The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest third-party web content, creating a potential surface for indirect prompt injection. However, it includes a comprehensive 'Untrusted content boundary' specification that instructs the agent to treat all retrieved data as non-instructional, use secure delimiters (<EXTERNAL_DATA>), and ignore any embedded requests to run tools or reveal secrets.
[COMMAND_EXECUTION]: Includes Python code that uses the subprocess module to interact with a local ArchiveBox installation. The implementation follows security best practices by passing arguments as a list rather than a shell string to prevent command injection.
[EXTERNAL_DOWNLOADS]: Provides instructions to download a configuration file from the official archivebox.io domain and install the archivebox package via standard registries (pip). These are well-known and trusted sources for this specific archiving tool.
[DATA_EXPOSURE_AND_EXFILTRATION]: Performs network requests to well-known web archiving services including the Internet Archive, Archive.today, and Perma.cc. The code includes support for authentication keys (such as IA S3 keys or Perma.cc API keys) as parameters but does not contain hardcoded credentials.

web-archiving