The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data which could potentially contain malicious instructions if obtained from untrusted sources.
Ingestion points: The slidej generate command parses JSON input files, and the slidej parse command parses .pptx files (via src/parsers/index.js and src/cli.js).
Boundary markers: None are explicitly implemented; the tool processes the full content of provided files.
Capability inventory: The tool has file-read capabilities (used to embed local images via src/generators/index.js) and file-write capabilities (to output the resulting .pptx or .json files).
Sanitization: The generator uses XML escaping (src/utils/xml-helpers.js) to sanitize text content before embedding it into the presentation's XML structure.
[DATA_EXPOSURE]: The tool can read arbitrary local files if their paths are provided in the src attribute of an image element in the JSON input. While this is the intended mechanism for embedding local images into slides, it presents a data exposure risk if the agent is instructed to process a JSON file containing sensitive file paths.

slidej