The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (validate_figure_spec.py, validate_render_audit.py, and clean_image_metadata.py) to perform essential validation and post-processing tasks. These scripts verify the integrity of the figure's scientific topology and strip sensitive metadata from the final image files.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content such as URLs, PDFs, and software repositories to extract scientific architecture. While this presents a potential surface for indirect prompt injection, the skill employs strong mitigations by requiring the agent to translate evidence into a structured FigureSpec that is validated by a script before rendering. The subsequent RenderAudit protocol provides a final check to ensure the generated image accurately reflects the evidence-backed plan.
[SAFE]: No malicious activity, such as unauthorized data exfiltration, credential harvesting, or prompt injection, was found. The skill demonstrates a security-conscious design by using absolute paths for artifact management and validating that references are regular files contained within a trusted workspace root.

fig-draw