The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from web pages (DOM, console logs) and documentation via tools such as extension_inspect, extension_logs, and extension_docs_search. Ingestion points include the analysis of web page DOM and console output in the extension_inspect and extension_logs tools. These surfaces are protected by clear boundary markers, specifically 'Core Rule 17', which mandates that the agent treat captured page content as untrusted evidence and forbids following instructions contained within that data. The skill's capability inventory includes file writing, CLI command execution, and network operations, but sanitization instructions require the agent to use captured data for behavior analysis only, effectively isolating it from the agent's command flow.\n- [DYNAMIC_EXECUTION]: The skill facilitates the evaluation of code within browser contexts to support extension debugging. Tools like extension_eval allow for code execution in page or content contexts. This is a standard and necessary feature for development environments and is documented as a debugging tool.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves project templates and configuration from official vendor-controlled sources. Project scaffolding utilizes templates from registry.extension.land and the extension-js/examples GitHub repository. These downloads are restricted to official framework resources and are documented neutrally as core functionality.

extension-dev