output-quality

AI-assisted review, revision, and slide-fit compression for structured markdown documents.

This package is useful when you generate long-form markdown with an LLM and want a final quality pass before exporting to slides, docs, or PDFs.

What It Does

• Reviews markdown for completeness, professionalism, and publish-readiness
• Revises content using explicit editorial instructions
• Compresses oversized ## sections so they fit better into slide-oriented layouts
• Fails open by returning the original content if the model call errors

Install

npm install output-quality @anthropic-ai/sdk

Quick Start

import { createOutputQuality } from "output-quality";

const quality = createOutputQuality({
  apiKey: process.env.ANTHROPIC_API_KEY,
});

const review = await quality.reviewContent(markdown);

if (!review.approved) {
  markdown = await quality.reviseContent(
    markdown,
    review.revision_instructions
  );
}

markdown = await quality.compressForPptx(markdown, styleExamples);

API

createOutputQuality(options?)

Creates an OutputQuality instance.

Options:

• apiKey: Anthropic API key. Defaults to process.env.ANTHROPIC_API_KEY.
• reviewModel: Model used for review and compression. Defaults to claude-sonnet-4-6.
• reviseModel: Model used for revision. Defaults to claude-sonnet-4-6.

reviewContent(content)

Returns a structured quality assessment:

interface ReviewResult {
  approved: boolean;
  score: number;
  issues: string[];
  revision_instructions: string;
}

Typical review criteria:

• complete sections rather than outlines
• specific and credible claims
• clean formatting without AI filler or preamble
• useful structure for an informed reader

reviseContent(content, instructions, systemPrompt?)

Rewrites content using the instructions generated by reviewContent.

compressForPptx(content, styleExamples?)

Condenses verbose sections to fit slide-oriented exports while preserving:

• numbers and metrics
• named entities
• quotes
• table rows

If the compressed output is suspiciously short, the package falls back to the original input.

Utilities

The package also exports:

• PPTX_SECTION_CHAR_LIMIT
• countBodyChars(section)

Style Examples

compressForPptx accepts an optional styleExamples string. This can contain short excerpts that represent your preferred output style. The model uses those examples as a formatting reference during compression.

Notes

• Designed for Node.js and Worker-style runtimes
• Uses Anthropic as a peer dependency so host apps can control versioning
• Best suited for structured markdown with headings and section bodies