The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from various project files (e.g., mkdocs.yml, docs/about.md, book-metrics.json) and interpolates these strings directly into a generated Node.js script. While intended for slide content, this represents an attack surface for indirect prompt injection if the source files are malicious.
Ingestion points: mkdocs.yml, docs/about.md, docs/learning-graph/book-metrics.json, docs/img/mascot/character-sheet.md.
Boundary markers: None explicitly defined in the generation patterns.
Capability inventory: File writing, network access (via npm/pip), and shell execution (node, python3, bk-capture-screenshot).
Sanitization: No explicit sanitization or escaping of ingested text is mentioned before it is inserted into the template.
[DYNAMIC_EXECUTION]: The skill generates a local Node.js script (generate.js) at runtime using provided patterns in references/slide-patterns.md. This script is then executed to produce the .pptx file. This is a low-risk pattern as it uses static, well-defined templates from the skill itself.
[COMMAND_EXECUTION]: The skill executes several shell commands, including bk-capture-screenshot (from microsim-utils), python3 (for image cropping), node (to run the generator), and soffice (for optional PDF conversion). These are standard operations for the skill's primary purpose.
[EXTERNAL_DOWNLOADS]: The skill installs standard, well-known libraries pptxgenjs via npm and pillow via pip. These are legitimate dependencies for slide generation and image processing.

linkedin-carousel-generator