The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/with_server.py executes arbitrary shell commands provided via command-line arguments. It uses subprocess.Popen with shell=True to start server processes and subprocess.run to execute the main automation command. While intended for local development workflows, this capability can be abused if the agent is influenced by malicious input.
[PROMPT_INJECTION]: The skill involves using Playwright to interact with web applications, creating a surface for Indirect Prompt Injection. The agent ingests untrusted data from the DOM or console logs, which could contain hidden instructions aimed at the agent. Evidence: (1) Ingestion points: Playwright's page.goto(), page.content(), and page.on("console", ...) in the examples. (2) Boundary markers: None explicitly implemented to separate web content from agent instructions. (3) Capability inventory: File system writes (screenshots/logs) and shell command execution via scripts/with_server.py. (4) Sanitization: No evidence of content sanitization before processing discovered elements or logs.

webapp-testing