verify-provenancelisted

# Skill: verify-provenance The model-free proof of qcobrain's one promise: **every claim in `wiki/` re-resolves, verbatim, in `sources/`.** `/wiki-doctor` describes this check as a reasoning workflow; this skill makes it a plain script anyone — a contributor, CI, or a skeptic evaluating the repo — can run to PROVE the graph holds with no LLM in the loop. If it exits `0`, cite-or-refuse is intact; if it exits non-zero, something drifted and the message says exactly what. ## Run it ```bash python3 skills/verify-provenance/verify_provenance.py # full report python3 skills/verify-provenance/verify_provenance.py --quiet # summary + failures only (CI mode) python3 skills/verify-provenance/verify_provenance.py --root . # explicit repo root ``` It needs nothing but Python 3 (standard library) and reads only `wiki/` and `sources/`. No network, no config, no API key, no model. ## What it checks For every `wiki/<id>.md` note (it skips `README.md` and any file without a frontmatter `id:`): **Hard checks — any failure exits non-zero:** 1. Every ` ```src ` block names a `source` key that is registered in `sources/_sources.md`. 2. The registered source file exists on disk. 3. The source's live `sha256[:12]` matches the hash recorded in `sources/_sources.md` (no silent drift). 4. The block's `quote` re-resolves as a **verbatim substring** of that source (whitespace-normalized). 5. A note carries **at least one** ` ```src ` block — a `wiki/` note with no citation belongs in