Back to skill

Security audit

WeryAI Music Generator

Security checks for vulnerabilities and agentic risk

Overview

This music-generation skill mostly matches its stated purpose, but it also ships undisclosed podcast-generation code and has risky endpoint and upload scoping around API credentials.

Review before installing. Use only the official WeryAI API host unless you control and trust the replacement endpoint, avoid local reference_audio paths unless you intend to upload that file, do not use arbitrary webhook URLs, and confirm paid submit or wait runs before spending credits. The bundled podcast code should be removed or explicitly disclosed before this is treated as a tightly scoped music-only skill.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/vendor/weryai-core/client.js:3
Finding

Configurable API Base URL Can Exfiltrate the API Key and User Content

Content
View full analysis
0) { const delay = Math.min(2000 * Math.pow(2, attempt - 1), 8000); if (verbose) log(`Retry ${attempt}/${retries} after ${delay}ms...`); await sleep(delay); } try { const controller = new AbortCo ...[truncated 3502 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/vendor/weryai-music/validators.js:123
Finding

Webhook Validation Accepts Loopback and Private Network Destinations

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (26)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Podcast text and audio generation with podcast-specific polling and result handling is inconsistent with the declared music skill. This mismatch is dangerous in context because the skill requests API credentials and may upload user-provided media, so inaccurate documentation can directly affect privacy, billing, and informed consent.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The manifest says this skill generates music, vocal songs, and instrumental tracks through a music API, but this file documents a shared runtime for podcast generation, including podcast speakers, podcast text generation, and podcast audio generation. That is a semantic mismatch in claimed behavior versus documented implementation scope, not an expected implementation detail of a music-generation skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest describes a skill for generating music, vocal songs, and instrumental tracks through a music API, but this file builds a podcast-generation payload and targets a /v1/generation/podcast/.../audio endpoint. That is a materially different content domain than music generation and indicates the implemented behavior exceeds or diverges from the declared skill purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This file exposes podcast speaker-listing functionality inside a skill described as a music generator, which is a capability/intent mismatch. Even though the code itself is simple and uses input validation plus URL encoding, the undocumented extra capability expands the skill's reachable surface and can mislead users, reviewers, or policy enforcement about what the skill actually does.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file belongs to a skill declared as a music generator, but it submits data to a podcast text-generation endpoint and returns script content. That capability mismatch is dangerous because it can route user input to an undeclared remote function, undermining user consent, review assumptions, and policy scoping for the skill.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This file implements a full podcast text/audio generation workflow even though the skill manifest is for music generation. That scope mismatch is dangerous because it can expose capabilities, data flows, and remote actions that users and reviewers would not reasonably expect from the declared skill, weakening consent and review boundaries. In a plugin/agent ecosystem, hidden or out-of-scope functionality increases the chance of unauthorized data submission or misuse.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
86% confidence
Finding

The skill declares network and environment-variable requirements but does not define an explicit tool scope such as allowed tools or permissions. In an agent runtime, that can let the skill invoke broader capabilities than users or orchestrators expect, increasing the chance of unintended network access or secret exposure through overly permissive execution.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The document gives conflicting instructions about whether wait-music.js should be the default execution path. Because wait both submits paid work and polls, inconsistent guidance can cause accidental duplicate charges, longer-running network activity, or use of the wrong execution mode in agent environments.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Forcing a default male vocal timbre without user opt-in bakes a sensitive preference choice into the workflow. In generative media systems this can lead to biased or unwanted outputs, and in some contexts it may misrepresent user intent or embed demographic assumptions without consent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The documentation both permits local file paths for reference_audio and later says not to use local file paths. For a skill that can upload local files automatically, that contradiction is dangerous because an agent may exfiltrate local files to a remote server based on ambiguous guidance.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Repeating a default male timbre in guidance reinforces a demographic default that users did not choose. This is risky in context because the skill is designed to automate content generation, so the bias can be systematically applied at scale across many requests.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Telling the agent to preserve a male default when the user has no preference encodes a systematic bias into automated behavior. While not a classic exploit, it is a genuine policy and safety issue because the agent will repeatedly make identity-related output decisions on behalf of users.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The reference documents a webhook_url callback field but does not warn that task status and generated result metadata may be transmitted to an external URL controlled by the caller. In a music-generation skill, prompts, lyrics, result URLs, and related task data may be sensitive or proprietary, so omitting an explicit warning increases the risk of unintended data exfiltration or privacy violations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The normalization tables explicitly support English and Chinese labels, including multiple Chinese literals for genre, emotion, instrument, and timbre. Under the policy criteria, locale-specific behavior should either offer user choice or be clearly justified; this file embeds locale handling but provides no natural-language indication of opt-in or scope.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Speaker discovery is not justified by the stated music-generation purpose, so it represents an unnecessary hidden capability. While it does not directly show code execution, injection, or credential theft in this snippet, hidden enumeration features are risky because they can enable unreviewed data access patterns and indicate the package contains unrelated vendor functionality.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The code sends user-provided text to an external API endpoint without any visible in-file disclosure, confirmation step, or data-sensitivity guard. If users provide private prompts, scripts, or proprietary material, the skill can exfiltrate that content to a third party in a way that may not be obvious from the skill's declared music-focused purpose.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The Supported Intents section says a generic music request should default to ONLY_MUSIC, while multiple later sections state the default mode is VOCAL_SONG and instruct operators to start there when unclear. This is a direct contradiction in intended routing behavior, not merely omitted detail.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The function sends the generated text request body to a remote API endpoint, which may include user-provided query and script content. In this file, there is no confirmation prompt, warning, or user-facing disclosure before transmitting that data over the network.

Content

No source excerpt is available for this finding.