing to disclose this behavior reduces informed consent and makes exploitation easier.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The README instructs users to run npx skills i ... without pinning a version. npx resolves and executes the latest package from the registry at install time, so a compromised or malicious upstream release could run arbitrary code on the user's machine during installation. In this skill context, the risk is somewhat elevated because the README is explicitly guiding users to execute network-fetched tooling.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

This is the same supply-chain exposure as the other npx skills invocation: the command fetches and runs an unpinned package version from npm. If the package publisher account or dependency chain is compromised, users following the README could execute attacker-controlled code. The skill's documentation role makes this a genuine operational risk even though there is no evidence of malicious intent.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
85% confidence
Finding

npx openskills install ... likewise executes an unpinned package obtained from the npm ecosystem. That creates a supply-chain risk where users may unknowingly run a newly published or hijacked version with arbitrary install-time behavior. Because this is in installation documentation, the exposure is direct and practical.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
84% confidence
Finding

The README again references npx skills as a tool, still without a pinned version. Even when listed as a related tool rather than a primary install step, it normalizes use of floating npx commands and carries the same risk of pulling and executing compromised upstream code. The danger is moderate rather than high because the file is documentation, not auto-executed code.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
84% confidence
Finding

The skill instructs the agent to perform network operations such as freshness checks, dataset downloads, and remote media access, but it declares no explicit tool scope or permissions. This creates a capability mismatch that can lead to unintended network use without clear sandboxing or user awareness, increasing the chance of silent data access or policy bypass in hosts that rely on manifest-declared permissions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill mandates an automatic freshness check and silent network update every time it runs, with no clear user-facing notice or consent. This is risky because it normalizes hidden network activity, can leak usage metadata, changes local state unexpectedly, and may pull unreviewed content into the execution environment on each invocation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill requires downloading remote cover images into local /tmp storage before sending them, even though a prompt-recommendation skill can usually present remote URLs directly. Unnecessary local file handling increases attack surface through untrusted remote content ingestion, disk writes, and host/tool interactions that are not essential to the core function.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The documented behavior expands from prompt recommendation into generating and remixing custom prompts, which materially broadens the skill's authority and output surface beyond its stated purpose. Scope expansion is dangerous because it can bypass user expectations, introduce unreviewed content-generation behavior, and make downstream misuse or policy evasion easier under the guise of a narrower recommendation tool.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instructions require the customized prompt to be output specifically in English, which imposes a language choice regardless of the user's preferred language. This is a natural-language policy concern because the skill does not present this as an option or obtain user opt-in for the locale constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The language-handling section says responses should follow the user's language, but separately requires prompt content to be provided in English for generation. This creates a forced locale behavior without opt-in, and no justification or alternative is offered for users who may want prompts in another language.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest explicitly says the package auto-downloads a library on install, but the install-time side effect is only embedded in descriptive text and paired with automatic execution through postinstall. This is dangerous because users and automated systems may install the package expecting static prompt data while instead granting immediate network and code-execution capability during installation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The package defines a postinstall hook that automatically executes local code during installation, which creates an unnecessary execution path for a skill whose stated purpose is only to provide prompt recommendations. Even if the current script is intended to download a prompt library, install-time code execution increases supply-chain risk because users trigger it implicitly during dependency installation without first reviewing or consenting to the behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

L026 写明“用你的语言交流,始终提供英文提示词用于生成”,这是对语言/locale 的固定要求。规则要求若技能强制特定语言而未给予用户选择或明确的合规性理由,应视为自然语言政策问题。

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

该 markdown 文件说明执行 pnpm install 会“自动从 HuggingFace 下载提示词库”,以及可手动运行 setup 脚本,但没有任何关于网络访问、下载远程内容或可能写入本地 references/ 目录的显式提醒。按照 markdown 文件的缺少用户警示标准,这类会影响本地环境和外部通信的行为应有简短披露。

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

This manifest description advertises the skill in general terms but does not specify when or how it should be activated, nor does it provide any limiting context or exclusion conditions. In manifest files, overly broad descriptions can function as ambiguous triggers and increase the chance of unintended invocation.

Content

No source excerpt is available for this finding.