legal-client-brief
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
subprocessmodule to run internal Python scripts for validation and building reports. Inscripts/build_report.pyandscripts/render.py, commands are executed using argument lists withshell=False, which effectively prevents command injection. - [DYNAMIC_EXECUTION]: The skill performs dynamic content generation through Jinja2 templates and HTML-to-PDF conversion using the Playwright library. These operations are core to the skill's utility and are implemented using established libraries.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external industry news and legal data. Ingestion points include text retrieved by the agent's search capabilities. Boundary markers such as 'DRAFT' and '待律师复核' (Awaiting Lawyer Review) are used in templates. Capability inventory includes subprocess execution, file system access, and headless browser automation. Sanitization is handled through Jinja2's auto-escaping and a custom validation script that restricts URLs to a local whitelist.
- [EXTERNAL_DOWNLOADS]: The skill requires downloading standard Python and Node.js packages from official registries (PyPI, NPM) and browser binaries for Playwright. These are documented as necessary dependencies for its report-generation functionality.
Audit Metadata