prototype
Pass
Audited by Gen Agent Trust Hub on Aug 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill involves the agent reading user prompts and surrounding project code to determine the scope and requirements of a prototype. This presents a potential surface for indirect prompt injection where instructions embedded in the codebase could influence the agent's behavior. However, the risk is minimized by the skill's instructions to generate self-contained, isolated artifacts.
- Ingestion points: User prompts and existing project source code (SKILL.md).
- Boundary markers: None explicitly defined in the instructions to separate external data from agent instructions.
- Capability inventory: The skill facilitates file creation (writing prototype code) and command execution for running and committing prototypes (SKILL.md, LOGIC.md, UI.md).
- Sanitization: No specific mechanisms for sanitizing or validating ingested code or prompt content are provided.
- [COMMAND_EXECUTION]: The instructions reference the use of project-specific task runners and runtimes, such as pnpm, python, bun, and git, to manage and execute the generated prototypes. These are standard development operations and are not used for silent or malicious background activities.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill explicitly advises against production database persistence, suggesting the use of in-memory state or local scratch files instead. This is a secure prototyping practice that prevents accidental exposure of production data.
Audit Metadata