The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes potentially untrusted data and renders it as HTML.
Ingestion points: The assets/deck_engine.py script reads data from data.json and various _decision_*.json files to populate slide content.
Boundary markers: There are no explicit delimiters or instructions to the agent to ignore embedded commands within the ingested data files.
Capability inventory: The SKILL.md file defines access to the Bash, Read, Write, Edit, and Glob tools, which are used to execute the deck engine and PDF conversion utilities.
Sanitization: While a basic escape function is used for some fields, the prose_slide and essay_slide functions in assets/deck_engine.py interpolate content directly into HTML paragraph tags without sanitization, allowing for potential XSS or instruction injection.
[COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands for deck generation and quality control.
Evidence: SKILL.md details the usage of python3, headless Google Chrome for PDF printing, and utilities such as pdfinfo and pdftoppm via the Bash tool.
[EXTERNAL_DOWNLOADS]: The skill's HTML output template fetches resources from a well-known external service.
Evidence: The render function in assets/deck_engine.py embeds a stylesheet link to https://fonts.googleapis.com to load the 'Inter' font family for the presentation.

mckinsey-deck