Changelog

Each release is a git tag (vX.Y.Z). Projects record the version they run in lineage.lock. Every entry says whether it changes anything that would alter already-generated stories (templates, the book template, style guides, skills); when it does, a project updating to it marks those stories stale rather than regenerating them.

0.7.0 — 2026-10-08

Alters generated stories: no. New tools, template helpers and a publishing path; the book template's existing functions render as before (records gains an internal flag only).

From a working family book's assembly and publication, in generic form (no names, places or material from that project):

Whole-book index without marks. #show: auto-index.with((heading: (spellings…), …)) in the book's master file marks every occurrence of the listed names as the book is set, longest name first, so chapter files stay untouched; make-index merges these with any hand-placed #idx. Names inside a #records list are skipped (records now sets an in-records state).
Exhibits. #exhibit(title:, note:, (path, [caption]), …) reproduces a document's scanned pages after a chapter, two to a row, numbered and captioned (a court file, a service record, a deed).
Readable editions of records. templates/records/readable-edition.typ: cover, plain-English summary and timeline, every original page beside its full transcription (typed text and handwriting distinguished), editor's notes.
Image manager (plugins/lineage/scripts/image_manager.py, skill image-manager): one catalogue of every image (data/image_catalog.csv: SHA-256, size, chapters, captions, origin, prompt hash, source URL, Drive id); check --id/--prompt/--url/--file before generating or fetching anything; missing (broken references, stand-ins, PLATE-TODO anchors), unused, and drive-sync (uploads new or changed images, record PDFs and the book to Google Drive with the drive.file scope; ids written back). generate_images.py now runs the guard before every generation and records each prompt in data/image_prompts.json.
Research tools. archive_fts.py: full-text search of the Internet Archive with the matching passages (yearbooks, gazettes, town reports, newspapers). patent_index_scan.py: every patent granted to a named person, from the Patent Office's annual indexes 1872–1969, with extracts and sources; confirm numbers on Google Patents' patent pages.
Publish a real family's dashboard (site/build_instance.py): the same read-only static snapshot as the public demo, for any project, with its book PDF. Refuses without --allow-real-names; always fails on local paths, emails, keys and localhost links; scrubs home and temp paths from published files; writes a vercel.json. See docs/DEPLOY.md.

Fixed: - The dashboard's home page crashed when a person's article stored its infobox as a list of [field, value] pairs (projects imported from older layouts); infobox reads now accept both forms.

0.6.0 — 2026-10-05

Alters generated stories: yes, for projects that use Lineage's skills: the book template, the writing rules and several skills changed (two columns by default, an epigraph per chapter, THE RECORDS on every chapter, a page minimum). Projects that carry their own copies of the skills are not affected (lineage update knows the difference). lineage update marks generated stories stale; it never regenerates them.

The innovations of a working family book's full first pass, brought up in generic form (no names, places or material from that project). Where the working book's rules and Lineage's disagreed, the working book's rules became Lineage's defaults:

Conflicts resolved (the working book's rule is now the default): - C1 Illustrations. Captions stay short (a name, or a scene title in the text's words; the author's caption word for word). An illustration is marked by kind: illustration in photo_index.csv, the chapter's note, and a copyright-page line ("The illustrations in this book are artist's renderings, not photographs."), which build_book.py front now writes whenever a placed image is an illustration (book.yaml → front.illustrations_note). The old rule that each caption must say "illustration" is retired. - C2 Epigraphs. One per chapter by default (chapters.epigraphs); two allowed in a family-history chapter; none if the author asks. A writer of the chapter's place and time. - C3 THE RECORDS on every chapter, listing 100% of the sources found and used (not only family-history chapters or chapters with context). - C4 No "general knowledge" sources. Every context fact cites a real source. - C5 Two columns by default. chapter.with(columns: 2) is the template default and assemble.py writes 2 when data/chapters.csv leaves columns blank; 1 stays available per chapter. Rows that already say 1 are unchanged. - C6 Archive thumbnails in drafts. #photo-addendum's show-images defaults to the draft flag: thumbnails print in drafts, and in the final book only once the holder's permission is recorded (show-images: true). - C7 Gates. Work stops at the author's chosen gates (three by default, workflow.gates; the project's CLAUDE.md may be stricter and wins), and a full first pass of the whole book runs when the author asks for it (the book swarm, below).

