The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as research papers and reports to generate academic materials, which is an inherent surface for indirect prompt injection.
Ingestion points: SKILL.md (Section 2.1) and scripts/inject-math.py process text and metadata from tagged external files.
Boundary markers: The instructions in SKILL.md explicitly direct the agent to treat all source files as factual evidence and never as instructions.
Capability inventory: The skill utilizes subprocess to call the pandoc utility and has access to standard file and shell operations through the agent.
Sanitization: The scripts/sanitize-svg.py script provides a mechanism to sanitize external SVG assets by rejecting active code, event handlers, and external references.
[COMMAND_EXECUTION]: The skill uses standard academic utilities via shell commands for document conversion and testing.
Evidence: scripts/inject-math.py executes pandoc to convert LaTeX expressions into native Office Math Markup Language (OMML) within PowerPoint files.
Evidence: The testing framework in the scripts/tests/ directory uses subprocess.run to execute the skill's own Python scripts for functional validation.
[DYNAMIC_EXECUTION]: The skill employs dynamic loading for internal script modularization.
Evidence: scripts/check-visual-plan.py uses importlib.util to load the local sanitize-svg.py script. This targets a specific, local file within the skill's own directory structure for validation tasks.

rk-academic-visualize