The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with external web content, which presents a surface for indirect prompt injection. The instructions include a robust 'Untrusted content protocol' in references/sourcing-rules.md that explicitly requires the agent to treat fetched data solely as observation and never as executable instructions, mitigating the risk of embedded malicious commands.\n- [EXTERNAL_DOWNLOADS]: The skill references documentation and source code from the Playwright project's official repository for research purposes. These links serve as a basis for defining visual comparison standards and do not involve the execution of untrusted scripts.

ditto