New: - Chapter dossiers (chapter-dossier skill, scripts/dossier.py, templates/dossier/): per chapter, the author's dated requests, a log of every search (blocked ones too), 100% of sources in SOURCES.csv, the evidence for and against the family's version, and lessons. dossier.py records generates THE RECORDS from it; dossier.py check fails when a URL in the chapter isn't logged or a cited source isn't listed. - Polite record fetching (scripts/fetch_records.py): targeted lookups (at most 50 URLs a run), robots.txt obeyed, at least 2 s per site, a project-only User-Agent (research.user_agent; an email address is refused), no logins or bot-check workarounds; raw copies with sha256 metadata; logged in the dossier; anything blocked goes on the manual-lookups list. - Research access rules (records-archives): automated fetching vs targeted lookups by a person in their own browser and accounts (or a browser assistant in the author's own session, at their direction); never CAPTCHA solving, bot-check tools, others' credentials or bulk harvesting. Hand-off templates: templates/research/MANUAL-LOOKUPS.md, manual-lookups-results.csv, and REQUESTS.md (archive requests and replies); make new copies them into research/. - The service-record method (family-history-chapters §6a): for anyone who served, the official record first as the spine; then the hometown and country, how the war drew them in and their reputation, enlistment, training and shipping, the unit's battles, weapons and generals, being sent home, and what veterans came home to; first-hand accounts from soldiers in or near the unit, quoted exactly and verified. - Sensitive record entries (memoir-style-guide §2): kept out of the prose until the author decides; then told plainly from the record, without moralising. - Page minimum and combining (style guide §6, chapter-index-builder): chapters of at least chapters.min_pages (default 10) in the layout, reached with real material; sparse chapters are combined, never padded. build_book.py chapter (make chapter) prints a note when a chapter is short. - Template: #family-tree(...), a three-generation tree (parents → couple → children, the line of descent continued under one child) that stands in for or beside #descent; #story-panel and #story-page, an eight-panel illustrated story over two facing pages, each panel captioned with the subject's exact words. - Illustration generator (scripts/generate_images.py, templates/images/): plans in photos/images-plan.yaml with period style presets, a one-line lock on every prompt, a required scene citation, and an optional likeness lock (real photographs as references through the edit endpoint); writes the original and a grayscale print copy; --dry-run prints every prompt. The key comes from OPENAI_API_KEY or ~/.openai_api_key (chmod 600, outside every repo) and is never printed or written. - The book swarm (docs/BOOK-SWARM.md, templates/swarm/book-swarm.js): a parameterized Workflow script for a full first pass: per chapter two researchers → writer → fidelity, taste and layout editors → two judges with revisions to a pass score → a 100% sources pass; then front matter, images plan, index, assembly and a whole-book continuity review. - book.yaml template: front.illustrations_note, chapters.min_pages, chapters.epigraphs, research.user_agent, workflow.gates. Project .gitignore: photos/generated/, photos/reference/, research/manual-lookups/. Project CLAUDE.md: gates, dossiers, research. - Docs: HOWTO (research access, dossiers, family tree, anyone who served, generating illustrations, the illustrated story, page minimum, gates and the full first pass), README, scripts/README, EXAMPLE-CHAPTER notes; the dashboard's standing rule on generated images. - Typst gotchas documented: a ; after an embedded call in markup is swallowed (write \u{3B}); a full-page float sized to the text block.

Also in this release (made on development after 0.5.0):

Dashboard: - Sources: a gallery and a lightbox (arrow keys, summary, people and places linked, provenance, notes, subject tagging). Roadmap #5. - Stories: what each story rests on, above its pages: every paragraph's citation chips, and every image's provenance with a toolbar (open, tag, copy reference). Roadmap #12. - An illustration is never a person's picture (tree, featured relative, person articles); "illustration" is a kind of source. - Wording: the dashboard speaks of the family's record; "the book" only where the export is meant.

Skills and scripts (generic improvements brought up from a working project): - Writing rules: locked passages (voice/verbatim.md) go in #verbatim word for word; never silently correct the subject; public-domain status checked per edition for epigraphs; an author's introduction to a part is a #bridge until approved; when the subject told a story two ways, keep both, attributed; contents lines are short plain nouns; propose an album section when a chapter has more good photographs than text. - Photo intake and archive workflow (photos/inbox/ → photos/archive/originals/…); dating cues; relative dates chained to the event they hang on. - make chapter PROJECT=… CHAPTER=… builds one story as a preview; PAGES=1 exports each page as an image for sharing. build_book.py falls back to the typst Python package without the CLI. - transcript/corrections.json → drop_word_runs: remove phrases the speech model echoed from its prompt, word by word. - lineage update: skill changes don't mark stories stale in projects with their own skills.

