skills-2026.08.21-73

Changes since skills-2026.08.21-71

• .gitignore update (12e2a48)
• lovdata-api: Enhance HTML to text conversion by adding table tags and improving line trimming (f063f3f)

skills-2026.08.21-71

Changes since skills-2026.06.10-69

• lovdata-api: Konverter HTML-fragment til ren tekst med bevart avsnittsstruktur (cfb2de1)
• ci: list all commits since last release in release body (134b79e)

skills-2026.06.10-69

nbno: split auth and OCR detail into sibling files to slim SKILL.md

SKILL.md had grown to ~786 lines / ~37 KB, all loaded into context whenever
the skill triggers — even for a trivial open-content download. Move the two
largest situational blocks into sibling files referenced by firm pointers
(same progressive-disclosure pattern as zotero-ready.md):

- auth.md: Step 2 Options A/B/C procedures (playwright MCP capture,
  capture_cookie.py, cookie format, bearer-vs-nbsso auth scope, _nblb tip).
  SKILL.md keeps the accessInfo pre-check, the which-option decision, the
  mandatory session-check, and an unmissable "read auth.md" pointer.
- reading-ocr.md: the whole "Inspecting pages" section (visual reading,
  tesseract/TESSDATA_PREFIX setup, ocr_chunked.py, shrink_pdf.py). SKILL.md
  keeps a stub that names the gotchas and points to it.

Wording is relocated verbatim, not rewritten, so no guidance is lost.
SKILL.md drops 786 -> 539 lines; description still 911 chars (<= 1024).
Also brought README's Layout block up to date (it was missing several files).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

skills-2026.05.20-65

nbno: improvements from live Bokhylla/pliktmonografi session

- IIIF: discover serveable widths via info.json, verify returned image
  dimensions, fall back to native-res regionByPx tiles when the resolver
  silently downsamples or refuses single-shot requests.
- IIIF: try both /items/<id>/manifest and /iiif/URN:NBN:no-nb_<id>/manifest
  so pliktmonografi items work without manual URL patching.
- Access pre-check: inspect accessInfo.viewability / legalDepositLoginText
  and refuse a no-auth download for FEIDE-restricted items (override
  with --force-auth).
- Persistent pip --target installs under <out>/_pylib so binaries survive
  Cowork's per-call ~/.local wipe.
- Tesseract language preflight: drop missing packs (e.g. nno) and warn.
- Ship ocr_chunked.py — resumable per-page ocrmypdf with pikepdf.open
  validation pass that drops files corrupted by killed mid-write runs.
  Cleans the per-page cache on successful merge.
- Ship shrink_pdf.py — pikepdf+JPEG image re-encode with defaults tuned
  for ~143 KB/page (≈50 MB on a 350-page book), probe-and-extrapolate
  size estimation, sibling-file output by default to avoid overwriting
  high-quality masters. zotero_book.py --shrink keeps the master at
  <basename>.original.pdf for non-destructive re-shrinking.
- SKILL.md: replace misleading "608 px is the max" claim and the
  "pliktmonografi: try no-auth first" heuristic; split the Zotero-ready
  workflow into a separate zotero-ready.md to keep SKILL.md under one
  context-window's worth of content.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

skills-2026.05.20-62

nbno: shorten description to keep under 1024 char limit

skills-2026.05.20-61

nbno: add windows sandbox info

skills-2026.05.20-60

nbno: add Zotero-ready download flow

skills-2026.05.14-59

lovdata-pro and nbno: force login before fetch

skills-2026.05.14-58

lovdata-pro: improvements after live session use (SIV/STR, js-respons…

skills-2026.05.14-57

nbno: improved download speed, bug fixes (based on live session exper…