natural-language-generation
Pass
Audited by Gen Agent Trust Hub on Apr 4, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill documentation describes a workflow that ingests natural language requirements from users to generate complex code structures, including component logic, interaction tests, and stories. This characterizes a Category 8 (Indirect Prompt Injection) attack surface.
- Ingestion points: User-provided component descriptions in plain English, as described in the
nameanddescriptionfields of the SKILL.md frontmatter. - Boundary markers: The provided documentation does not define specific boundary markers (like XML tags or delimiters) to isolate user input from the generation instructions, although the workflow uses an intermediate JSON structured requirement step.
- Capability inventory: The skill (via the
component-generator-agent) generates TypeScript implementation code, React hooks, and Storybook interaction tests ('play functions') which are executable within a development environment. - Sanitization: While the documentation mentions that the agent 'Validates output' and the
accessibility-remediationskill 'Auto-fixes issues', there is no explicit description of input sanitization to prevent adversarial instructions from influencing the code generation logic.
Audit Metadata