Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
affaan-m avatar

Regex Vs Llm Structured Text

  • 5.8k installs
  • 238k repo stars
  • Updated August 5, 2026
  • affaan-m/everything-claude-code

Decision framework and hybrid pipeline for structured text parsing using regex first and LLM only on low-confidence edge cases.

About

This skill provides a practical decision framework for parsing structured text such as quizzes, forms, invoices, and documents by choosing between regular expressions and large language models. The core insight is that regex handles 95 to 98 percent of cases deterministically at low cost, reserving expensive LLM calls for low-confidence edge cases only. The documented architecture runs a regex parser first, applies confidence scoring to flag items below a 0.95 threshold, and routes only flagged items to a cheap LLM validator such as Claude Haiku. Production metrics from a 410-item quiz pipeline show 98 percent regex success with about five LLM calls and roughly 95 percent cost savings versus full-LLM parsing. Best practices mandate starting with regex even if imperfect, using TDD for parsers, returning new instances from cleaning steps, and recording regex success rate plus LLM call counts. Anti-patterns include sending all text to LLM when regex already covers 95 percent, using regex on highly variable free-form text, and skipping confidence scoring.

  • Decision tree: start regex when patterns are consistent, use LLM only for free-form text
  • Hybrid pipeline: regex parser, confidence scorer, then LLM validator for items below 0.95
  • Production quiz metrics: 98% regex success, ~5 LLM calls, ~95% cost savings on 410 items
  • Confidence flags for few choices, missing answers, and short text before LLM review
  • TDD-first parser development with metrics tracking for regex success and LLM call volume

Regex Vs Llm Structured Text by the numbers

  • 5,755 all-time installs (skills.sh)
  • +222 installs in the week ending Aug 5, 2026 (Skillselion tracking)
  • Ranked #135 of 16,546 AI & Agent Building skills by installs in the Skillselion catalog
  • Security screen: MEDIUM risk (skills.sh audit)
  • Data as of Aug 5, 2026 (Skillselion catalog sync)
At a glance

regex-vs-llm-structured-text capabilities & compatibility

Capabilities
decide regex vs llm · build regex parser · score extraction confidence · route edge cases to llm · track parser metrics
Works with
anthropic
Use cases
data analysis · token optimization · pdf parsing
From the docs

What regex-vs-llm-structured-text says it does

正規表現は低コストかつ決定論的に95〜98%のケースを処理できる。コストのかかるLLM呼び出しは残りのエッジケースに留める。
SKILL.md
正規表現が95%以上を処理できる場合に全テキストをLLMに送る(コスト高・低速)
SKILL.md
npx skills add https://github.com/affaan-m/everything-claude-code --skill regex-vs-llm-structured-text

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs5.8k
repo stars238k
Security audit3 / 3 scanners passed
Last updatedAugust 5, 2026
Repositoryaffaan-m/everything-claude-code

Should I parse structured text with regex or an LLM, and how do I minimize cost while keeping accuracy?

Choose regex or LLM for structured text parsing and build a hybrid pipeline that uses LLM only on low-confidence extractions.

Who is it for?

Teams parsing repetitive structured text like quizzes, forms, invoices, or document tables with cost constraints.

Skip if: Highly variable free-form prose where regex cannot reach 90 percent pattern coverage.

When should I use this skill?

User chooses between regex and LLM for text extraction, builds hybrid parsers, or optimizes parsing cost versus accuracy.

What you get

A regex-first extraction pipeline with confidence scoring and targeted LLM validation for flagged items only.

  • parsing strategy
  • regex patterns
  • LLM fallback rules

By the numbers

  • Regex handles an estimated 95–98% of structured text parsing cases

Files

SKILL.mdMarkdownGitHub ↗

構造化テキスト解析における正規表現 vs LLM

構造化テキスト(クイズ、フォーム、請求書、ドキュメント)を解析するための実用的な意思決定フレームワーク。核心的な洞察:正規表現は低コストかつ決定論的に95〜98%のケースを処理できる。コストのかかるLLM呼び出しは残りのエッジケースに留める。

使用場面

  • 繰り返しパターンを持つ構造化テキスト(設問、フォーム、表)の解析
  • テキスト抽出に正規表現とLLMのどちらを使うかの判断
  • 両方のアプローチを組み合わせたハイブリッドパイプラインの構築
  • テキスト処理におけるコスト/精度のトレードオフの最適化

意思決定フレームワーク

テキスト形式は一貫していて繰り返しがあるか?
├── はい (>90% が何らかのパターンに従う) → 正規表現から始める
│   ├── 正規表現が 95%+ を処理 → 完了、LLM は不要
│   └── 正規表現が <95% を処理 → エッジケースのみ LLM を追加
└── いいえ (自由形式、高度に可変) → LLM を直接使用

アーキテクチャパターン

[正規表現パーサー] ─── 構造を抽出(95〜98% の精度)
    │
    ▼
[テキストクリーナー] ─── ノイズを除去(マーカー、ページ番号、アーティファクト)
    │
    ▼
[信頼度スコアラー] ─── 信頼度の低い抽出結果にフラグを立てる
    │
    ├── 高信頼度(≥0.95)→ 直接出力
    │
    └── 低信頼度(<0.95)→ [LLM バリデーター] → 出力

