wobble-bibble 🕌

TypeScript library for Islamic text translation prompts with LLM output validation and prompt stacking utilities.

Installation

npm install wobble-bibble
# or
bun add wobble-bibble

Features

• Bundled Prompts: 8 optimized translation prompts (Hadith, Fiqh, Tafsir, etc.) with strongly-typed access
• Translation Validation: Catch LLM hallucinations like malformed segment IDs, Arabic leaks, forbidden terms
• Prompt Stacking: Master + specialized prompts combined automatically
• Gold Standards: High-fidelity reference dataset for benchmarking

Quick Start

Get Translation Prompts

import { getPrompt, getPrompts, getPromptIds } from 'wobble-bibble';

// Get a specific stacked prompt (strongly typed)
const hadithPrompt = getPrompt('hadith');
console.log(hadithPrompt.content); // Master + Hadith addon combined

// Get all available prompts
const allPrompts = getPrompts();

// Get list of prompt IDs for dropdowns
const ids = getPromptIds(); // ['master_prompt', 'hadith', 'fiqh', ...]

Validate LLM Output

import { validateTranslationResponse } from 'wobble-bibble';

const segments = [
    { id: 'P1234', text: '... Arabic source for P1234 ...' },
    { id: 'P1235', text: '... Arabic source for P1235 ...' },
];

const llmOutput = `P1234 - Translation of first segment
P1235 - Translation of second segment`;

const result = validateTranslationResponse(segments, llmOutput);
if (result.errors.length > 0) console.error(result.errors);

API Reference

Prompts

| Function | Description | |----------|-------------| | getPrompt(id) | Get a specific stacked prompt by ID (strongly typed) | | getPrompts() | Get all available stacked prompts | | getStackedPrompt(id) | Get just the prompt content string | | getMasterPrompt() | Get raw master prompt (for custom addons) | | getPromptIds() | Get list of available prompt IDs | | stackPrompts(master, addon) | Manually combine prompts |

Validation

| Function | Description | |----------|-------------| | validateTranslationResponse(segments, response) | Unified validator for LLM translation responses (IDs, Arabic leak, invented IDs, gaps, speaker-label drift, etc.) | | VALIDATION_ERROR_TYPE_INFO | Human-readable descriptions for each ValidationErrorType (for UI/logging) |

Utilities

| Function | Description | |----------|-------------| | formatExcerptsForPrompt(segments, prompt) | Format segments for LLM input | | normalizeTranslationText(text) | Normalize newlines and split merged markers onto separate lines | | extractTranslationIds(text) | Extract all segment IDs from "ID - ..." markers |

Available Prompts

| ID | Name | Use Case | |----|------|----------| | master_prompt | Master Prompt | Universal grounding rules | | hadith | Hadith | Isnad-heavy texts, Sharh | | fiqh | Fiqh | Legal terminology | | tafsir | Tafsir | Quranic exegesis | | fatawa | Fatawa | Q&A format | | encyclopedia_mixed | Encyclopedia Mixed | Polymath works | | jarh_wa_tadil | Jarh Wa Tadil | Narrator criticism | | usul_al_fiqh | Usul Al Fiqh | Legal methodology |

Prompt Development

See docs/refinement-guide.md for the methodology used to develop and test these prompts. See AI_REVIEW_PROMPT.md for the peer-review prompt template used when sending round packets to external agents.

License

MIT