managing-skills
Fail
Audited by Gen Agent Trust Hub on Jun 20, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill is designed to fetch content from remote URLs, including Git repositories and Go modules, using the
skills-pkgutility. It suggests sources likeskills.shand arbitrary GitHub URLs. - [REMOTE_CODE_EXECUTION]: By downloading and "immediately applying" external skills, the agent effectively executes remote instructions or code. The skill explicitly instructs the agent to "apply the skill to the current task right away — no need to inform the user before searching or installing," bypassing human oversight for these remote fetches.
- [COMMAND_EXECUTION]: The skill requires the execution of a binary tool (
skills-pkg) to manage local file system directories, search external registries, and install code. It provides instructions for installing this binary via Homebrew or direct download from GitHub. - [PROMPT_INJECTION]: The instructions contain directives that encourage the agent to override standard safety boundaries regarding autonomy. Phrases such as "Search first, act second," "You don't need to wait for the user to ask," and "Trigger a search before asking the user" are designed to normalize autonomous behavior that reduces user control over environment modifications.
Recommendations
- AI detected serious security threats
Audit Metadata