Security audit
Security checks for vulnerabilities and agentic risk
This skill is a cloud video/audio transcription tool that uploads user-specified media to a remote service, and the inspected behavior is mostly disclosed and aligned with that purpose.
Install only if you are comfortable sending the selected video/audio files, media URLs, prompts, filenames, and API token-authenticated requests to the GuaiKei cloud service and its HTTPS upload storage. Avoid using it for confidential meetings, private recordings, or copyrighted media unless the provider's retention and privacy claims meet your needs.
The skill claims constrained behavior but apparently supports general network download flows, local file write/delete, resume, redirects, and retries. Those capabilities materially expand the attack surface: user-supplied URLs can lead to unreviewed remote fetching and local filesystem modification beyond a narrowly described transcription action.
The skill claims constrained behavior but apparently supports general network download flows, local file write/delete, resume, redirects, and retries. Those capabilities materially expand the attack surface: user-supplied URLs can lead to unreviewed remote fetching and local filesystem modification beyond a narrowly described transcription action.
The skill claims constrained behavior but apparently supports general network download flows, local file write/delete, resume, redirects, and retries. Those capabilities materially expand the attack surface: user-supplied URLs can lead to unreviewed remote fetching and local filesystem modification beyond a narrowly described transcription action.
The skill claims constrained behavior but apparently supports general network download flows, local file write/delete, resume, redirects, and retries. Those capabilities materially expand the attack surface: user-supplied URLs can lead to unreviewed remote fetching and local filesystem modification beyond a narrowly described transcription action.
Although the README mentions cloud processing and privacy claims, it does not prominently warn users at the point of use that submitted local videos and third-party links are sent to a remote service controlled outside the local environment. Users may mistakenly believe processing is local or equivalent to offline tools, leading to inadvertent disclosure of confidential recordings, meetings, or copyrighted content.
The README defines very broad natural-language trigger phrases and auto-conversion rules that can map ordinary user requests directly into a command that uploads local files or remote video URLs for cloud processing. This creates a prompt/skill overreach risk: users may invoke the skill unintentionally, causing unintended data transfer or processing of sensitive media.
Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.
4. 同时传入文件路径与任务ID，优先执行 `--id`，忽略 `--file`
5. 无自定义 prompt 时，默认完整转录视频全部文字
| 用户自然语言指令 | 自动生成命令 |
| -------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
| 视频提取 https://example.com/video.mp4 中的文字 | `node scripts/video2text/index.js --file "https://example.com/video.mp4"` |
| 把本地 /path/to/your/video.mp4 改成小红书风格的文案 | `node scripts/video2text/index.js --file "/path/to/your/video.mp4" --prompt "改写成小红书风格的文案"` |
The skill requires access to a sensitive environment variable (GUAIKEI_API_TOKEN) but does not declare any explicit tool scope or permissions boundary. That weakens reviewability and policy enforcement, because the runtime can access secrets without the manifest clearly constraining or documenting that capability.
The trigger description is extremely broad and covers many generic content-processing tasks, increasing the chance the skill activates for requests beyond simple speech-to-text. In an agent ecosystem, overbroad activation can cause unintended exfiltration of local files or remote media to a third-party service when a more limited or local tool would have been appropriate.
The guidance says to prioritize this skill when any matching intent appears, but the non-applicable cases are not strong enough to prevent over-selection. That can route broad analysis or writing tasks into a networked media-upload workflow, exposing data to a third-party service without sufficiently precise user intent.
Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.
当用户用自然语言下达指令时，按以下映射生成命令，保证识别与执行一致：
| 用户自然语言指令 | 生成的命令 |
| ---------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
| 提取 https://example.com/video.mp4 里的文字 | `node scripts/video2text/index.js --file "https://example.com/video.mp4"` |
| 总结这个视频的核心观点 https://example.com/video.mp4 | `node scripts/video2text/index.js --file "https://example.com/video.mp4" --prompt "总结这个视频的核心观点"` |
This code file contains multiple natural-language strings for errors, warnings, and help output entirely in Chinese, such as the module description, thrown errors, and generated CLI help text. Because the skill forces a specific language for user interaction without any opt-in or documented locale limitation, it violates the language/locale policy criterion.
This code builds and sends an HTTPS POST request containing arbitrary JSON payload data plus a TOKEN header. There is input validation and retry logging, but no confirmation prompt, disclosure message, or warning that user/system data and credentials will be transmitted to a remote host.
This code emits all user-facing status and warning text in Chinese only, including configuration guidance and recovery instructions. The file provides no indication of user opt-in, locale detection, or an alternative language path, which is a natural-language locale policy violation under the stated rules.
Multiple thrown error messages are hard-coded in Chinese, and the file provides no opt-in, locale selection, or justification for restricting output language. This can violate language/locale policy when the skill is expected to operate for users in other languages.
The banner text is hard-coded in Chinese, which imposes a specific language on all users regardless of their locale or preferences. This is a natural-language policy concern because the file provides no mechanism to choose or opt into the language.
No suspicious patterns detected.