The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches assets from several well-known and legitimate external sources to support its themes.
Downloads web fonts from Google Fonts (fonts.googleapis.com) and Alibaba CDN (at.alicdn.com).
Retrieves font assets from Apple's S3 storage (applesocial.s3.amazonaws.com).
Loads styles for code highlighting from Cloudflare's CDN (cdnjs.cloudflare.com).
[COMMAND_EXECUTION]: The skill utilizes shell commands via the Bash tool to manage the presentation project lifecycle.
Uses mkdir to initialize workspace directories for organization.
Invokes marp-cli via npx to convert Markdown sources into HTML, PNG, PDF, and PPTX formats.
Recommends the use of brew to install system packages like libreoffice for advanced export features.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection within its content analysis and review workflow.
Ingestion points: The skill reads untrusted user-provided content (PDFs, research papers, notes) in Stage 1 to generate presentation outlines.
Boundary markers: The instructions for subagents in Stage 3 (e.g., Content Review Expert) interpolate the processed content into prompts using simple placeholders like [插入presentation.md内容] without utilizing strong delimiters or safety warnings to ignore embedded instructions.
Capability inventory: The skill possesses capabilities including Bash (executing the Marp CLI), Write, and Edit (modifying the presentation code).
Sanitization: There is no explicit logic described for sanitizing or filtering input data before it is presented to the subagents for processing.

marp-slides-creator