The Agent Skills Directory

[PROMPT_INJECTION]: The references/pitfalls.md document provides strategies for navigating content moderation filters on external image generation platforms. It suggests rephrasing descriptions to avoid triggering safety blocks (e.g., using neutral physical descriptions instead of emotional or 'threat-adjacent' vocabulary), which is a common prompt engineering technique to avoid false positives.
[DATA_EXPOSURE]: The scripts/generate-with-ref.py utility accesses configuration and environment files located in the ~/.hermes/ directory. It specifically reads the OPENAI_API_KEY from a local .env file to authenticate requests to the OpenAI API.
[COMMAND_EXECUTION]: The skill includes two Python scripts, generate-with-ref.py and postprocess.py, which are intended to be executed by the agent to perform image generation and manipulation. These scripts use standard libraries (Pillow, NumPy) for image processing and urllib for API communication.
[EXTERNAL_DOWNLOADS]: The skill makes network requests to api.openai.com to perform image edits. This is the intended primary purpose of the generation script and targets a well-known service provider.

nous-branding