The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill takes untrusted user input via the prompt argument in scripts/generate_schematic.py and passes it to scripts/generate_schematic_ai.py, which then interpolates it into an LLM prompt for image generation and quality review. This creates a surface for indirect prompt injection.
Ingestion points: User-provided description in scripts/generate_schematic.py and scripts/generate_schematic_ai.py (CLI arguments).
Boundary markers: Absent; the user prompt is appended directly to the SCIENTIFIC_DIAGRAM_GUIDELINES template.
Capability inventory: Subprocess execution via subprocess.run (in generate_schematic.py), network access via requests.post (in generate_schematic_ai.py), and file system write access via open() (for saving images and logs).
Sanitization: No sanitization or filtering is applied to the user-supplied prompt before it is sent to the OpenRouter API.
[DYNAMIC_EXECUTION]: The wrapper script scripts/generate_schematic.py uses subprocess.run() to execute the internal AI generation script scripts/generate_schematic_ai.py. This is a standard architectural pattern for the skill's functionality, though it involves dynamic process spawning.
[COMMAND_EXECUTION]: The skill includes a bash script scripts/review_poster.sh that executes local system utilities (such as pdfinfo, pdffonts, pdfimages, and gs) to perform quality checks on generated PDF posters. These operations are restricted to the local environment and intended for PDF metadata analysis.

latex-posters