krillinai-subtitle
Warn
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill's primary command template in
SKILL.mddirectly interpolates the<input>variable into a shell execution block:"$KRILLINAI_CLI" subtitle "<input>". While the input is enclosed in double quotes, it does not prevent subshell execution or command substitution if the input contains characters like backticks (`) or$(). This could allow a malicious input to execute arbitrary commands in the host environment. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from platforms like YouTube and Bilibili.
- Ingestion points: External subtitles, platform captions, and video metadata are fetched from remote URLs provided in the
<input>field (identified inSKILL.md). - Boundary markers: No delimiters or instructions are present to ensure the agent ignores malicious instructions that might be embedded in the downloaded captions or metadata.
- Capability inventory: The skill has the capability to execute a CLI tool (
$KRILLINAI_CLI) and perform file system operations (read/write) within the specified$WORKDIR(identified inSKILL.md). - Sanitization: There is no evidence of sanitization or validation of the content retrieved from external sources before processing.
- [EXTERNAL_DOWNLOADS]: The skill downloads media and caption data from well-known services including YouTube and Bilibili. This is part of the intended functionality and targets established platforms.
Audit Metadata