The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided PowerPoint files, which constitutes a surface for indirect prompt injection where malicious instructions could be embedded in the presentation text.
Ingestion points: The SKILL.md and references/editing.md files describe workflows for reading and analyzing external .pptx presentations using markitdown to inform slide outlines and requirements.
Boundary markers: The skill instructions do not specify explicit boundary markers or "ignore instructions" delimiters for the extracted text when it is provided to the agent.
Capability inventory: The skill utilizes shell commands (python, node, npm, pip), file system writing, and library-based network fetching for images.
Sanitization: The skill recommends using the defusedxml library in references/editing.md to safely handle XML parsing and mitigate XML External Entity (XXE) vulnerabilities.
[DYNAMIC_EXECUTION]: The skill employs a workflow where it generates JavaScript files for individual slides and then executes them at runtime to compile the final presentation.
Evidence: SKILL.md instructs the agent to create individual slide-XX.js files and then run a compile.js script that uses require() to load and execute these dynamically generated modules.
[EXTERNAL_DOWNLOADS]: The skill installs several external libraries to facilitate presentation generation, text extraction, and icon processing.
Evidence: SKILL.md lists dependencies including markitdown (from Microsoft, a trusted organization), pptxgenjs, sharp, and various react libraries via standard npm and pip installation commands.

pptx-generator