The Agent Skills Directory

[Prompt Injection]: The instructions utilize <system-reminder> tags containing directives to "ignore any existing tests" or "ignore tests entirely." While these are intended to guide the agent toward a specific debugging workflow, the use of "ignore/disregard" patterns is a common indicator of attempts to override standard operational behavior.
[Indirect Prompt Injection]: The skill involves reading web application frontend content and logs to identify bugs. This represents a potential attack surface where instructions embedded in the processed web data (such as HTML comments or log outputs) could influence the agent's actions.
Ingestion points: Browser interactions via Playwright (page.goto), reading server/UI logs, and analyzing screenshots.
Boundary markers: The skill does not define specific delimiters to distinguish between its own instructions and the data being analyzed.
Capability inventory: The skill can execute shell commands (npm run, python server.py), write configuration files, and install packages.
Sanitization: No explicit sanitization or validation of the webapp content is mentioned before it is processed by the agent.
[Dynamic Execution]: The skill operates by writing and then running Playwright scripts in Python or Node.js. The generation and execution of scripts at runtime is a security consideration, although it is the primary method for the skill's intended functionality of automated web testing.
[Command Execution]: The skill uses subprocess calls to start development servers and run tests. While these commands target the local environment for testing purposes, they demonstrate significant control over the execution environment.

webapp-testing