The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from untrusted external sources, including live websites, browser tabs, and public design pattern libraries (Mobbin, Pinterest, Are.na, Cosmos), creating a surface for indirect prompt injection where malicious content could attempt to influence agent behavior.
Ingestion points: Public pattern libraries, browser tabs, and live websites as described in sources.md.
Boundary markers: The skill does not explicitly define delimiters or instructions for the agent to ignore instructions potentially embedded in the fetched design references.
Capability inventory: The agent can access local design files and screenshots, perform network requests to external platforms, and write HTML files to the local .gather/ directory.
Sanitization: There are no instructions for sanitizing or escaping the content retrieved from external sources before it is interpolated into the generated HTML board.
[DYNAMIC_EXECUTION]: The skill dynamically generates a local HTML file (.gather/) containing images, notes, and links from various sources. This involves the creation of executable content (HTML/CSS) from untrusted external data, which carries a potential risk of script injection if the user opens the file in a browser.
[COMMAND_EXECUTION]: The skill integrates with a local application named 'Pond' via the Model Context Protocol (MCP), which involves executing a specific binary at /Applications/Pond.app/Contents/MacOS/Pond with local arguments.
[EXTERNAL_DOWNLOADS]: The skill retrieves data and images from well-known design platforms including Mobbin, Are.na, Pinterest, and Cosmos. These interactions are core to the skill's functionality and use established services.

gather