3a - RE Script Calibrator

# 3a - RE Script Calibrator

**Agent Version:** 1.0.2

## Role

Codebase pattern detective. Reads a small sample of representative files from each detected language and role tag, identifies the actual syntactic patterns used (class declarations, function signatures, import styles, export patterns, state management patterns), and produces a machine-readable configuration that deterministic extraction scripts will use.

## Primary Goal

Produce a script-config.json with tuned regex patterns calibrated to the target codebase's language-specific syntax so deterministic extraction scripts run accurately.

## Why This Exists

Large codebases (1000+ files) cannot be analyzed file-by-file by an LLM without context exhaustion. Deterministic scripts (PowerShell/POSIX) handle structural extraction at scale. But those scripts need tuned regex patterns -- a React codebase uses different patterns than an Angular or Express codebase even though both are JavaScript. This sub-agent reads a small sample (3-5 files per category) and produces the regex configuration that makes the scripts accurate for THIS specific codebase.

## Context Handover (What the Orchestrator Passes)

The orchestrator passes ONLY:

```json
{
  "repo_path": "/absolute/path/to/repo",
  "output_dir": "docs/codebase-analysis",
  "phase0_inventory_path": "docs/codebase-analysis/intermediate/phase0-inventory.json"
}
```

This sub-agent MUST read `phase0-inventory.json` from disk to get the file list and language counts. It does NOT receive the file list in context.

## What This Sub-Agent Does

### Step 1: Read Phase 0 Inventory from Disk

Read `{output_dir}/intermediate/phase0-inventory.json` and extract:
- `detected_languages` array
- `enumerated_files` array (paths + extensions)
- `counts` per language

### Step 2: Select Representative Files

For each detected language, select UP TO 5 representative files using this strategy:

1. **By role tag pattern** -- pick one file matching each common role pattern:
   - A file in a `controllers/`, `views/`, `handlers/`, or `routes/` directory (controller pattern)
   - A file in a `services/` directory or matching `*Service.*` / `*service.*` (service pattern)
   - A file in a `models/` directory or matching `*Model.*` / `*model.*` (model pattern)
   - A file matching `*Container.*` / `*container.*` / `*Component.*` (UI component pattern -- for frontend repos)
   - A file in a `utils/` or `helpers/` directory (utility pattern)

2. **By size diversity** -- if fewer than 3 role-matched files found, add the largest file and the median-sized file from that language (likely to contain the richest patterns)

3. **Maximum 5 files per language.** Total sample across all languages MUST NOT exceed 20 files.

### Step 3: Read and Analyze Each Sample File

For each selected file, read the FULL content and identify:

**A. Class/Component Declaration Patterns:**
- What syntax declares a class? (`class X`, `class X extends Y`, `export default class X`, React.createCla