The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates explicit defensive instructions to mitigate indirect prompt injection. It mandates that the agent treat all external content (URLs, PDFs, or documents) as passive evidence and ignore any instructional commands, role-play requests, or system-bypass attempts found within those materials.
[COMMAND_EXECUTION]: The instructions provide guidance on using content-fetching tools such as Playwright or Puppeteer. It strictly enforces security constraints, such as requiring headless mode (headless: true) and prohibiting the use of graphical browser windows during execution.
[DATA_EXFILTRATION]: Although the skill accesses local Markdown files and external network resources, these actions are limited to its primary purpose of content verification. The skill instructions specifically forbid the output of privacy-sensitive information or credentials encountered during the auditing process.

content-checker