The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by ingesting user-provided diagram descriptions and LLM-generated critiques into its internal prompt templates for iterative refinement.\n
Ingestion points: User-supplied diagram descriptions are passed as command-line arguments to scripts/generate_schematic.py and subsequently to scripts/generate_schematic_ai.py.\n
Boundary markers: The skill uses clear text labels (e.g., USER REQUEST:, ORIGINAL REQUEST:, ITERATION:) to delimit user or previous-step content within prompts.\n
Capability inventory: The skill can execute local Python scripts via subprocess.run, perform network POST requests to OpenRouter via the requests library, and write binary (PNG) and text (JSON) files to the local file system.\n
Sanitization: There is no evidence of specific input validation or sanitization of the natural language prompts before they are interpolated into the templates sent to the LLM.\n- [COMMAND_EXECUTION]: The tool uses subprocess.run in scripts/generate_schematic.py to launch its internal AI generation script. The execution uses a list-based argument format and targets a statically resolved path within the skill's own directory, which follows security best practices for process spawning.\n- [EXTERNAL_DOWNLOADS]: The skill communicates with OpenRouter's official API endpoints to generate images and conduct vision-based quality reviews. These network operations are transparently documented in the skill's instructions and are necessary for its primary function.

scientific-schematics