The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external social media platforms which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Tools like twitter_web_trending, tiktok_app_v3_home_feed, reddit_app_popular_feed, and google_news_bulk_articles fetch content from public, user-generated sources into the agent's context.
Boundary markers: The instructions include guidelines to 'Separate factual observations from analysis' and 'Cite sources', which provide some structural separation, but there are no explicit data delimiters mentioned to prevent the agent from obeying instructions embedded in the retrieved text.
Capability inventory: The skill is limited to calling sandbase_describe_tool and sandbase_call_tool for research purposes. It does not possess file-writing, system command execution, or general network capabilities beyond the defined API tools.
Sanitization: There is no mention of filtering, escaping, or sanitizing the retrieved social media content before the agent processes it.
[NO_CODE]: The skill consists only of instructional markdown files and does not contain any executable scripts or binary files.

trend-spotter