prototype
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes building prototypes that process untrusted data from external sources to drive application state and navigation.
- Ingestion points: In
LOGIC.md, the prototype is instructed to read keystrokes or lines of input from the terminal (Read one keystroke (or one line) at a time). InUI.md, the prototype is instructed to read thevariantvalue directly from URL search parameters (searchParams.get('variant')). - Boundary markers: Absent. The skill instructions do not require the generated prototype code to include delimiters or instructions to ignore embedded commands within the captured input.
- Capability inventory: The skill defines capabilities to
execute the core process, and the generated prototypes are capable of state mutation and conditional UI rendering based on the ingested data. - Sanitization: Absent. The skill lacks instructions for validating or sanitizing the input captured from the URL or terminal before it is used to determine which code paths are executed or how the state is mutated.
Audit Metadata