The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interpolates raw user input into larger narrative prompt templates.
Ingestion points: The user_prompt argument in scripts/prompt_engine.py (via humanize_prompt) and scripts/generate.py.
Boundary markers: None. User-provided text is directly concatenated with system-level instructions in prompt_engine.py without delimiters or 'ignore embedded instructions' warnings.
Capability inventory: The resulting prompt is sent to Google's Generative AI API (generate_with_gemini and generate_with_imagen in scripts/generate.py). The skill also has the capability to write generated images and metadata to the local file system.
Sanitization: No sanitization or filtering is performed on the user-provided prompt before interpolation.
[EXTERNAL_DOWNLOADS]: The skill requires external dependencies and makes network calls to legitimate services.
The scripts/requirements.txt file specifies standard, well-known libraries: google-genai, Pillow, and python-dotenv.
The scripts/generate.py script communicates with Google's official AI Studio endpoints using the google-genai client library to generate images.
[COMMAND_EXECUTION]: The documentation provides instructions for local command execution to operate the skill.
SKILL.md instructs the user to execute local Python scripts (generate.py, prompt_engine.py, templates.py) and install dependencies via pip.
[METADATA_POISONING]: The SKILL.md frontmatter contains a risk: critical field.
While this appears to be a local configuration or classification field for the author, such high-severity markers in metadata can be used to mislead automated scanners or users about the actual safety profile of the skill.

ai-studio-image