The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection because its primary function is to take user-supplied text and send it to external Generative AI models via Google Cloud APIs.
Ingestion points: Processed data (user prompts) enters the system via scripts such as scripts/gemini_genai_sdk.py and scripts/openmaas_openai_sdk.py.
Boundary markers: The SKILL.md file defines a "Safety & Confirmation Tier" (Tier R) which requires the agent to obtain explicit interactive confirmation from the user before performing inference and prohibits execution in the same turn as the confirmation request.
Capability inventory: The skill performs network operations to Google Cloud endpoints and executes local Python and shell scripts.
Sanitization: There is no evidence of input validation, escaping, or filtering of the user-provided prompts in the included scripts.
[EXTERNAL_DOWNLOADS]: The skill installs standard Python packages from well-known and trusted providers.
Evidence: scripts/requirements.txt lists google-genai, google-cloud-aiplatform, openai, and google-auth as dependencies to be installed via pip.
[COMMAND_EXECUTION]: The skill executes shell and Python scripts for environment verification and model inference.
Evidence: The scripts/verify_all.sh script programmatically iterates through and executes all Python files located in the scripts/ directory to verify the environment.

agent-platform-inference