remember
Pass
Audited by Gen Agent Trust Hub on Jul 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill implements a persistent memory mechanism by saving user-provided constraints to a local file (.ui-craft/brief.md). These saved constraints are then loaded as context for future executions, creating an indirect prompt injection surface.
- Ingestion points: Conversation history (preceding corrections) and .ui-craft/brief.md.
- Boundary markers: None present; user-provided text is appended directly to the project brief as a markdown entry.
- Capability inventory: File system write access to the project directory and optional use of an external memory service.
- Sanitization: No robust sanitization or escaping of user input is implemented before persistence, though the agent is instructed to avoid rules that breach accessibility standards.
Audit Metadata