Hugging Face Spaces Builder
Build, deploy, debug, and maintain machine-learning apps on Hugging Face Spaces.
The documentation covers login, token scope, confirmation for paid storage, the distinction between secrets and public environment variables, and warns against storing PII in public buckets. However, deployment, upload, restart, resource creation, and external inference create material external effects; unified confirmation gates, rollback procedures, least-privilege checks, and complete data-flow disclosure are incomplete, so points are deducted.
It provides CLI preflight checks, dependency-conflict guidance, log diagnosis, known errors, and an end-to-end smoke-test workflow, and the main paths are broadly self-consistent on static inspection. However, it assumes current platform, CLI, and external-service behavior, contains no committed test suite or CI coverage, and some version or hardware claims may become stale, so the score remains below the static ceiling.
The frontmatter clearly identifies Space creation, hosting, debugging, ZeroGPU, Docker, Static SDK, and hf spaces scenarios, with fairly specific triggers. Still, non-fit boundaries, input/output contracts, and protection against semantic false triggers are incomplete; Chinese-language support is absent; and the core function depends on Hugging Face external services whose reachability from mainland China is not addressed, so points are deducted.
The information architecture is clear, with progressive disclosure through reference files and coverage of installation, dependencies, examples, troubleshooting, limitations, licensing, and update suggestions. Maintenance ownership, version policy, changelog, and a formal update path are not clear, while several claims depend on external documentation, so points are deducted.
The skill offers reusable commands, code patterns, and troubleshooting guidance for creating, implementing, debugging, and verifying Spaces, making the core value proposition clear. This review did not execute the instructions or verify a real Space output, and users must still adapt the guidance to the specific model, dependencies, and account constraints, so the static ceiling applies.
Commands, error substrings, configuration examples, and links to Hugging Face primary documentation provide some auditability. However, there is no committed test suite, CI evidence, version-locked verification, or independent reproduction record; key claims remain primarily author-provided guidance, so the score stays below the static ceiling.
- The skill can create, upload, restart, and reconfigure Spaces, set secrets or environment variables, and create paid buckets; confirm targets, visibility, token scope, cost, and data exposure before execution, and prepare a recoverable path.
- Core workflows depend on huggingface.co, the Hugging Face CLI, the Spaces runtime, and external inference providers; stable reachability from mainland-China networks is not established by the source material.
- The document says to look up current versions, but its examples and hardware, Python, and Torch compatibility claims can change with the platform; recheck them before use.
- Do not place real tokens, PII, or other sensitive data in public environment variables or public buckets.
What does this skill do, and when should you use it?
This skill guides agents through creating, building, debugging, and maintaining applications hosted on Hugging Face Spaces. It covers Gradio, Docker, and Static SDKs, along with ZeroGPU, dedicated hardware, model loading, persistent buckets, inference providers, and community grants. It uses the hf CLI, Space README frontmatter, the spaces Python package, logs, and live API checks as part of the workflow. It is a strong fit for developers who need to host working Hugging Face applications, with hardware and storage options constrained by account permissions and cost.
Checks whether the hf CLI is installed and whether the user is authenticated; searches existing Spaces for prior art; selects an SDK and hardware tier; creates Spaces and configures README frontmatter, application code, and dependencies; applies ZeroGPU patterns using spaces.GPU and Gradio MCP launch settings; uploads files with hf upload; reads build and runtime logs; calls Gradio APIs to verify endpoints and outputs; configures buckets for persistent storage; and diagnoses failures using known-error references and an iterative debugging ladder.
- A developer wants to publish a public machine-learning demo using Gradio and ZeroGPU.
- A model author wants to adapt an existing GitHub inference project into a Hugging Face Space.
- A Space maintainer needs to diagnose a failed build, broken startup, or incorrect API output.
- A developer needs a non-Python container or a browser-side application and must choose Docker or Static SDK.
- A Space owner needs persistent storage for uploads, generations, or logs through a Hugging Face bucket.
What are this skill's strengths and limitations?
- Covers the workflow from Space creation through live verification.
- Explains how to choose among Gradio, Docker, Static, ZeroGPU, and dedicated hardware.
- Includes concrete guidance for ZeroGPU, logs, API smoke tests, buckets, and inference-provider fallbacks.
- Calls out operational pitfalls such as ZeroGPU import order and hf upload repository targeting.
- ZeroGPU is Gradio-only, PyTorch-first, and requires the creator to use a PRO, Team, or Enterprise plan.
- Dedicated GPUs and buckets are paid capabilities and may require canPay=True.
- The workflow depends on the hf CLI, network access, and a live Space environment; local testing is intentionally limited.
- The supplied material does not document an automated test suite or cross-platform test coverage.
How do you install this skill?
Obtain the skill folder at skills/huggingface-spaces/ and copy or symlink it into a standard .agents/skills location supported by the Agent Skills client. For Codex, the README gives repository-level or user-level .agents/skills locations as examples. The README also documents on-demand installation with hf skills add <skill-name>; the repository supports Claude Code, Codex, Gemini CLI, and Cursor.
How do you use this skill?
Trigger it explicitly in a coding agent, for example: “Use the Hugging Face Spaces skill to deploy this PyTorch model as a ZeroGPU Gradio Space and verify its live API.” Check the hf CLI and authentication first; create a Space with hf repos create and upload files with hf upload <namespace>/<name> . --repo-type space. The exact implementation depends on the model, SDK, hardware choice, and account permissions.
How does this skill compare with similar options?
Compared with the repository’s hf-cli skill, huggingface-spaces focuses on building, deploying, debugging, and maintaining Space applications. The README positions hf-cli as the recommended companion for the broader Hugging Face Hub command set.