hf-free-maxxing

hf-free-maxxing — an Agent Skill for the free Hugging Face tier

An Agent Skills-format skill (installable folder with a SKILL.md) that gives any compatible coding agent — Claude Code, Codex, Cursor, Gemini CLI, OpenCode, … — verified, measured access to every FREE Hugging Face capability through one CLI (hfx):

• Storage: 100 GB private + 8.7 TB public per entity (user + each org), max file 500 GB
• Bandwidth: 20 TB/mo (user) + 30 TB/mo per org on CloudFront
• Static hosting: unlimited always-on static Spaces, CORS *
• Media CDN: permanent uploads outside both storage and bandwidth quotas
• LLM API: $0.10/mo credits ≈ 5M tokens on the cheapest pinned lane ($0 promos appear & vanish — monthly re-check built in)
• GPU: ZeroGPU bursts — 8 runs + 300 GPU-s per rolling 24h (2000+ MCP-enabled Spaces)
• Data backend: SQL-ish filter/search/stats over dataset repos + DuckDB full-SQL lane
• Identity: full OIDC/OAuth2 provider (PKCE, device flow, RFC 7591), no MAU cap
• Docker: read-only pull library of every public Docker Space image
• Events: webhooks CRUD + replay, bucket change-feed SSE, notifications

Every quota number, endpoint, and gotcha was live-verified in Sep 2026 by the hf-free-maxxing research project; the skill distills it into an agent-loadable operating manual + a 16-command CLI (kit/bin/hfx).

Install

Into any skills directory (.agents/skills/ is the cross-agent standard):

# with the hf CLI (v1.x):
hf download landogayatri/hf-free-maxxing --local-dir .agents/skills/hf-free-maxxing

# or plain git:
git clone https://huggingface.co/landogayatri/hf-free-maxxing .agents/skills/hf-free-maxxing

# once merged into the huggingface/skills marketplace:
hf skills add hf-free-maxxing

Then follow SKILL.md (or just mention the skill to your agent). Deps: pip install "huggingface_hub<2.0" boto3 gradio_client and an HF token in HF_TOKEN. Verify with bash kit/bin/hfx doctor.

Layout

hf-free-maxxing/
├── SKILL.md      # agent-facing skill instructions (start here)
├── kit/          # the hfx CLI + full consumer docs
│   ├── AGENTS.md # 1200-line verified reference (per-capability deep docs)
│   ├── bin/hfx   # the CLI (16 commands, --help everywhere)
│   ├── lib/      # pure-Python modules
│   └── docs/     # AGENTS.md skeleton + per-topic fragments
└── LICENSE       # Apache-2.0

Provenance

Live-verified Sep 2026; evidence transcripts (findings/, playbooks/, data/kit-tests/) live in the research repo. Quotas move in real time — hfx status / hfx infer budget are the source of truth.