Importing Existing AI Instructions¶

The prs import command converts existing AI instruction files into PromptScript format. This allows teams to adopt PromptScript without losing their existing configurations.

Supported Formats¶

Quick Start¶

# Import a single file
prs import CLAUDE.md

# Preview the conversion
prs import CLAUDE.md --dry-run

# Import with roundtrip validation
prs import CLAUDE.md --validate

# Specify output directory
prs import CLAUDE.md --output ./my-project

How It Works¶

1. Detect Format - Automatically identifies the source format from the filename
2. Parse Sections - Extracts markdown headings and content blocks
3. Classify - Heuristically maps sections to PromptScript blocks (@identity, @standards, @restrictions, @knowledge, @context)
4. Score - Assigns a confidence score (0-100%) to each classification
5. Emit - Generates .prs output with # REVIEW: comments on low-confidence sections

Confidence Scoring¶

Each section receives a confidence score:

• HIGH (>80%) - Strong pattern match (e.g., "You are..." maps to @identity)
• MEDIUM (50-80%) - Ambiguous patterns requiring review
• LOW (<50%) - No clear match, classified as @context

Sections below HIGH confidence include # REVIEW: comments in the generated output. Always review the imported file before using it in production.

CLI Options¶

prs import <file> [options]

Options:
  -f, --format <format>   Force source format (claude, github, cursor, generic)
  -o, --output <dir>      Output directory (default: .promptscript)
  --dry-run               Preview output without writing files
  --validate              Run roundtrip validation after import

Roundtrip Validation¶

The --validate flag parses the generated .prs output with the PromptScript parser to verify it is syntactically valid. This catches any edge cases where the import produces invalid PromptScript syntax.

After Import¶

1. Review the generated imported.prs file
2. Check # REVIEW: comments and adjust block classifications
3. Add @meta fields (team, description) as needed
4. Set up promptscript.yaml with your compilation targets
5. Run prs compile to generate output for your AI tools

Programmatic Usage¶

import { importFile } from '@promptscript/importer';

const result = await importFile('./CLAUDE.md');
console.log(result.prsContent); // Generated .prs content
console.log(result.totalConfidence); // Average confidence (0-1)
console.log(result.warnings); // Low-confidence warnings
console.log(result.sections); // Classified sections