The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from social media platforms (Xiaohongshu, Douyin, WeChat) which serves as a vector for indirect prompt injection.
Ingestion points: The scripts/cn_last30days.py script fetches post titles, descriptions, and comments from external social media platforms via the RedFox API. Additionally, SKILL.md specifies using WebSearch to gather further data.
Boundary markers: The instructions do not define explicit boundary markers or XML-style delimiters to separate untrusted social media content from system instructions when presented to the agent for analysis.
Capability inventory: The skill has access to Bash, WebSearch, and Write tools, which could be abused if the agent follows malicious instructions embedded in the processed social media data.
Sanitization: While the Python script performs basic HTML entity escaping for report generation, the raw text content is passed to the agent's context for summarization and analysis without robust filtering for instructional patterns.

cn-last30days