The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text data and URLs from users to generate meme images and Markdown snippets, creating a potential attack surface where malicious instructions could be embedded in data processed by the agent.\n
Ingestion points: User-provided inputs for template, top_text, bottom_text, and custom style URLs (identified in SKILL.md and scripts/meme_generator.py).\n
Boundary markers: The skill does not implement specific delimiters or instructions to prevent the agent from interpreting text within the generated memes as instructions.\n
Capability inventory: The skill uses a Python script for URL construction and generates Markdown image links for display in the agent's output.\n
Sanitization: The provided meme_generator.py script employs urllib.parse.quote to escape special characters for URL safety, which mitigates common URL-based injection risks but does not address potential natural language instruction injection within the meme captions.\n- [EXTERNAL_DOWNLOADS]: The skill generates references to the api.memegen.link domain to fetch meme images. This is a well-known public service used for its intended purpose, and the references are documented neutrally.

meme-factory