Ebook Library Skill

Use this skill when a user wants to look up, search, or resolve books in a local Calibre collection.

This skill is read-only. It can inspect Calibre databases and file paths but does not write to library metadata.

Library discovery and validation

Use a real Calibre library only. The source of truth is metadata.db and, for text lookup, full-text-search.db. Never create substitute library content, fake JSON catalogs, invented book files, or ad hoc metadata when the real library is missing.

1. If you already know the library location, set these first:

export CALIBRE_LIBRARY_ROOT="/path/to/Calibre Library"
export CALIBRE_METADATA_DB="$CALIBRE_LIBRARY_ROOT/metadata.db"
export CALIBRE_FTS_DB="$CALIBRE_LIBRARY_ROOT/full-text-search.db"

2. If the library root is unknown, discover DB paths first:

find "$HOME" -name metadata.db 2>/dev/null
find "$HOME" -name full-text-search.db 2>/dev/null

3. Before running any lookup, verify what is actually readable:

test -r "$CALIBRE_METADATA_DB" && echo "metadata ok"
test -r "$CALIBRE_FTS_DB" && echo "fts ok"

• If the task names a specific library path, check that exact path first.
• If metadata.db is missing or unreadable, stop and report the missing path.
• If full-text-search.db is missing or unreadable, restrict yourself to metadata/path tasks and say content search is unavailable.

Decision tree

• User asks for books by metadata (title, author, topic) → use find_books.py.
• User asks for a quote/passage (global or in one book) → use search_content.py (prefer --book-id when possible).
• User asks for nearby context around a hit → use get_excerpt.py.
• User asks for file path → use resolve_book.py.
• User asks to browse when search is vague → use list_books.py.
• User asks for quick sanity checks → use inspect_calibre_metadata.py.

Common commands

Use these directly for the most common tasks.

Find books by title, author, or topic

python3 scripts/find_books.py \
  --db-path "$CALIBRE_METADATA_DB" \
  --query "Dune" \
  --limit 5

Returns a JSON array like:

[{"id": 4, "title": "Dune", "authors": "Frank Herbert"}]

Search for a phrase inside one known book

python3 scripts/search_content.py \
  --fts-db "$CALIBRE_FTS_DB" \
  --metadata-db "$CALIBRE_METADATA_DB" \
  --book-id 4 \
  --query "knowledge" \
  --context 400

Use this first when you already know the target book_id.

Search for a phrase across the whole library

python3 scripts/search_content.py \
  --fts-db "$CALIBRE_FTS_DB" \
  --metadata-db "$CALIBRE_METADATA_DB" \
  --query "artificial intelligence" \
  --limit 10

Pull a longer excerpt around a hit

python3 scripts/get_excerpt.py \
  --fts-db "$CALIBRE_FTS_DB" \
  --metadata-db "$CALIBRE_METADATA_DB" \
  --book-id 4 \
  --around "knowledge" \
  --chars 800

Resolve a Calibre book ID to a file path

python3 scripts/resolve_book.py \
  --metadata-db "$CALIBRE_METADATA_DB" \
  --library-root "$CALIBRE_LIBRARY_ROOT" \
  --book-id 4 \
  --format EPUB

Browse books when search terms are vague

python3 scripts/list_books.py \
  --db-path "$CALIBRE_METADATA_DB" \
  --limit 50

Sanity-check database access

python3 scripts/inspect_calibre_metadata.py \
  --db-path "$CALIBRE_METADATA_DB" \
  --limit 20

Quick response rules

• find_books.py and search_content.py return [] for honest no-match cases.
• Invalid book IDs return structured JSON errors such as {"error": "Book 999 not found", "error_code": "BOOK_NOT_FOUND"}.
• Prefer search_content.py --book-id ... over global content search whenever possible.
• When a hit needs proof, follow with get_excerpt.py and quote the returned snippet.
• If you had to discover the library location, report the path you actually used.

Orchestration (fallback rules)

After the library discovery step, use the decision tree and common commands above. Apply these extra fallback rules only when the first pass does not resolve the task:

1. If find_books.py returns no results, run inspect_calibre_metadata.py to confirm DB accessibility, then retry with a shorter or partial query.
2. If a global phrase search returns no results, confirm the target with inspect_calibre_metadata.py, then retry with a narrower phrase.
3. If find_books.py returns multiple plausible titles, run scoped search_content.py searches to disambiguate before choosing a candidate.
4. If a hit needs proof, run get_excerpt.py; if it errors, retry with a broader search in the same source before concluding failure.
5. If resolve_book.py returns exists: false, verify the book_id via find_books.py, then retry path resolution with the preferred format.
6. Return transparent evidence fields: book id, title, author, method used, and snippet or path as appropriate.

Result handling

• Empty arrays ([]) are valid "no result" responses.
• search_content.py/get_excerpt.py/resolve_book.py can return structured errors for invalid IDs, missing text, or bad positions.
• If a book_id seems stale, rerun metadata lookup first (find_books.py by title/author context), then retry resolve_book.py before concluding the library is broken.

Boundaries

• Do not infer file conversions or imports from search output.
• Do not guess book titles/authors when lookup returns no match.
• Prefer precise scoped searches to reduce unnecessary full-library scans.

For exact script arguments, output formats, examples, and error payloads, see references/scripts.md.