The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (URLs, uploaded documents, meeting transcripts) and interpolates this content into prompts for secondary LLM calls and HTML generation without sanitization or boundary markers.
Ingestion points: SKILL.md (Step 3) acquires content via web_fetch, pdfplumber, pandas, and from /mnt/transcripts/.
Boundary markers: Absent. Content is directly embedded into the design_prompt string in mode-infographic.md and into HTML templates in mode-publish.md and mode-visualize.md.
Capability inventory: Includes tool execution (Excalidraw:create_view, HeyGenverse:create_app, visualize:show_widget), file writing (/mnt/user-data/outputs/), and external network access (Google Gemini API).
Sanitization: No validation or escaping of the ingested content is performed before processing.
[COMMAND_EXECUTION]: The mode-infographic.md reference uses an embedded Python script to process information, interact with an external API, and perform file operations in the local environment.
[EXTERNAL_DOWNLOADS]: The skill performs several network operations across different modes:
Fetches content from user-provided URLs via web_fetch in SKILL.md.
Communicates with Google's official Gemini API endpoint to generate visual content in mode-infographic.md.
Loads the Chart.js library from the Cloudflare CDN in mode-publish.md and mode-visualize.md.

viz