reproduce
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, such as bug tickets and live web pages. This creates a surface where malicious content in a ticket or on a website could attempt to redirect the agent's actions. \n
- Ingestion points: Ticket body, comments, and website content (text, console, network) accessed via Playwright. \n
- Boundary markers: Instructions require a predefined plan and explicit expected outcomes. \n
- Capability inventory: Local file system writes and shell command execution (npx playwright). \n
- Sanitization: Explicit rules to redact PII and avoid saving tokens or cookies. \n- [DYNAMIC_EXECUTION]: The skill generates and executes Playwright test scripts based on a provided template. \n
- Evidence: Phase 2 provides a TypeScript template for reproduction scripts that use fs and @playwright/test. \n- [COMMAND_EXECUTION]: The skill uses shell commands to run the reproduction harness. \n
- Evidence: Execution of npx playwright test in Phase 3.
Audit Metadata