The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/with_server.py is designed to execute arbitrary shell commands passed as command-line arguments. It uses subprocess.run() to execute the main command and subprocess.Popen() to start background servers.
[DYNAMIC_EXECUTION]: In scripts/with_server.py, the subprocess.Popen() call uses shell=True to execute server startup commands. This allows the execution of complex shell strings (e.g., using && or cd), but also significantly increases the risk of command injection if arguments are derived from untrusted sources.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external web pages during testing sessions. Specifically, examples/console_logging.py captures browser console messages (msg.text) and examples/element_discovery.py extracts text from page elements (inner_text()). If the web application under test is malicious or compromised, it could provide instructions to the agent through these channels.
Ingestion points: Browser console logs in examples/console_logging.py and page element text content in examples/element_discovery.py.
Boundary markers: None observed in the implementation to delimit untrusted page content from agent instructions.
Capability inventory: Execution of shell commands via scripts/with_server.py using subprocess.run and subprocess.Popen.
Sanitization: No evidence of sanitization or validation of the content extracted from the browser before it is processed or logged.

webapp-testing