The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and processing of external PowerPoint (.pptx) files from untrusted sources such as user uploads, local disk, and remote URLs, which introduces a surface for indirect prompt injection attacks.
Ingestion points: The Presentation.open(data) method documented in references/loading-and-saving.md is designed to process external binary data.
Boundary markers: No delimiters or specific instructions for the agent to ignore content-based instructions are provided in the documentation.
Capability inventory: The skill demonstrates capabilities for file reading (fs.readFileSync), file writing (fs/promises.writeFile), and network fetching (fetch).
Sanitization: There is no mention of sanitization or structural validation for the content within the PowerPoint files.
[METADATA_POISONING]: The file references/loading-and-saving.md contains extraneous text that appears to be internal logs from a development tool, including a specific local file path (file:///d:/FeatureAnalysis/...) from the developer's environment. While likely an accidental artifact of the documentation process, it constitutes an information leak and a failure of metadata integrity.

syncfusion-javascript-powerpoint