neta-suggest
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses
npx -yto download the@talesofai/neta-skillspackage from the NPM registry at runtime. This dependency is not pinned to a specific version, allowing potentially unverified code to be fetched during execution. Evidence: Found throughoutSKILL.mdin commands likenpx -y @talesofai/neta-skills@latest. - [REMOTE_CODE_EXECUTION]: The use of
npxto fetch and immediately run a package constitutes remote code execution, as the package content is hosted externally and executed in the local environment. Evidence: Commands such asnpx -y @talesofai/neta-skills@latest suggest_keywords. - [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands, including environment variable configuration and file system operations like writing to and reading from temporary files in
/tmp. Evidence:DEBUG=* npx ...andnpx ... > /tmp/categories_level1.jsonin the optimization and debugging sections. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data and interpolates it into shell command arguments, creating an attack surface for command injection and prompt injection.
- Ingestion points: User-supplied strings enter via
--prefix,--keyword, and--search_keywordsparameters inSKILL.md. - Boundary markers: No delimiters or sanitization warnings are present to isolate user input from the command structure.
- Capability inventory: Subprocess execution via
npxand shell redirection. - Sanitization: No sanitization or escaping mechanisms are documented for the interpolated inputs.
- Ingestion points: User-supplied strings enter via
Audit Metadata