cuawright-web
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Command Execution via Bash: The skill uses the
Bashtool to perform environment setup (e.g., installing Playwright) and to execute browser automation scripts. This is a core feature enabling the skill to interact with the local environment and drive web tasks. - Dynamic Script Generation and Execution: The skill's workflow involves generating Python scripts (
final_script.py) and executing them to perform multi-step automation. It includes automated checks, such as an import-safety smoke test, to ensure the generated scripts are well-formed. - Indirect Prompt Injection Surface: As the skill processes data from external websites (via ARIA snapshots and screenshots), it is exposed to untrusted content that could theoretically contain instructions designed to influence the agent's behavior. The skill's design includes a self-verification step against a predefined plan to help ensure task integrity.
- Scoped File System Operations: Instructions direct the agent to work within a specific
WORKSPACE_DIR, where all generated code, screenshots, and logs are stored. This practice helps manage the skill's impact on the file system by keeping task-related artifacts localized.
Audit Metadata