The Agent Skills Directory

[PROMPT_INJECTION]: Vulnerability to indirect prompt injection from external web content. The skill utilizes agent-browser snapshot and agent-browser console to ingest data from user-specified target URLs. There are no defined boundary markers or instructions to isolate the website's content from the agent's internal logic, making it possible for a malicious site to influence the agent's behavior or trigger unintended actions.
[COMMAND_EXECUTION]: The skill uses shell commands (mkdir, cp, agent-browser) to manage the testing environment and report generation. While these commands are necessary for the skill's stated purpose, they provide a capability that could be abused if the agent's instructions are overridden by content encountered during the testing process.

dogfood