@peacethinking/mediation-prompts

Production-tested prompt templates for AI-assisted conflict resolution.

Parameterized TypeScript functions that generate structured system/user message pairs for LLM-based mediation. Each prompt includes a Zod output schema for runtime validation.

Install

npm install @peacethinking/mediation-prompts

Usage

Generate a Verdict Prompt

import { buildVerdictPrompt, VerdictOutputSchema } from '@peacethinking/mediation-prompts';

const { system, user } = buildVerdictPrompt({
  viewpoint1: "My manager assigns me more work than my colleagues...",
  viewpoint2: "The project had tight deadlines and tasks were assigned by availability...",
  languageCode: 'en',
  tone: 'neutral',
});

// Send to any LLM
const response = await llm.chat([
  { role: 'system', content: system },
  { role: 'user', content: user },
]);

// Validate output
const result = VerdictOutputSchema.safeParse(JSON.parse(response));

Analyze a Single Perspective

import { buildPerspectivePrompt } from '@peacethinking/mediation-prompts';

const { system, user } = buildPerspectivePrompt({
  viewpoint: "I've been feeling overwhelmed since the reorganization...",
  languageCode: 'de',
});

Synthesize Two Viewpoints

import { buildMediationPrompt } from '@peacethinking/mediation-prompts';

const { system, user } = buildMediationPrompt({
  viewpoint1: "...",
  viewpoint2: "...",
  languageCode: 'en',
});
// Output: { summary, solution_proposals: string[] }

Generate Clarification Questions

import { buildClarificationPrompt } from '@peacethinking/mediation-prompts';

const { system, user } = buildClarificationPrompt({
  viewpoint: "...",
  languageCode: 'en',
});

API

| Function | Output | Description | |---|---|---| | buildVerdictPrompt() | Scores, outcome, reframing | Full 6-dimension verdict with scoring | | buildPerspectivePrompt() | Per-dimension analysis | Single viewpoint analysis | | buildMediationPrompt() | Summary + 3-5 proposals | Two-viewpoint synthesis | | buildClarificationPrompt() | 5-dimension questions | Follow-up clarification |

Each function returns { system: string, user: string } and exports a corresponding *OutputSchema for Zod validation.

License

MIT