feat(skills): add 12 Hugging Face skill entries (OCI distribution)

[JAORMX]

Summary

Adds the 12-skill Hugging Face pack from huggingface/skills to the ToolHive catalog. Content is Apache-2.0, pinned upstream to commit 061ab49 and packaged by Dockyard to ghcr.io/stacklok/dockyard/skills/<name>:0.1.0 (Dockyard PR #501).

Follows the OCI-distribution pattern established by PR #1093 (claude-api), PR #1094 (toolhive-cli-user), PR #1095 (trailofbits security skills), and PR #1109 (gemini skills).

Skills added

• hf-cli — Hugging Face Hub CLI (hf) for models, datasets, spaces, buckets, repos, jobs, inference endpoints, etc.; replaces deprecated huggingface-cli
• hf-mcp — Use the Hugging Face Hub via MCP server tools (search, repo details, docs, compute jobs, Gradio Spaces as tools)
• huggingface-community-evals — Run HF Hub model evaluations locally with inspect-ai or lighteval (vLLM / Transformers / accelerate backends)
• huggingface-datasets — Dataset Viewer API workflows (rows, pagination, filters, parquet URLs, SQL via parquetlens)
• huggingface-gradio — Build Gradio web UIs in Python (Interface, Blocks, ChatInterface, gradio CLI)
• huggingface-llm-trainer — Train/fine-tune LLMs with TRL (SFT, DPO, GRPO, reward modeling) or Unsloth on HF Jobs cloud GPUs; GGUF conversion
• huggingface-paper-publisher — Publish and manage research papers on HF Hub (pages, links, authorship, citations)
• huggingface-papers — Look up and read HF paper pages in markdown and use the papers API (arXiv IDs, structured metadata)
• huggingface-tool-builder — Build reusable CLI scripts using the HF API and hf CLI (composable NDJSON output, auth hygiene)
• huggingface-trackio — Track and visualize ML training experiments with Trackio (metrics, alerts, HF Space dashboards)
• huggingface-vision-trainer — Train vision models on HF Jobs (object detection, image classification, SAM/SAM2 segmentation)
• transformers-js — Run ML models in JavaScript/TypeScript via Transformers.js (NLP, vision, audio, multimodal; WebGPU/WASM)

Notes for review

• License: Apache-2.0. Confirmed at the huggingface/skills repo root; upstream does not embed an SPDX identifier in per-skill SKILL.md frontmatter, which is tracked as an allowed Dockyard manifest issue.
• MCP dependencies satisfied. The huggingface MCP server entry already in this catalog (registries/official/servers/huggingface and registries/toolhive/servers/huggingface) satisfies the runtime dependency for hf-mcp, huggingface-llm-trainer, and huggingface-vision-trainer (they call hf_jobs, hf_whoami, hf_doc_search, hf_doc_fetch).
• No allowedTools. None of the 12 upstream SKILL.md files declare allowed-tools / allowedTools in frontmatter. huggingface-llm-trainer references mcp__huggingface__hf_whoami in the body but not in frontmatter, so no allowedTools is added per the frontmatter-not-body policy.
• Non-flat upstream layout for hf-mcp. Its packages[].subfolder is hf-mcp/skills/hf-mcp (matches upstream repo layout); the other 11 use the standard skills/<name> subfolder.
• No skill/SKILL.md subfolder. Following the claude-api / toolhive-cli-user / gemini-* precedent — OCI is authoritative, commit-pinned git package is included as a fallback.
• Icons. All 12 skills share the same monochrome 24x24 smiley SVG (thematic for Hugging Face). Happy to differentiate per-skill if reviewers prefer.

Test plan

• task catalog:validate — 32 skills total (20 existing + 12 new), all valid (both upstream and toolhive formats)
• task catalog:build — all 12 new skills present under data.skills in build/toolhive/registry-upstream.json
• task catalog:validate in CI
• Post-merge: skills appear in the published registry-upstream.json feed

Signed-off-by: Juan Antonio Osorio ozz@stacklok.com

🤖 Generated with Claude Code

[github-actions]

Registry Build Preview

All registries built successfully!

Registry: official

• Total Servers: 68
  • Container-based: 47
  • Remote: 21
• Skills: 0
• File Size: 1.5M
• Last Updated: 2026-04-21T09:30:25Z

Registry: toolhive

• Total Servers: 113
  • Container-based: 82
  • Remote: 31
• Skills: 32
• File Size: 1.9M
• Last Updated: 2026-04-21T09:30:26Z