The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides detailed instructions and examples for running the crawlberg interact command to automate browser tasks.
[DYNAMIC_EXECUTION]: The executeJs action allows the agent to run arbitrary JavaScript code within the context of the loaded web page. The skill includes a warning to use only trusted scripts for this operation.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the automated ingestion and interaction with arbitrary external web content, which presents a surface for indirect prompt injection. If the agent processes instructions found within the scraped content, it could be induced to perform unintended actions.
Ingestion points: Data is ingested via the crawlberg interact <url> command and the scrape action in SKILL.md.
Boundary markers: None are defined to separate ingested content from system instructions.
Capability inventory: The tool includes powerful capabilities such as click, type, executeJs, and screenshot (documented in SKILL.md).
Sanitization: No specific sanitization or filtering of external content is mentioned.

automating-the-browser