The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is designed to package and execute local Python code on remote Runpod workers using decorators like @Endpoint. This involves shipping function bodies via cloudpickle and installing remote dependencies like torch or pandas.
[COMMAND_EXECUTION]: The instructions guide the agent to perform local shell operations, including starting background processes (flash dev), monitoring logs with grep, and testing endpoints via curl. These are standard development tasks for the provided CLI tool.
[REMOTE_CODE_EXECUTION]: The skill requires installing the runpod-flash tool via uv or pip and managing environment-specific dependencies. All installation targets and resource URLs (PyPI, HuggingFace, Runpod) are trusted or well-known services.
[INDIRECT_PROMPT_INJECTION]: As the skill ingests user code and processes data for remote execution, it possesses an attack surface for indirect injection. However, this is inherent to its primary purpose as a developer tool and is documented within standard operating patterns.

flash