The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external RSS feeds defined in feeds.json through the scripts/fetch-feeds.js script. This content is then interpolated into a prompt for Gemini to synthesize a newsletter digest. Maliciously crafted articles within the RSS feeds could attempt to influence the LLM's behavior or output.
Ingestion points: scripts/fetch-feeds.js fetches and parses external RSS/Atom feed items (title, content, excerpts).
Boundary markers: The prompt structure in SKILL.md uses instructions like 'Sources (use ONLY these — do not invent facts):' but lacks robust delimiters or 'ignore embedded instructions' warnings for the interpolated data.
Capability inventory: The agent executes Node.js scripts capable of performing authenticated network POST requests to a user-configured Ghost CMS instance (scripts/ghost-publish.js).
Sanitization: The fetch-feeds.js script performs basic HTML tag stripping using regex, but does not sanitize against prompt injection patterns within the text.
[COMMAND_EXECUTION]: The skill requires the agent to execute local Node.js scripts using the node command to fetch feeds and publish content. It also triggers npm install to manage dependencies.
[CREDENTIALS_UNSAFE]: The skill manages sensitive environment variables GEMINI_API_KEY and GHOST_ADMIN_KEY. While the use of a .env file is standard practice, the scripts/ghost-publish.js script processes the GHOST_ADMIN_KEY (formatted as key_id:secret) to generate JWT tokens for administrative access to the user's Ghost instance.

newsletter-digest