実装

1. 正規表現パーサー(大半のケースを処理)

import re
from dataclasses import dataclass

@dataclass(frozen=True)
class ParsedItem:
    id: str
    text: str
    choices: tuple[str, ...]
    answer: str
    confidence: float = 1.0

def parse_structured_text(content: str) -> list[ParsedItem]:
    """Parse structured text using regex patterns."""
    pattern = re.compile(
        r"(?P<id>\d+)\.\s*(?P<text>.+?)\n"
        r"(?P<choices>(?:[A-D]\..+?\n)+)"
        r"Answer:\s*(?P<answer>[A-D])",
        re.MULTILINE | re.DOTALL,
    )
    items = []
    for match in pattern.finditer(content):
        choices = tuple(
            c.strip() for c in re.findall(r"[A-D]\.\s*(.+)", match.group("choices"))
        )
        items.append(ParsedItem(
            id=match.group("id"),
            text=match.group("text").strip(),
            choices=choices,
            answer=match.group("answer"),
        ))
    return items

2. 信頼度スコアリング

LLMによるレビューが必要かもしれない項目にフラグを立てる:

@dataclass(frozen=True)
class ConfidenceFlag:
    item_id: str
    score: float
    reasons: tuple[str, ...]

def score_confidence(item: ParsedItem) -> ConfidenceFlag:
    """Score extraction confidence and flag issues."""
    reasons = []
    score = 1.0

    if len(item.choices) < 3:
        reasons.append("few_choices")
        score -= 0.3

    if not item.answer:
        reasons.append("missing_answer")
        score -= 0.5

    if len(item.text) < 10:
        reasons.append("short_text")
        score -= 0.2

    return ConfidenceFlag(
        item_id=item.id,
        score=max(0.0, score),
        reasons=tuple(reasons),
    )

def identify_low_confidence(
    items: list[ParsedItem],
    threshold: float = 0.95,
) -> list[ConfidenceFlag]:
    """Return items below confidence threshold."""
    flags = [score_confidence(item) for item in items]
    return [f for f in flags if f.score < threshold]

3. LLM バリデーター(エッジケースのみ)

def validate_with_llm(
    item: ParsedItem,
    original_text: str,
    client,
) -> ParsedItem:
    """Use LLM to fix low-confidence extractions."""
    response = client.messages.create(
        model="claude-haiku-4-5-20251001",  # Cheapest model for validation
        max_tokens=500,
        messages=[{
            "role": "user",
            "content": (
                f"Extract the question, choices, and answer from this text.\n\n"
                f"Text: {original_text}\n\n"
                f"Current extraction: {item}\n\n"
                f"Return corrected JSON if needed, or 'CORRECT' if accurate."
            ),
        }],
    )
    # Parse LLM response and return corrected item...
    return corrected_item

4. ハイブリッドパイプライン

def process_document(
    content: str,
    *,
    llm_client=None,
    confidence_threshold: float = 0.95,
) -> list[ParsedItem]:
    """Full pipeline: regex -> confidence check -> LLM for edge cases."""
    # Step 1: Regex extraction (handles 95-98%)
    items = parse_structured_text(content)

    # Step 2: Confidence scoring
    low_confidence = identify_low_confidence(items, confidence_threshold)

    if not low_confidence or llm_client is None:
        return items

    # Step 3: LLM validation (only for flagged items)
    low_conf_ids = {f.item_id for f in low_confidence}
    result = []
    for item in items:
        if item.id in low_conf_ids:
            result.append(validate_with_llm(item, content, llm_client))
        else:
            result.append(item)

    return result

実際のメトリクス

本番のクイズ解析パイプライン(410項目)より:

メトリクス
正規表現の成功率98.0%
低信頼度項目8 (2.0%)
必要なLLM呼び出し回数~5
全件LLM比のコスト節約~95%
テストカバレッジ93%

ベストプラクティス

  • 正規表現から始める — 不完全な正規表現でも改善のベースラインになる
  • 信頼度スコアリングを使用して、LLMの助けが必要なものをプログラムで特定する
  • 最も安価なLLMを使用して検証する(Haikuクラスのモデルで十分)
  • 解析済み項目を変更しない — クリーニング/検証ステップから新しいインスタンスを返す
  • TDDは解析器に効果的 — まず既知のパターンのテストを書き、次にエッジケースを書く
  • メトリクスを記録(正規表現の成功率、LLM呼び出し回数)してパイプラインの健全性を追跡する

避けるべきアンチパターン

  • 正規表現が95%以上を処理できる場合に全テキストをLLMに送る(コスト高・低速)
  • 自由形式で高度に可変なテキストに正規表現を使用する(LLMの方が適切)
  • 信頼度スコアリングをスキップして正規表現が「うまくいく」ことを期待する
  • クリーニング/検証ステップで解析済みオブジェクトを変更する
  • エッジケースをテストしない(不正な入力、欠損フィールド、エンコーディング問題)

適用場面

  • クイズ/試験問題の解析
  • フォームデータの抽出
  • 請求書/レシートの処理
  • ドキュメント構造の解析(見出し、セクション、表)
  • 繰り返しパターンがあり、コストが重要なあらゆる構造化テキスト

Related skills

Forks & variants (1)

Regex Vs Llm Structured Text has 1 known copy in the catalog totaling 1.4k installs. They canonicalize to this original listing.

FAQ

When should I start with regex instead of an LLM?

When more than 90 percent of the text follows repeatable patterns; regex often handles 95 to 98 percent deterministically.

How does the hybrid pipeline limit LLM cost?

Regex extracts first, confidence scoring flags items below 0.95, and only those items go to a cheap LLM validator.

What anti-pattern should I avoid?

Sending all text to an LLM when regex already handles 95 percent or more; that is costly and slow.

Is Regex Vs Llm Structured Text safe to install?

skills.sh reports 3 of 3 security scanners passed. Review the Security Audits panel on this page before installing in production.

AI & Agent Buildingllmautomation

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.