webapp-testing
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/with_server.pyutility executes shell commands provided as arguments to start and manage web server processes during automation tasks. - [DYNAMIC_EXECUTION]: The script
scripts/with_server.pyusessubprocess.Popenwithshell=Trueto execute server startup commands. While this supports shell-native features like command chaining (e.g.,cd backend && python server.py), it also introduces a risk of command injection if the input strings are influenced by malicious data. - [INDIRECT_PROMPT_INJECTION]: The skill's primary workflow involves the agent ingesting and analyzing rendered DOM content from web pages using Playwright. This creates an attack surface where instructions embedded in a web application's content could influence the agent's actions.
- Ingestion points:
page.content(),page.locator().all(), andinner_text()calls described inSKILL.mdand implemented inexamples/element_discovery.py. - Boundary markers: The instructions do not define clear boundaries or markers to separate untrusted web content from system instructions.
- Capability inventory: The skill possesses the ability to execute shell commands and manage subprocesses via
scripts/with_server.py, and can write files such as screenshots and logs (as seen inexamples/console_logging.pyandexamples/static_html_automation.py). - Sanitization: There is no evidence of sanitization or filtering of the content retrieved from the browser before it is processed by the agent.
- [PROMPT_INJECTION]: Instructions in
SKILL.mdexplicitly discourage the agent from reading the source code of the provided scripts, suggesting they be treated as black boxes. While justified as a context-window optimization, this instruction could potentially hide implementation details from the agent's inspection.
Audit Metadata