The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the bash tool to execute shell commands, specifically curl for fetching data and jq or python3 -c for parsing structured formats like JSON and XML.
[EXTERNAL_DOWNLOADS]: The skill routinely performs network requests to retrieve data from user-specified external URLs through WebFetch, WebSearch, and curl.
[REMOTE_CODE_EXECUTION]: The skill uses a javascript_tool to execute dynamically constructed JavaScript code in a browser environment to extract complex data structures from the DOM.
[PROMPT_INJECTION]: The skill exhibits a high susceptibility to indirect prompt injection due to its core functionality of ingesting and processing untrusted web data.
Ingestion points: Data enters the agent context through WebFetch, WebSearch, and browser navigation (navigate).
Boundary markers: The instructions lack robust boundary markers or "ignore instructions" warnings when passing raw fetched HTML or text to the AI for reconnaissance and extraction.
Capability inventory: The skill has significant capabilities including system command execution (bash), browser automation (computer tool), and the ability to save data to the local file system.
Sanitization: While the skill includes a "Transform" phase for cleaning data, this occurs after the data has already been processed by the agent in the "Recon" and "Extract" phases, meaning malicious instructions in the source content could influence the agent's behavior before they are cleaned.

web-scraper