The Agent Skills Directory

[SAFE]: The skill is well-structured and follows security best practices for internal tool execution and secret management.
[COMMAND_EXECUTION]: The script scripts/generate_schematic.py uses subprocess.run to call an internal helper script. This is implemented correctly using a list of arguments (shell=False by default), which ensures that user-provided diagram descriptions cannot lead to command injection.
[EXTERNAL_DOWNLOADS]: The skill interacts with the OpenRouter API (openrouter.ai) for its schematic generation feature. This is a functional requirement and targets a well-known service for LLM and image generation capabilities.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it is designed to analyze untrusted scientific data (Category 8).
Ingestion points: External scientific claims and research paper content provided by the user for critique.
Boundary markers: None explicitly defined in the prompt instructions for isolating external data.
Capability inventory: The skill can execute shell commands via scripts/generate_schematic.py (Bash) and has file system access (Write, Edit).
Sanitization: No specific sanitization or escaping is performed on the input data before processing, though the analytical frameworks provided guide the model toward objective critique, which mitigates simple injection attempts.

scientific-critical-thinking