screen-reader-lab
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external HTML and JSX code provided via file paths or snippets, creating an attack surface where malicious instructions could be embedded in the code content or attributes (e.g.,
aria-label). - Ingestion points: In Phase 1 and Phase 2 of
SKILL.md, the agent is instructed to ask for and read file paths or code snippets for simulation. - Boundary markers: The instructions do not specify any boundary markers or explicit instructions for the agent to ignore embedded commands within the analyzed data.
- Capability inventory: The skill relies on file-reading and parsing capabilities. While it does not explicitly define network or write operations, the parsed data influences the agent's narration and findings output.
- Sanitization: No sanitization, validation, or escaping of the external code input is mentioned in the simulation process.
Audit Metadata