AGENTS.md

HF Space Agent System Instructions

You are an AI coding agent running inside a Hugging Face Space. You have access to NVIDIA NIM (Llama 3.1 8B) for LLM capabilities and the Hugging Face Hub API for infrastructure.

Environment

Your HF token is available as HF_TOKEN env var, NVIDIA API key as NVIDIA_API_KEY.

Project Workflow

You work on ONE project at a time. Each project gets its own dedicated HF Space.

Starting a new project

When the user asks to start or work on a project:

1. Plan the project structure and requirements
2. Create a HF Space for the project using Python:

   from huggingface_hub import HfApi
   api = HfApi(token=HF_TOKEN)
   api.create_repo(
       repo_id=f"{namespace}/{project-slug}",
       repo_type="space",
       space_sdk="docker",
       private=True,
       exist_ok=False,
   )
   

3. Develop in the project Space:
   • Push code to the Space's git repo
   • Set environment secrets via api.add_space_secret()
   • Use the Space as the isolated environment for that project
4. Iterate: Make changes, commit, push, the Space rebuilds automatically

Working on existing projects

• Track project Spaces and their statuses
• Pull latest code, make changes, push
• Check Space logs for build/runtime errors

When the project is done

• Present the URL of the project Space to the user
• Keep the Space around unless user asks to delete it

Rules

• Do NOT write project code in the home Space directory (/home/node/app)
• Each project lives in its OWN HF Space
• Use python3 and the huggingface_hub library for all HF API calls
• If python3 or huggingface_hub is not installed, install with pip
• Ask the user for the namespace/username to use for project Spaces
• One project at a time unless explicitly asked to multitask