skill-creator
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection (Category 8) within the
scripts/init_skill.pyscript. Untrusted user data, such as the skill name, is ingested and interpolated directly into script templates and file paths without sufficient sanitization or the use of boundary markers. This vulnerability exists alongside the skill's capabilities to create directories, write files, and modify file permissions. - [REMOTE_CODE_EXECUTION]: The documentation in
references/scripts.mdguides users to install theuvtool by executing a remote script fromastral.shvia a shell pipe. This represents a remote code execution pattern, though it originates from a well-known service domain. - [COMMAND_EXECUTION]: The initialization and packaging scripts perform various shell-equivalent operations, including the creation of directory structures, writing of Python and Markdown files, and the modification of file execution bits using
chmod. - [EXTERNAL_DOWNLOADS]: The skill's design promotes the integration of third-party libraries through the
uvpackage manager, which involves downloading software from public package repositories.
Audit Metadata