0.5.0 — 2026-10-01

Alters generated stories: no. Eight skills' description: lines changed (what triggers them, not what they tell the writer); lineage update now recognizes a description-only change and doesn't mark stories stale for it.

Lineage is described as what it is: a cooperative family documentation and research platform whose outputs include stories, a book and narration. - README rewritten around the platform, with a tour of the real dashboard; docs/HOWTO.md restructured (set up a lineage, contributors, sources, records, genealogy and Familypedia, then stories and a book); docs/FACTORY.md, the plugin, marketplace and skill descriptions reframed. - examples/demo-lineage/: an invented family (the Calders) with recordings, scanned records, a ship, a route, a genealogy, contributors and a narrated story, built by scripts/demo_lineage.py. make demo opens the dashboard on it. - make screenshots (scripts/screenshots.py, Playwright) captures the dashboard against the demo lineage into docs/images/, with an empty HOME so no keys or accounts appear. - Timeline and Familypedia tiers: a record citation (R001), manifest or ledger now counts as "what the records show" (the check was case-sensitive after lower-casing).

0.4.0 — 2026-10-01

Alters generated stories: no. Nothing under plugins/lineage/book/, fonts/ or skills/.

Projects are consumers of Lineage: - bin/lineage version | update | dashboard. update fetches the releases, prints the changelog between the pinned release and the target and the files of the project it would touch; with --apply it installs the release read-only under ~/.lineage/releases/<tag>, pins it in lineage.lock, re-points .claude/skills when it links to a release, and marks stale (never regenerates) the stories a release would alter. Material is never touched. - The project template gains a Makefile (make dashboard, make lineage-version, make update-lineage [APPLY=1] [TO=vX.Y.Z]), lineage.lock, local-overrides/, the dashboard's runtime files in .gitignore, and the arrangement at the top of its CLAUDE.md. - CLAUDE.md for this repo: platform changes are made here and flow down; content stays in the project; one-off things go in the project's local-overrides/. - Familypedia: empty Sources, Records and Photographs sections are hidden; a lone word that only names a stub person (a surname off a roster) is no longer linked automatically.

0.3.2 — 2026-10-01

Alters generated stories: no. Dashboard engine only (app/familypedia.py).

A place on a route now links back to the ships and voyages whose route it is ("On the route of"), so a vessel, its ports and its voyage are reachable from each other.

0.3.1 — 2026-10-01

Alters generated stories: no. Dashboard engine only (app/familypedia.py).

Familypedia fixes found while building a real project's articles: - After two articles are folded together ("same as"), the names of the folded one now find the surviving article, so its coordinates, routes, events and relations carry over. - The person article's link into the genealogy tree reads the tree's people as a mapping (as the Genealogy tab writes it), not a list. Articles no longer fail on that. - Also called: in a person profile keeps the names (and names in quotation marks) and drops commentary written on the same line.

0.3.0 — 2026-10-01

Alters generated stories: no. Nothing under templates/, book/, plugins/lineage/book/, plugins/lineage/templates/ or plugins/lineage/skills/ changed since 0.2.0.

The dashboard (app/), new since 0.2.0: - Home, a settings gear (Family details · Connectors · Contributors · Project settings) and a terminal drawer; Sources, Familypedia, Genealogy, Stories and Timeline tabs. - Source intake end to end: SHA-256 dedupe, extraction and OCR, summaries, thumbnails, edits kept apart from derived values, trash and restore. No demo content unless --demo. - Stories with states, rendering to book pages, narration; Timeline with tiers, conflicts and gaps; Genealogy rebuilt from the material with evidence, review, edits and GEDCOM. - Familypedia article types: person · place · event · vessel · organization/unit · object · publication · occupation · theme, with typed infoboxes, tier sections, passages, sources, typed records, photographs and marked illustrations, related articles, backlinks, open questions and "Beyond the family"; tagging of sources, records, photographs and events to any type (picker, suggestions with evidence, bulk); browse, search, [[links]], and a map drawn from the records' own coordinates. Inputs and outputs: app/README.md. - Example chapter: the Hannibal route and voyage details.

0.2.0

The project renamed to Lineage; a Claude Code plugin marketplace with two plugins (lineage, factory).