mobile-ui-design
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions to modify project-wide configuration and instruction files used by AI tools. It directs the agent to update files such as
CLAUDE.md,AGENTS.md,.cursor/rules/*.mdc, and.github/copilot-instructions.mdwith imperative pointers (e.g., "Before implementing any screen, read..."). This behavior is intended to ensure future agent interactions are guided by the generated design document, effectively persisting specific instructions across sessions. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its data ingestion and file-writing capabilities.
- Ingestion points: The skill ingests data from external sources including the project brief, codebase, routes, plans, and screenshots (documented in
SKILL.md). - Boundary markers: While the skill attempts to preserve user decisions, it lacks explicit technical boundary markers or instructions to ignore malicious prompts embedded within the ingested data.
- Capability inventory: The skill performs file system writes to both documentation directories (
docs/design/) and critical project instruction files (CLAUDE.md,AGENTS.md, etc.). - Sanitization: There is no evidence of sanitization or validation of the content extracted from the project environment before it is propagated into persistent instruction files.
Audit Metadata