The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a research phase using Perplexity Sonar to gather facts from the web, which are then interpolated into generation and review prompts in scripts/generate_infographic_ai.py. This creates an attack surface for indirect prompt injection.
Ingestion points: External data from perplexity/sonar-pro search results is stored in research_data['content'] and included in the prompt for the image generation model.
Boundary markers: The skill uses basic headers like RESEARCHED DATA AND FACTS (use these in the infographic): to separate research data from instructions, but lacks explicit directives for the model to ignore embedded malicious instructions within that data.
Capability inventory: The skill has capabilities to write files (images and JSON logs) to the local filesystem and perform network operations to OpenRouter APIs.
Sanitization: No sanitization or filtering is performed on the content retrieved from web searches before it is included in the prompt.
[COMMAND_EXECUTION]: The orchestration script scripts/generate_infographic.py uses subprocess.run to execute the local AI generation script generate_infographic_ai.py. This is implemented using secure practices (list-based arguments and a restricted environment), but it represents a dependency on local script execution.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to openrouter.ai to access Gemini and Perplexity models for research, image generation, and quality review. This is required for its primary functionality.

infographics