Skill: recover-outline-bridge

Recover PDF outline bookmarks by OCR-ing the printed table of contents via the Arcane-PP server, then using the parsed entries as seeds for recovery.

When to use

• The PDF has no bookmarks and no reference edition is available
• The printed TOC is the only reliable source of chapter structure
• The user mentions "Arcane-PP", "OCR", "TOC pages", or "bridge"
• Heuristic recovery (recover-outline-heuristic) gave poor results

Prerequisites

• Arcane-PP server must be running. Verify:

  bash .claude/commands/recover-outline-bridge/scripts/check-server.sh
  

• User must provide (or agree to identify) the TOC page range.

Steps Claude must follow

1. Probe the PDF:

   arcane analyze probe "<pdf>" --json
   

   Abort if document_kind == "Scanned".

2. Identify TOC page range — ask the user if unknown. Alternatively, detect automatically:

   arcane analyze layout "<pdf>" --json | python3 -c \
     "import sys,json; d=json.load(sys.stdin); \
      toc=[a for a in d['anchors'] if a['kind']=='TocEntry']; \
      print(f\"TOC pages: {min(a['page_index']+1 for a in toc)}-{max(a['page_index']+1 for a in toc)}\") if toc else print('No TOC anchors detected')"
   

3. Manual path — extract and inspect (recommended for first time):

   arcane pdf extract-pages "<pdf>" --start N --end M toc-extract.pdf
   arcane process-toc "<pdf>" --toc-pages "N-M" --output seeds.json
   

   Show the user the seeds.json content. Ask: "Does this look like the correct chapter list?" Then run recovery with the seeds:

   arcane recover-outline "<pdf>" --seed-file seeds.json --toc-pages "N-M" \
     --dry-run --depth 2
   

4. Automated path (when user is confident):

   arcane recover "<pdf>" --toc-pages "N-M" --output "<fixed.pdf>" --depth 2
   

5. For a whole project (when all sources have TOC ranges set):

   arcane recover-project --project "<name>" --dry-run
   arcane recover-project --project "<name>"
   

6. Verify:

   arcane analyze outline "<fixed.pdf>" --depth 2
   

7. Fill in template.md and present the summary.

Setting TOC ranges for batch recovery

When adding sources, include TOC page ranges so recover-project works:

arcane add "<project>" "<pdf>" --textbook --toc-start-page N --toc-end-page M