The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from Markdown files and vision model responses, creating a surface for indirect instructions to influence the agent.
Ingestion points: The assets/build-deck.mjs script reads the deck.md source file; the assets/deck-review.mjs script ingests answers.json, which contains text transcriptions provided by external agents or vision models.
Boundary markers: The skill does not implement explicit delimiters or instruction-blocking warnings when processing these external data sources.
Capability inventory: The skill uses execFileSync to run the d2 diagramming tool and playwright to capture browser screenshots of the generated slides.
Sanitization: The build-deck.mjs script includes an inline function that performs HTML entity escaping on prose text before embedding it into the slide HTML, which provides basic protection against browser-based injection.
[COMMAND_EXECUTION]: The skill executes external command-line tools to perform its primary functions.
The assets/build-deck.mjs script uses execFileSync to invoke the d2 diagramming tool to render diagram fences into SVG format.
[DYNAMIC_EXECUTION]: The skill uses dynamic loading for optional dependencies.
The assets/deck-review.mjs script performs a dynamic import("playwright") when the screenshot functionality is triggered via the --out flag.

d2-slides