The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a Python helper script (scripts/generate_image.py) to interact with the Novoads API, upload reference images, and download generated creatives. This script uses only standard Python libraries and performs expected API and file operations.
[EXTERNAL_DOWNLOADS]: The skill instructions provide links to the vendor's official GitHub repository (github.com/novoads/agent-skills) to fetch configuration files or prompting guides if they are missing from the local installation.
[CREDENTIALS_UNSAFE]: The skill manages the NOVOADS_API_KEY by reading from a .env file or environment variables. This follows standard local development practices for secret management and is documented clearly in the prerequisites.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a surface for processing untrusted user prompts and reference images. However, it implements mitigation strategies such as a mandatory prompt rewrite phase and the enforcement of safety suffixes (e.g., NO_CHROME_SUFFIX, SAFE_ZONE_SUFFIX) to constrain the image generator's output. The capabilities are limited to vendor API communication and writing image files to a designated folder.
Ingestion points: User-provided seed prompts and reference image paths.
Boundary markers: The script automatically appends structural safety suffixes to the final prompt to ensure layout and content constraints.
Capability inventory: The script performs network POST/PUT requests to the Novoads API and writes PNG files to the ./generated directory.
Sanitization: Prompt instructions are rewritten by the agent for clarity and safety before being passed to the generation script.

nano-banana-image-ad