The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill uses user-provided text (prompts, brand names) to generate SVG icons and brand assets via LLM APIs. This ingestion of untrusted data creates a surface for indirect prompt injection where a malicious user could attempt to override the generation logic. The risk is minimized by the use of structured system instructions and safety filters provided by the AI service.
Ingestion points: User arguments via CLI in scripts/logo/generate.py, scripts/cip/generate.py, and scripts/icon/generate.py.
Boundary markers: Prompts are formatted within instruction-based templates (e.g., SVG_PROMPT_TEMPLATE).
Capability inventory: The skill generates file content (SVG/HTML) and saves images to disk using standard Python file I/O.
Sanitization: Relies on the AI provider's native safety filters and configured safety_settings in the generation scripts.
[EXTERNAL_DOWNLOADS]: The skill downloads image assets from documented image generation providers (Atlas Cloud and MuAPI). The implementation demonstrates security maturity by including explicit URL validation logic to prevent Server-Side Request Forgery (SSRF) and unauthorized access to local or internal network resources.
Validation logic: The _validate_public_https_url function in scripts/logo/generate.py verifies HTTPS schemes, filters out local hostnames (e.g., localhost, .local, .internal), and ensures IP addresses are global/public.

design