Ingest Substack, Medium, and Ghost exports as first-class writing sources

[telivity-otaip]

Why

LinkedIn export + local notes cover a lot, but many writers keep their longform on Substack / Medium / Ghost. Broader ingest = richer article-channel voice indexes and a bigger potential user base discovering the repo.

Scope

Extend ingest so these sources land as proper Piece rows (prefer linkedin_article-equivalent longform or a new source label that write --channel article can use).

Suggested sources

Source	Input shape	Notes
Substack	export ZIP or posts/ markdown	Dates + titles from front matter
Medium	export ZIP (posts/ HTML)	Strip Medium chrome/CSS (see corpus_text.py patterns)
Ghost	JSON export or markdown	Map lexical/html → clean text

CLI sketch:

personality-protect ingest --substack ~/Exports/substack.zip
personality-protect ingest --medium ~/Exports/medium.zip
personality-protect ingest --ghost ~/Exports/ghost.json

Acceptance criteria

• At least two of the three sources fully supported (third can be follow-up)
• Pieces get stable ids, dates/years when present, and a clear source field
• HTML/CSS chrome stripped so style measurement sees real prose
• Synthetic Contoso fixtures + unit tests (no real author exports in the repo)
• README ingest section updated
• Article channel can retrieve these as longform rhythm refs when labeled appropriately

Pointers

• src/personality_protect/ingest.py
• src/personality_protect/corpus_text.py (Medium/Ghost CSS stripping already exists)
• src/personality_protect/models.py (Piece.source)

Privacy

Never commit real exports. Ship only synthetic fixtures under tests/fixtures/.

Difficulty

Medium — parsing + tests; high user value.