migration-scoring

Comprehensive quality evaluation for code migration projects.

Overview

This plugin evaluates completed migrations through multiple lenses:

1. Mapping — Document source-to-target file relationships
2. Quality Scoring — Measure coverage and correctness
3. Style Scoring — Evaluate code quality and conventions
4. Reporting — Generate executive summary with recommendations

Prerequisites

• Completed migration with both source and target code present
• Python 3.13 with uv
• LLM API key (Anthropic or OpenAI)
• Optional: Custom style rubric file

Quick Start

export LLM_API_KEY="your-api-key"
export LLM_MODEL="anthropic/claude-3-5-sonnet-20241022"

uv run python -m lc_sdk_examples.migration_scoring \
  --src-path /path/to/migration/project \
  --rubric-path /path/to/style_rubric.txt

PowerShell equivalent for the environment setup:

$env:LLM_API_KEY = "your-api-key"
$env:LLM_MODEL = "anthropic/claude-3-5-sonnet-20241022"

uv run python -m lc_sdk_examples.migration_scoring `
  --src-path C:\path\to\migration\project `
  --rubric-path C:\path\to\style_rubric.txt

Workflow Phases

Phase 1: Migration Mapping

See ../migration-mapping/SKILL.md

Creates a source→target file mapping:

• Identifies which target files implement each source file
• Supports many-to-many relationships
• Flags unmigrated source files

Output: migration_mapping.json

{
  "CALC001.cbl": ["InvoiceCalculator.java", "TaxCalculator.java"],
  "CUST002.cbl": ["CustomerService.java"]
}

Phase 2: Quality Scoring

See ../score-quality/SKILL.md

Scores each source file on:

• Coverage (1-5): How much functionality was migrated
• Correctness (1-5): How accurately behavior was preserved

Output: migration_score.json

{
  "CALC001.cbl": {
    "coverage": 4,
    "correctness": 5,
    "justification": "All calculation logic migrated..."
  }
}

Phase 3: Style Scoring