# Hf Provenance

> Compare a downloaded Hugging Face JSON-weight snapshot with the authoritative Botte micro-NN source. Produces SHA-256 inventories and blocks publication on missing, extra, changed, invalid, or unproven model files. Offline and dependency-free.

- Skill: `zedarvates/hf-provenance` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add zedarvates/hf-provenance`
- Raw SKILL.md: https://api.skillmd.com/api/skills/zedarvates/hf-provenance/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: zedarvates (https://skillmd.com/u/zedarvates)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/zedarvates/hf-provenance

---


# Hugging Face provenance gate

Download the Hub JSON files into a bounded temporary directory, then run:

```bash
python -m skills.hf_provenance.cli /tmp/botte-hub-models \
  --hub-revision <immutable-hub-sha> \
  --source-revision <git-sha> \
  --output distribution/huggingface/micro-nn/model-snapshot.json
```

Exit code `0` means the complete inventories, SHA-256 digests, JSON validity,
and source provenance checks all pass. Exit code `2` is a publication block.
Never replace Hub weights by treating a filename match as proof.