The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions recommend that the agent download a favicon from a user-specified or brand-related URL (e.g., https://company.com/favicon.ico) to the local /tmp directory. This pattern involves retrieving content from untrusted remote sources determined by the user's prompt.
[COMMAND_EXECUTION]: The skill provides a command-line workflow for the agent to execute, using curl to download a remote file followed by a python3 script to process that file using the PIL (Pillow) library. While intended for extracting brand colors, processing untrusted remote files with image libraries can be exploited if the library has unpatched vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The generated browser extension is designed to ingest and scrape data from LinkedIn profiles (name, headline, photo). This creates an indirect prompt injection surface where a malicious profile could attempt to influence the extension's behavior or output, although the script includes escapeHtml to mitigate basic injection attacks.
[SAFE]: The extension template demonstrates good security practices by storing API keys in chrome.storage.local and restricting sensitive API calls to the background service worker, ensuring credentials are not exposed to the content script or the web page DOM.

browser-extension-builder