prompts.example.yaml

# EPUB LLM Cleaner - Example Prompts Configuration
# ================================================
#
# This file contains example prompts for the EPUB cleaning workflow.
# Copy this file to 'prompts.yaml' and customize for your specific needs.
#
# The cleaning process works in two stages:
#   1. ANALYSIS: Quickly scan each chapter to decide if it needs cleaning
#   2. CLEANING: Selectively rewrite only the problematic paragraphs
#
# PLACEHOLDERS:
#   {chapter_text}     - Used in analysis_prompt, contains the full chapter text
#   {numbered_chapter} - Used in cleaning_prompt, contains paragraphs with [PARAGRAPH N] markers
#
# CUSTOMIZATION TIPS:
#   - Be specific about what issues to look for (OCR errors, formatting, etc.)
#   - Provide examples of common problems you've seen in your ebook
#   - Define clear criteria for when to filter vs. when to skip
#   - For cleaning, specify exactly how you want issues fixed
#
# ============================================================================

# ANALYSIS PROMPT
# ---------------
# Purpose: Quickly determine if a chapter needs any cleaning
# Input: {chapter_text} - The full text content of the chapter
# Expected Output: Single word - either "FILTER" or "CLEAN"
#
# Tips for customization:
#   - List specific patterns or issues to look for
#   - Include examples of problematic text if helpful
#   - Define what should NOT trigger filtering to reduce false positives
#   - Keep the prompt focused for faster, more accurate decisions

analysis_prompt: |
  You are analyzing a chapter from a digitized book to determine if it contains issues that need correction.

  FILTER if the chapter contains ANY of these issues:

  OCR and Scanning Artifacts:
  - Misread characters (e.g., "rn" read as "m", "cl" as "d", "1" as "l")
  - Random punctuation or symbols from scanning errors
  - Words split incorrectly across lines (e.g., "to-gether" or "some thing")
  - Missing spaces between words or extra spaces within words
  - Garbled text or nonsense character sequences

  Formatting Problems:
  - Headers or footers repeated within body text
  - Page numbers embedded in paragraphs
  - Running headers mixed into content
  - Inconsistent paragraph breaks or missing line breaks
  - Bullet points or list markers that didn't convert properly

  Text Quality Issues:
  - Repeated words or phrases (e.g., "the the", "and and")
  - Incomplete sentences that appear cut off
  - Obvious typos or misspellings that appear systematic
  - Filler text or placeholder content

  DO NOT FILTER if the chapter only contains:
  - Intentional stylistic choices by the author
  - Archaic or unusual spelling that is period-appropriate
  - Technical terms or proper nouns that may look unusual
  - Normal punctuation and formatting

  Analyze this chapter:

  {chapter_text}

  Respond with ONLY ONE WORD:
  - "FILTER" if it needs correction
  - "CLEAN" if it does not


# CLEANING PROMPT
# ---------------
# Purpose: Identify and rewrite only the paragraphs that need fixing
# Input: {numbered_chapter} - Chapter text with [PARAGRAPH N] markers
# Expected Output: Either "NONE" or list of "PARAGRAPH N: [corrected text]"
#
# Tips for customization:
#   - Explain the specific fixes you want applied
#   - Provide before/after examples for common issues
#   - Specify what to preserve (author's voice, style, etc.)
#   - Define when to rewrite vs. when to remove a paragraph
#
# OUTPUT FORMAT:
#   - If no changes needed: respond with "NONE"
#   - To rewrite a paragraph: "PARAGRAPH N: [corrected text here]"
#   - To remove a paragraph: "PARAGRAPH N: [PARAGRAPH REMOVED]"

cleaning_prompt: |
  You are editing a digitized book to fix OCR errors, scanning artifacts, and formatting issues while preserving the author's original voice and intent.

  INSTRUCTIONS:

  1. FIX OCR ERRORS:
     - Correct misread characters: "rn" misread as "m", "cl" as "d", "0" as "O"
     - Fix split words: "to-gether" should be "together"
     - Correct missing/extra spaces: "some thing" to "something", "alot" to "a lot"
     - Remove garbled characters or symbols from scanning errors

  2. FIX FORMATTING ISSUES:
     - Remove embedded page numbers, headers, or footers from body text
     - Fix paragraph breaks that were incorrectly merged or split
     - Correct list formatting that didn't convert properly
     - Remove duplicate content from scanning overlap

  3. FIX TEXT QUALITY:
     - Remove repeated words: "the the" to "the"
     - Complete obviously truncated sentences when possible
     - Fix systematic typos while preserving intentional spelling

  WHAT TO PRESERVE:
     - The author's original voice, style, and word choices
     - Intentional formatting like emphasis or paragraph structure
     - Period-appropriate spelling or archaic language
     - Technical terms, proper nouns, and specialized vocabulary
     - Dialogue formatting and attribution

  WHAT NOT TO CHANGE:
     - Do not "improve" the writing or modernize language
     - Do not change sentences that are correct but unusual
     - Do not add content that wasn't there
     - Do not change the meaning or intent of any passage

  RESPONSE FORMAT:

  If NO paragraphs need changes, respond with:
  NONE

  If paragraphs need changes, respond with:
  PARAGRAPH N: [complete corrected text]

  For paragraphs that should be removed entirely (e.g., duplicate content, page headers):
  PARAGRAPH N: [PARAGRAPH REMOVED]

  Example response:
  PARAGRAPH 3: The corrected text for paragraph three goes here with all OCR errors fixed.
  PARAGRAPH 7: [PARAGRAPH REMOVED]
  PARAGRAPH 12: Another corrected paragraph with proper spacing and punctuation.

  IMPORTANT:
  - Only include paragraphs that actually need changes
  - Provide the complete corrected paragraph text, not just the corrections
  - Preserve the exact paragraph structure - do not merge or split paragraphs
  - When in doubt, preserve the original text

  Here is the chapter with numbered paragraphs:

  {numbered_chapter}

  Provide your response:


# ============================================================================
# ADDITIONAL EXAMPLE USE CASES
# ============================================================================
#
# Below are alternative prompt ideas for different cleaning scenarios.
# Uncomment and modify as needed, or use as inspiration for your own prompts.
#
# ---
# EXAMPLE: Removing Watermarks and Copyright Notices
# ---
# analysis_prompt: |
#   Analyze if this chapter contains watermarks, copyright notices, or
#   promotional content that should be removed...
#
# ---
# EXAMPLE: Standardizing Dialogue Formatting
# ---
# analysis_prompt: |
#   Check if this chapter has inconsistent dialogue formatting such as
#   mixed quote styles, missing attribution, or incorrect punctuation...
#
# ---
# EXAMPLE: Fixing Character Encoding Issues
# ---
# analysis_prompt: |
#   Scan for character encoding problems like wrong quotation marks,
#   missing accented characters, or garbled special characters...
#
# ---
# EXAMPLE: Removing Duplicate Content
# ---
# analysis_prompt: |
#   Check if paragraphs or sentences are repeated due to scanning
#   overlap or conversion errors...
#
# ============================================================================