The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from untrusted social media platforms and online articles.
Ingestion points: Fetches content via xhs.get_note, wechat_article.py, xreach, reddit.read_post, bilibili.get_video_info, and r.jina.ai as defined in SKILL.md.
Boundary markers: There are no boundary markers or instructions to ignore embedded commands within the fetched content.
Capability inventory: The skill has the ability to write to the user's Zotero library, execute shell commands, and perform network requests.
Sanitization: No sanitization, filtering, or validation logic is defined for the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill workflow involves executing shell commands that interpolate external URLs.
Evidence: Instructs the execution of python3 ~/.agent-reach/.venv/bin/wechat_article.py "URL", xreach tweet URL --json, and curl -s "https://r.jina.ai/URL". These patterns represent a command injection surface if the URL parameters are not safely escaped by the underlying platform.
[EXTERNAL_DOWNLOADS]: The skill utilizes the well-known service Jina Reader (r.jina.ai) to fetch and convert web content into markdown format for processing.

social-media-paper-triage