The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Multiple code templates for model loading (e.g., in templates/runpod-worker.py and reference/model-deployment.md) use the trust_remote_code=True parameter. This allows the execution of arbitrary Python code included in the model's remote repository on Hugging Face.
[EXTERNAL_DOWNLOADS]: The CI/CD instructions in reference/cicd.md download the runpodctl binary from RunPod's official GitHub releases during the deployment process.
[COMMAND_EXECUTION]: Shell scripts within the documentation use sudo to install the runpodctl CLI tool and execute deployment commands as part of the GitHub Actions workflow.
[CREDENTIALS_UNSAFE]: The skill requires sensitive credentials such as RUNPOD_API_KEY and HF_TOKEN to be provided as environment variables or GitHub Secrets for authentication with external services.
[PROMPT_INJECTION]: The LLM handler patterns process user-supplied prompt and messages data. While the skill employs Pydantic for input validation, it lacks specific sanitization layers or boundary markers to mitigate indirect prompt injection or jailbreak attempts within the processed data.

runpod-deployment