The Agent Skills Directory

[Indirect Prompt Injection] (MEDIUM): The skill possesses a vulnerability surface for indirect prompt injection via external data ingestion.
Ingestion points: The skill ingests untrusted content from the NewsArticle database via the get_recent_news function and the db.query in gather_quick_data.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the news processing logic.
Capability inventory: The agent generates a structured JSON response containing reasoning, action, and risk_factors which are presented directly to the user/UI.
Sanitization: No sanitization or filtering of news headlines or content is performed before the agent processes them for sentiment or summary generation.
[Data Exposure] (LOW): The skill accesses local database models (NewsArticle) and external market data APIs (YahooClient). While these are standard for the stated purpose, they represent a data ingestion path from potentially untrusted external sources (news providers).
[Unverifiable Dependencies] (LOW): The skill references internal modules such as backend.ai.skills.base_agent and backend.data.yahoo_client. These appear to be part of a larger, trusted local architecture rather than remote/untrusted downloads.

quick-analyzer-agent