The Agent Skills Directory

Data Ingestion Surface: The skill is designed to retrieve and format external content from sources like M365 searches, calendar views, and web searches (Step 1 in SKILL.md). This ingestion of untrusted data creates a potential surface for indirect prompt injection, where malicious instructions within the source material could attempt to influence the agent's behavior during the slide generation process. The skill provides grounding instructions and suggests using placeholders for missing data to maintain factual accuracy.
Self-Contained Execution Environment: The skill instructions (SKILL.md) and template (references/template.html) strictly enforce a self-contained architecture. It requires all CSS and JavaScript to be inlined and forbids the use of external CDN links or remote images. This design choice enhances security by preventing the loading of third-party resources at runtime.
Controlled Script Delivery: The navigation and theme engine is delivered as five small, self-contained JavaScript blocks. The instructions mandate that these blocks remain verbatim and split (Step 2 in SKILL.md) to comply with character limits and isolation policies in embedded preview environments like Microsoft Teams and Outlook. This granular approach ensures the logic remains simple and avoids the risks associated with large, consolidated script payloads.
Artifact and File Management: The skill utilizes host-specific artifact tools (CreateArtifact, EditArtifact) where available and implements manual versioning for file-based delivery on other platforms (Step 4a in SKILL.md). This ensures clear tracking of generated content and prevents data confusion during the iterative presentation creation process.

rich-html-presentation