The Agent Skills Directory

[COMMAND_EXECUTION]: The provided Python helper script in SKILL.md uses subprocess.run to invoke the marp command-line utility. It correctly passes arguments as a list rather than a single string, which is a security best practice that prevents shell command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The installation section directs users to download the @marp-team/marp-cli package from the official NPM registry or via Homebrew. These are well-known, trusted package management services used to acquire the necessary conversion engine.
[PROMPT_INJECTION]: The skill processes user-provided Markdown and HTML content, creating an indirect prompt injection surface. A malicious input could attempt to include directives that exploit the Marp rendering engine or inject malicious CSS, although this is a common risk for conversion tools and is mitigated by the scope of the underlying rendering engine.
Ingestion points: Markdown/HTML content provided by the user is passed to the markdown_to_pptx function.
Boundary markers: None present in the example implementation.
Capability inventory: The skill uses file-system writes (temp files) and executes the marp binary via subprocess.
Sanitization: The script performs a basic check for Marp-specific frontmatter but does not sanitize the body of the markdown content.

html-to-ppt