The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from the user (such as topics, outlines, and research notes) to automate the creation of HTML presentations. This capability introduces a surface for indirect prompt injection, where malicious instructions embedded in the source material could attempt to influence the agent's behavior during the generation process.
Ingestion points: The skill reads data from conversation history, user-supplied arguments (e.g., the topic arg), and various local files in the working directory (markdown files, notes, research).
Boundary markers: The skill architecture requires the agent to confirm context and requirements with the user before building, which serves as a natural checkpoint.
Capability inventory: The agent generates self-contained HTML files with embedded JavaScript and links to external CDN-hosted libraries (such as Chart.js and GSAP) for animations and data visualization.
Sanitization: The SKILL.md file contains a specific 'Content sanitization' rule instructing the agent to escape <script>, </style>, and raw HTML tags in user-provided text to prevent document structure breakage or Cross-Site Scripting (XSS) within the generated deck.
[SAFE]: The skill utilizes well-known external libraries (e.g., Google Fonts, GSAP, Chart.js, Lucide) hosted on reputable Content Delivery Networks (CDNs) like fonts.googleapis.com, cdn.jsdelivr.net, and unpkg.com. These references are standard for web development and follow the established trust guidelines.

slides