The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a function to disable SSL certificate verification globally within the execution context. This is achieved by monkey-patching the httpx library's client initialization and overriding the default SSL context in the Python ssl module. This feature is intended for environments with corporate proxies but can be exploited to facilitate man-in-the-middle attacks.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted content from user-provided prompts and Google Search results.\n
Ingestion points: The prompt argument in scripts/generate.py and scripts/batch_generate.py, and the output of Google Search grounding when enabled via the --search flag.\n
Boundary markers: There are no protective delimiters or specific instructions provided to the underlying model to distinguish between the developer's instructions and potentially malicious instructions embedded in the prompt or search data.\n
Capability inventory: The skill possesses the ability to read local image files (scripts/generate.py), write generated images to the local file system (scripts/generate.py), and communicate with external APIs (Google Gemini).\n
Sanitization: The skill does not perform any validation, escaping, or sanitization of the prompt strings before they are passed to the Gemini API.\n- [COMMAND_EXECUTION]: The skill provides Python scripts (generate.py, batch_generate.py) that take arbitrary string prompts and file paths as arguments. This allows the agent to perform file system operations and network requests based on model output.\n- [EXTERNAL_DOWNLOADS]: The documentation instructs users to install external Python packages (google-genai, Pillow) from public registries to enable the skill's functionality.

nanobanana