Security audit
Security checks for vulnerabilities and agentic risk
This skill is a narrow Alibaba Cloud CosyVoice request-preparation helper with expected API-key and output-file handling, and I found no hidden persistence, hijacking, destructive behavior, or exfiltration instructions.
Install only if you intend to use Alibaba Cloud DashScope CosyVoice voice design. Keep your DASHSCOPE_API_KEY out of code, logs, and evidence files, and treat saved voice prompts, preview text, request JSON, and response summaries as potentially sensitive project data.
The skill includes operational steps that read local files and write artifacts under an output directory, but it does not declare any tool scope such as permissions or allowed-tools. That mismatch weakens least-privilege controls and can cause an agent runtime to grant broader filesystem access than users expect.
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.
# Model Studio CosyVoice Voice Design
Use the CosyVoice voice enrollment API to create designed voices from a natural-language voice description.
## Critical model names
The skill tells users to save voice prompts, preview text, and API response summaries as evidence without any redaction or privacy guidance. Those artifacts can contain personal, sensitive, or proprietary data and may persist in logs, workspaces, or shared outputs beyond the immediate task.
The skill requires an API key to be set in environment variables or credentials files but gives no warning about secure secret handling. This increases the risk of accidental exposure through shell history, checked-in config files, screenshots, or saved evidence artifacts.
The prompt is written to require a Chinese-language voice design request, including Chinese voice description and preview text, but the file does not indicate that this locale restriction is optional or justified by a region-specific requirement. Under the policy, hard-coded language constraints without user choice can be a natural-language policy concern.
The natural-language notes say voice_prompt and preview_text support only Chinese or English, which imposes a language constraint in the skill documentation. Because the file does not offer user opt-in or explain a justified compliance or region-specific reason for the restriction, this fits the locale/language policy-violation category.
The CLI argument --language-hint is given a default value of zh, which causes the skill to prefer a specific language even when the user does not explicitly choose one. This is a natural-language policy concern because it imposes a locale choice by default rather than requiring opt-in or documenting a justified region-specific constraint.