state-machine
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read existing component code and data hooks to identify UI states. This creates a surface for indirect prompt injection if the repository code being analyzed contains malicious instructions designed to influence the agent's behavior during the generation of the throwaway page.
- Ingestion points: The agent reads local source code files and data fetching hooks to determine possible component states (documented in SKILL.md, Section 2).
- Boundary markers: There are no specific delimiters or instructions provided to the agent to distinguish between the code it is auditing and the instructions it should follow.
- Capability inventory: The skill utilizes file-writing capabilities to create new project folders and routes, and file-deletion capabilities to remove them (documented in SKILL.md, Sections 3 and 'Remove it in one step').
- Sanitization: The skill does not define specific sanitization or validation steps for the code content it ingests before using it to inform the generation of new files.
- [DYNAMIC_EXECUTION]: The skill generates new executable content in the form of a 'throwaway page' (a new route or component file) which is then loaded within the application environment.
- Pattern: The skill involves generating new JavaScript/TypeScript code based on templates and existing project components and then directs the agent to verify the execution of this new code in a browser context (documented in SKILL.md, Section 6).
Audit Metadata