The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it evaluates untrusted academic content which can influence subsequent agent actions, such as scoring and diagram generation. A malicious paper could include instructions to bias the evaluation or exploit the image generation loop.
Ingestion points: Research papers and proposals processed in SKILL.md.
Boundary markers: No delimiters or 'ignore' instructions are used when processing external documents.
Capability inventory: Execution of Python scripts via subprocess.run and API requests to OpenRouter for image generation.
Sanitization: No input filtering or validation is implemented for the analyzed content.
[COMMAND_EXECUTION]: The skill executes local Python scripts using subprocess.run to automate the evaluation workflow. Although implemented safely without a shell, this provides an execution surface for data-driven arguments.
[EXTERNAL_DOWNLOADS]: The skill communicates with openrouter.ai to perform AI-assisted schematic generation and review. This is documented as a well-known service and used for its primary purpose.

scholar-evaluation