Back to skill

Security audit

Seedance 2.0 Video Gen

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent paid WeryAI Seedance video-generation skill, with a real but disclosed API-endpoint override risk users should control.

Install only if you are comfortable sending prompts, settings, task IDs, and public HTTPS reference-image URLs to WeryAI for paid generation. Keep WERYAI_API_KEY out of files, use a scoped or low-balance key where possible, and do not set WERYAI_BASE_URL or WERYAI_MODELS_BASE_URL unless you fully trust the endpoint.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/video_gen.js:19
Finding

Unrestricted API Endpoint Overrides Can Disclose the WeryAI API Key

Content
View full analysis
controller.abort(), 30000); try { const res = await fetch(fullUrl, { method, headers, body: body != null ? JSON.stringify(body) : undefined, signal: controller.signal, }); ``` ```js async function apiRequest(method, path, body, apiKey) { return httpJson(method, BASE_URL + path, body, apiKey); } async function fetchModelsRegistry(apiKey) { return httpJson('GET', MODELS_BASE_URL + MODELS_API_PATH, null, apiKey); } ``` The related vendored client follows the same pattern: ```js const url = `${baseUrl}${path}`; const headers = { Authorization: `Bearer ${apiKey}`, 'Content-Type': 'application/json', }; ``` ```js const res = await fetch(url, { method, headers, body: body ? JSON.stringify(body) : undefined, signal: controller.signal, }); ``` ### Technical Analysis The Skill legitimately requires network access and must send `WERYAI_API_KEY` to WeryAI to authenticate video-generation and model-registry requests. Sending the credential to the documented default WeryAI hosts is therefore necessary for the declared functionality. However, `WERYAI_BASE_URL` and `WERYAI_MOD ...[truncated 2788 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (22)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Yes, this is a mismatch. The declared purpose centers on Seedance 2.0 video generation workflows, but the actual code only queries the user's WeryAI generation balance. That is a materially different primary purpose and capability. While balance checking may be related to the same vendor ecosystem, it is not a supporting implementation detail of video generation in this code chunk; it is a separate function with distinct behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The code chunk does not implement or expose video generation behavior. It simply imports and re-exports a balance-related executor from weryai-core/balance.js. That suggests the actual function is to check or retrieve WeryAI account/API balance, which is materially different from the declared purpose of generating Seedance 2.0 videos. This is a clear description-behavior mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

There is a clear description-behavior mismatch. The description promises a Seedance 2.0 video generator with multiple generation workflows, including controlled start/end frame transitions. However, the code only calls fetchModelRegistry(ctx), validates input.mode against three mode names, and returns arrays of registry entries. No video generation request, image handling, storyboard processing, or frame transition logic appears in this chunk. Accessing the WeryAI API for model metadata is related to the broader domain, but the primary behavior of this code is model enumeration, not video creation.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The description presents a broad Seedance 2.0 video generation skill supporting several generation modes. However, this code chunk specifically posts to an image-to-video endpoint and only includes fields consistent with single-image image-to-video generation. There is no evidence in this chunk of text-to-video, multi-image inputs, storyboard handling, or first/last-frame transition support. This is a material description-to-behavior mismatch because the declared primary capabilities are broader than what the supplied code actually implements.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 16)May include surrounding context.

md
**Dependencies:** `scripts/video_gen.js` in this directory + `WERYAI_API_KEY` + Node.js 18+. No other Cursor skills are required.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 68)May include surrounding context.

md
**Dependencies:** `scripts/video_gen.js` in this directory + `WERYAI_API_KEY` + Node.js 18+. No other Cursor skills are required.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 76)May include surrounding context.

md
**Dependencies:** `scripts/video_gen.js` in this directory + `WERYAI_API_KEY` + Node.js 18+. No other Cursor skills are required.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 77)May include surrounding context.

md
**Dependencies:** `scripts/video_gen.js` in this directory + `WERYAI_API_KEY` + Node.js 18+. No other Cursor skills are required.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 81)May include surrounding context.

md
**Dependencies:** `scripts/video_gen.js` in this directory + `WERYAI_API_KEY` + Node.js 18+. No other Cursor skills are required.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill explicitly requires environment access and network connectivity, but it does not declare a restrictive tool scope such as allowed tools or permissions. In an agent environment, that widens the effective execution surface and can permit unintended use of secrets or outbound requests beyond the narrowly described workflow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instruction says to turn the user's brief into a complete production prompt "in English" before every paid call. This forces a specific language choice and does not offer the user an opt-in or alternative, which conflicts with the language/locale policy criteria.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 199)May include surrounding context.

md
- Do not use local file paths for references; only public `https` URLs are supported.
- Do not invent undocumented API fields beyond the JSON surface in this skill.
- Do not assume multi-image or first/last-frame support without checking live metadata when the behavior matters.
- Do not store the secret value of `WERYAI_API_KEY` inside the repository.
- Do not copy parameter names blindly from another provider without checking the normalized fields in this document.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

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

md
- Do not use local file paths for references; only public `https` URLs are supported.
- Do not invent undocumented API fields beyond the JSON surface in this skill.
- Do not assume multi-image or first/last-frame support without checking live metadata when the behavior matters.
- Do not store the secret value of `WERYAI_API_KEY` inside the repository.
- Do not copy parameter names blindly from another provider without checking the normalized fields in this document.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This flow instruction directs the operator to expand the brief into a full English prompt. As written, it imposes a language requirement on all users instead of providing a choice or documenting a justified locale constraint.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The client sends JSON request bodies and authorization headers via fetch(), which may include user or system data. While verbose logging exists, it is optional and developer-facing rather than a user disclosure, and there is no confirmation prompt, comment, or docstring warning about this transmission.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The code posts the assembled request body to a remote /v1/generation/image-to-video endpoint, and that body includes user-supplied fields such as prompt and image. While there is validation and error handling, this file does not provide a confirmation prompt, user-visible notice, or explanatory comment/docstring disclosing that user content is transmitted over the network.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The code sends a request body containing prompt text and image inputs to a remote API via client.post(apiPath, body). While there is some warning logging for model-validation issues, there is no user-facing confirmation, disclosure comment, or explanatory docstring here indicating that user content will be transmitted off-system.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

This code performs an HTTP POST to a generation endpoint using a body that may include prompt text and image data. While there is internal logging for validation warnings, this file does not include a confirmation prompt, user-facing disclosure, or explanatory comment/docstring warning that user content will be transmitted to an external service.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The default parameters specify a "concise cinematic English prompt" as the prompt style. This is another natural-language instruction that fixes the language to English without documenting user choice or opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The user-facing text in this README is entirely in Chinese, with no indication that language selection is optional or that the skill is intended only for a Chinese-language audience. This can violate language/locale policy expectations for skills unless the locale restriction is explicit and justified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The code performs a network call with client.get(MODELS_API_PATH, { retries: 1 }), which can transmit system or user context through the client configuration, but this file provides no user-facing print, comment, or docstring describing that network access. Under the code-file warning rule, network operations should have some visible disclosure unless clearly documented elsewhere or obviously inherent to the skill's stated purpose, which is not evident from this file alone.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The code performs an HTTP POST to submit the assembled request body to a remote text-to-video endpoint. Although validation and dry-run handling are present, this file does not include a confirmation prompt or explicit user-facing notice that prompt content and related parameters will be transmitted off-system.

Content

No source excerpt is available for this finding.