peach-e2e-run
Pass
Audited by Gen Agent Trust Hub on Apr 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local shell scripts such as
./e2e.shand./e2e/pwc.shto manage environment setup and run testing scenarios. - [EXTERNAL_DOWNLOADS]: The environment setup process (
./e2e.sh setup) involves the automated installation of external components, including Chrome Beta, agent-browser, and playwright-cli. - [REMOTE_CODE_EXECUTION]: The skill makes extensive use of
agent-browser evalandplaywright-cli evalto execute dynamically generated JavaScript code within the context of a web browser. - [DATA_EXFILTRATION]: The instructions enable the agent to extract and potentially report information from any open browser tab, including sessions where the user may be authenticated to sensitive services.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests data from external websites through browser evaluation tools without implementing security boundaries or sanitization.
- Ingestion points: Browser tab content and DOM structures accessed via
agent-browser eval. - Boundary markers: None identified in the provided instructions.
- Capability inventory: Execution of local shell scripts and arbitrary JavaScript evaluation in the browser.
- Sanitization: No evidence of input validation or escaping for data retrieved from web pages.
Audit Metadata