The Agent Skills Directory

Indirect Prompt Injection (LOW): The scripts component_generator.py and style_generator.py interpolate user-provided strings (like --description and --name) directly into code templates using simple string replacement.
Ingestion points: Command-line arguments --name, --description, and --props in the generator Python scripts.
Boundary markers: None. The input is directly placed inside JSDoc comments or code structures (e.g., `/** * ${COMPONENT_NAME}
${DESCRIPTION} /). An attacker could provide a description like / import { exec } from 'child_process'; /*` to break out of the comment and inject malicious code into the generated file.
Capability inventory: The generator scripts use open(args.output, 'w').write() to create new files on the local filesystem.
Sanitization: While name_validator.py exists as a standalone tool to check naming conventions, it is not integrated into the generation logic to sanitize input against injection characters (like comment delimiters or newline characters).
File System Access (LOW): The scripts component_generator.py, style_generator.py, and test_generator.py accept an --output argument that is used to write files without any path traversal checks or restrictions to a specific workspace directory. This could allow an agent to overwrite arbitrary files if the path is not controlled.

frontend-component