The Agent Skills Directory

[COMMAND_EXECUTION]: The skill mandates the use of a local Python script (scripts/generate_schematic.py) to generate scientific diagrams. The instructions label this as a mandatory step for completing a grant proposal, which involves executing code on the local system.
[EXTERNAL_DOWNLOADS]: The schematic generation tool (scripts/generate_schematic_ai.py) makes network requests to the OpenRouter API (openrouter.ai) to interact with image generation and language models.
[CREDENTIALS_UNSAFE]: The scripts require an OPENROUTER_API_KEY. While the skill correctly suggests using a .env file for secret management, the script logic also permits passing the API key as a command-line flag, which can expose secrets in process history or logs.
[INDIRECT_PROMPT_INJECTION]: The skill accepts arbitrary user input as a diagram description and incorporates it into a prompt sent to an external AI model.
Ingestion points: The prompt argument in generate_schematic.py provided by the user.
Boundary markers: The script uses a standard prompt template (SCIENTIFIC_DIAGRAM_GUIDELINES) but does not apply strict delimiters or escaping to the user-supplied description.
Capability inventory: The agent has access to Bash, Read, Write, and Edit tools, as well as the ability to execute the provided Python scripts.
Sanitization: No sanitization is performed on the user-provided prompt before it is sent to the external API.

research-grants