The Agent Skills Directory

[COMMAND_EXECUTION]: The workflow executes a local script using the bun runtime (Tools/ValidateDeck.ts) to validate the generated HTML output.
[DATA_EXFILTRATION]: The skill performs a network operation by sending a POST request to http://localhost:31337/notify via curl to announce workflow execution. While targeting a local address, this constitutes an external network call.
[DYNAMIC_EXECUTION]: The internal validation tool Tools/ValidateDeck.ts utilizes new Function(script) to verify the syntax of JavaScript embedded within the generated presentation files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text data (Orgmode, Markdown, or plain text) from local files or URLs to generate slides, creating an injection surface.
Ingestion points: External text files or web content fetched at runtime (documented in Workflows/Generate.md).
Boundary markers: The instructions explicitly mandate the use of functional replacers (() => value) and HTML escaping to prevent template injection attacks (documented in RenderingSpec.md).
Capability inventory: Local file writing to ~/Downloads, shell command execution via bun, and local network notifications via curl.
Sanitization: Employs escapeHtml and safe JSON serialization to sanitize user-provided content before it is embedded in the HTML template.

ljg-present