The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses dynamic code loading and injection patterns at runtime. In assets/animations/fx-runtime.js, the script dynamically generates paths and appends <script> tags to the document head to load canvas animation modules from the local assets/animations/fx/ directory. In assets/runtime.js, the skill utilizes document.write() to inject a programmatically generated HTML string, which includes scripts for navigation and window synchronization, into the presenter mode popup window.
[COMMAND_EXECUTION]: The skill includes shell scripts for file scaffolding and slide rendering. The scripts/render.sh utility executes Google Chrome in headless mode with the --no-sandbox flag to take screenshots of local HTML slide files. The scripts/new-deck.sh script uses sed to create new presentation instances by rewriting asset paths in templates.
[EXTERNAL_DOWNLOADS]: The skill references external assets from well-known service providers. It imports web fonts from fonts.googleapis.com and loads the chart.js and highlight.js libraries from the jsdelivr.net CDN in multiple template files.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes untrusted user data to generate presentation content. 1. Ingestion points: User-provided content and notes are interpolated into HTML templates via the agent. 2. Boundary markers: Use of <section> tags for slides and <aside class="notes"> for speaker notes provides structural separation. 3. Capability inventory: The skill uses shell scripts to invoke a headless browser for rendering. 4. Sanitization: The skill does not perform explicit sanitization of interpolated HTML content.
[SAFE]: The skill contains example security policies in templates/full-decks/testing-safety-alert/index.html that display potentially destructive shell commands (e.g., rm -rf /) as static text for demonstration purposes; these are not executable code.

html-ppt