dineway-building-clone
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses Playwright to browse user-provided source URLs for reconnaissance, screenshot capture, and data extraction. It also includes an optional integration with
api.atlascloud.aifor generating visual asset fallbacks if the original assets are unrecoverable. - [COMMAND_EXECUTION]: The workflow involves executing local scripts bundled with the skill, such as
capture-source-screenshots.mjs(Node.js) andstitch-browser-screenshots.py(Python), to process screenshots and validate data models. These executions are constrained to local, skill-specific scripts. - [INDIRECT_PROMPT_INJECTION]: As the skill ingests content from external websites, it possesses an attack surface for indirect prompt injection. However, the skill implements robust mitigations, including:
- Sanitization: URLs are scanned for sensitive query parameters (auth, token, etc.) and redacted.
- Data Scanning: Extracted data is scanned using regular expressions (
SENSITIVE_DATA_PATTERN,PRIVATE_PII_PATTERN) to prevent the ingestion of credentials or PII. - Validation: Generated 'clone agents' must pass strict validation via
validate_clone_agent.pyandvalidate_data_model_artifacts.pybefore any implementation goal is created.
Audit Metadata