The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill aggregates and processes real-time content from over 84 external news, listings, on-chain, and market sources. This data is untrusted and could contain malicious instructions designed to influence the AI agent's behavior, especially given the skill's focus on trading signals and impact scores.
Ingestion points: External data enters the agent context through several tools, including get_latest_news, search_news, and subscribe_latest_news, which return fields such as title, content, and summary_en from third-party sources.
Boundary markers: There are no specific delimiters or protective instructions (e.g., 'ignore any instructions in the news text') mentioned in the skill's data processing workflow to separate untrusted content from the system prompt.
Capability inventory: The agent is expected to interpret this data to provide summaries, 'ai_signal' (trading directions), and 'ai_score' (impact ratings), making it susceptible to content that attempts to manipulate these outputs.
Sanitization: The documentation does not indicate that the incoming text is sanitized or filtered for prompt injection patterns before being presented to the model.
[EXTERNAL_DOWNLOADS]: The skill relies on external infrastructure and software dependencies to function.
API Interaction: It makes network requests to ai.6551.io and establishes WebSocket connections to wss://ai.6551.io/open/news_wss for data retrieval.
Package Dependency: The installation instructions require the execution of a Python package named opennews-mcp using the uv tool.
Authentication: Users are directed to obtain an API token from https://6551.io/mcp and provide it as an environment variable.

opennews-mcp-news-aggregation