The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface through the ingestion and processing of external research publications.
Ingestion points: The skill fetches research paper content, abstracts, and metadata from external websites (such as arXiv or general web search results) via the web_fetch tool.
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" boundary markers to separate the untrusted external paper content from the agent's visualization generation logic.
Capability inventory: The skill uses the show_widget tool to render HTML, SVG, and JavaScript. It also utilizes sendPrompt() to allow user-driven (or data-driven) follow-up queries, which could be exploited by malicious content in a paper.
Sanitization: While the skill demonstrates awareness of the platform's Content Security Policy (CSP) regarding inline event handlers, it does not explicitly instruct the agent to sanitize or escape the research data before interpolating it into the widget code.
[DYNAMIC_EXECUTION]: The skill dynamically generates HTML, SVG, and JavaScript code at runtime based on the analysis of external papers. This code is executed via the show_widget tool to provide interactive simulations and architecture diagrams.
[EXTERNAL_DOWNLOADS]: The skill references and fetches the Chart.js library from Cloudflare's official CDN repository (cdnjs.cloudflare.com) to support quantitative comparisons in its interactive widgets.

research-survey-visualizer