prototype
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill guides the agent to create interactive terminal and UI components that ingest external data. 1. Ingestion points: The instructions in LOGIC.md involve reading keystrokes and terminal lines, while UI.md uses URL search parameters for switching variants. 2. Boundary markers: There are no instructions to use delimiters or warnings to treat embedded content as untrusted. 3. Capability inventory: Prototypes are executed using project-level task runners (pnpm, python, bun) and can interact with the local environment. 4. Sanitization: The instructions lack guidance on sanitizing these inputs before processing.
- [COMMAND_EXECUTION]: The skill requires the agent to add and execute scripts through standard task runners like package.json or Makefiles, involving the use of shell commands within the project context.
Audit Metadata