The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to prompt injection because it interpolates user-controlled strings directly into the AI system prompt without sanitization. In scripts/generate_schematic_ai.py, the user_prompt is appended to a block of guidelines, which could allow an attacker to override scientific instructions.
Ingestion points: The prompt command-line argument in scripts/generate_schematic.py.
Boundary markers: Absent.
Capability inventory: File system write (images and JSON logs), network requests to OpenRouter API, and subprocess execution.
Sanitization: None; the input is used verbatim.
[COMMAND_EXECUTION]: The wrapper script scripts/generate_schematic.py uses subprocess.run to invoke internal Python scripts with arguments derived from user input. Additionally, documentation in references/README.md and references/QUICK_REFERENCE.md encourages users to manually modify shell profiles like ~/.bashrc to store API keys.
[EXTERNAL_DOWNLOADS]: The skill requires the requests library and performs network operations targeting the OpenRouter API at https://openrouter.ai/api/v1 to generate and review scientific schematics.
[DATA_EXFILTRATION]: The _load_env_file function in scripts/generate_schematic_ai.py recursively searches for .env files in parent directories up to five levels deep. This behavior can lead to the accidental exposure of sensitive environment variables if the skill is executed within a nested directory structure containing unrelated secrets.

scientific-schematics