developing-genkit-go
Pass
Audited by Gen Agent Trust Hub on Apr 30, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches the Genkit CLI installation script from
cli.genkit.dev(e.g.,curl -sL cli.genkit.dev | bashinSKILL.md). - [REMOTE_CODE_EXECUTION]: Executes the downloaded CLI installation script directly in the shell. This is a standard but high-privilege operation for developer tooling.
- [PROMPT_INJECTION]: The skill documents patterns for interpolating untrusted user data directly into AI prompts, which creates an indirect prompt injection surface.
- Ingestion points: Untrusted data enters the agent context through flow arguments (e.g.,
topicinjokeFlowinSKILL.md) and prompt input types (e.g.,JokeRequestinreferences/prompts.md). - Boundary markers: Examples do not include boundary markers or explicit instructions to ignore embedded commands in the user-provided data.
- Capability inventory: The skill uses
genkit.GenerateText,genkit.Generate, andgenkit.DefineToolwhich can perform actions based on the potentially injected instructions. - Sanitization: No sanitization or validation of the input variables before interpolation into the prompt string is shown in the examples.
Audit Metadata