ThomasMoreAI/legal-skills-open / br/litigation/skills/juriscan-brunobracaioli
juriscan-brunobracaioli
VerifiedAnálise forense de processos judiciais brasileiros. Extrai, classifica e cruza peças processuais, detecta contradições, calcula prazos CPC e gera relatórios de risco com exportação para Obsidian. Use quando mencionar: processo judicial, petição, sentença, acórdão, contradições, prazos, timeline, análise forense, Obsidian.
scripts/integrity_check.py
View on GitHub#!/usr/bin/env python3
"""
integrity_check.py — Document integrity and OCR quality verification.
Detects:
- OCR quality issues (garbled text, encoding anomalies)
- Metadata anomalies (missing pieces, date inversions, duplicate labels)
- Page gaps (missing pages based on markers)
Usage:
python3 integrity_check.py --input ./analysis/
"""
from __future__ import annotations
import argparse
import json
import os
import re
import sys
from pathlib import Path
# Add utils to path
sys.path.insert(0, os.path.dirname(__file__))
from utils.dates import extract_all_dates, validate_chronology
# Portuguese dictionary words for OCR confidence (common legal terms)
_LEGAL_TERMS = {
'processo', 'autor', 'réu', 'juiz', 'vara', 'sentença', 'acórdão',
'petição', 'contestação', 'réplica', 'recurso', 'apelação', 'agravo',
'decisão', 'despacho', 'certidão', 'mandado', 'intimação', 'citação',
'direito', 'lei', 'artigo', 'código', 'tribunal', 'justiça', 'foro',
'comarca', 'advogado', 'procurador', 'ministério', 'público', 'federal',
'estadual', 'civil', 'penal', 'trabalho', 'consumidor', 'constituição',
'danos', 'morais', 'materiais', 'indenização', 'condenação', 'honorários',
'custas', 'valor', 'causa', 'pedido', 'prova', 'testemunha', 'perito',
'laudo', 'parecer', 'audiência', 'julgamento', 'votação', 'relator',
'ementa', 'voto', 'dispositivo', 'procedente', 'improcedente', 'parcial',
'prazo', 'dias', 'úteis', 'recurso', 'embargos', 'contrarrazões',
}
# Pattern for consecutive consonants (OCR artifact indicator)
_CONSONANT_CLUSTER = re.compile(r'[bcdfghjklmnpqrstvwxyz]{5,}', re.IGNORECASE)
# Page marker patterns
_PAGE_MARKERS = [
re.compile(r'(?i)fls?\.\s*(\d+)'),
re.compile(r'(?i)Página\s+(\d+)\s+de\s+(\d+)'),
re.compile(r'(?i)pág(?:ina)?\.\s*(\d+)'),
]
# Expected piece ordering for anomaly detection
_EXPECTED_PIECES = [
'PETIÇÃO INICIAL', 'CONTESTAÇÃO', 'RÉPLICA', 'SENTENÇA',
'APELAÇÃO', 'CONTRARRAZÕES', 'ACÓRDÃO',
]
def calculate_ocr_confidence(text: str) -> float:
"""Calculate heuristic OCR confidence score (0.0-1.0).
Factors:
- Ratio of recognized words vs garbage
- Presence of expected legal patterns
- Character encoding anomalies
- Consecutive consonant clusters (OCR artifacts)
"""
if not text or len(text) < 50:
return 0.0
words = re.findall(r'\b[a-záàâãéèêíïóôõöúçñ]{3,}\b', text.lower())
if not words:
return 0.0
# Factor 1: Legal term density (0-0.3)
legal_count = sum(1 for w in words if w in _LEGAL_TERMS)
legal_density = min(legal_count / max(len(words), 1), 0.3)
legal_score = legal_density / 0.3 * 0.3
# Factor 2: Vowel presence ratio (0-0.4)
# Real Portuguese words contain vowels; OCR garbage typically has low vowel ratios
all_tokens = text.split()
if not all_tokens:
return 0.0
vowel_pattern = re.compile(r'[aeiouáàâãéèêíïóôõöú]', re.IGNORECASE)
tokens_with_vowels = sum(1 for t in all_tokens if len(t) >= 3 and vowel_pattern.search(t))
long_tokens = sum(1 for t in all_tokens if len(t) >= 3)
vowel_ratio = tokens_with_vowels / max(long_tokens, 1)
word_score = vowel_ratio * 0.4
# Factor 3: No consonant clusters (0-0.15)
clusters = _CONSONANT_CLUSTER.findall(text[:5000])
cluster_penalty = min(len(clusters) * 0.03, 0.15)
cluster_score = 0.15 - cluster_penalty
# Factor 4: Encoding quality (0-0.15)
# Check for replacement characters, null bytes, control chars
bad_chars = len(re.findall(r'[\x00-\x08\x0b\x0c\x0e-\x1f\ufffd]', text[:5000]))
encoding_penalty = min(bad_chars * 0.03, 0.15)
encoding_score = 0.15 - encoding_penalty
total = legal_score + word_score + cluster_score + encoding_score
return round(min(max(total, 0.0), 1.0), 3)
def detect_metadata_anomalies(chunks: list[dict]) -> list[dict]:
"""Flag metadata anomalies in the chunk set.
Checks for:
- Date chronology violations
- Missing expected pieces
- Duplicate piece labels
- Unusually short/long pieces relative to type
"""
anomalies = []
# Date chronology
chrono_issues = validate_chronology(chunks)
anomalies.extend(chrono_issues)
# Missing expected pieces (only flag if we have both petição and sentença)
labels = [c.get('label', '') for c in chunks]
has_peticao = 'PETIÇÃO INICIAL' in labels
has_sentenca = 'SENTENÇA' in labels
if has_peticao and has_sentenca and 'CONTESTAÇÃO' not in labels:
anomalies.append({
'tipo': 'PECA_FALTANTE',
'descricao': 'Contestação não detectada entre Petição Inicial e Sentença',
'impacto': 'MÉDIO',
'recomendacao': 'Verificar se é revelia ou se a contestação não foi detectada pelo chunking',
})
# Duplicate labels
label_counts: dict[str, int] = {}
for label in labels:
label_counts[label] = label_counts.get(label, 0) + 1
for label, count in label_counts.items():
if count > 1 and label not in ('DESPACHO', 'CERTIDÃO', 'OFÍCIO', 'MANDADO/CARTA'):
anomalies.append({
'tipo': 'PECA_DUPLICADA',
'descricao': f"'{label}' aparece {count} vezes — verificar se são peças distintas ou chunking duplicado",
'impacto': 'BAIXO',
'recomendacao': 'Revisar manualmente se são peças distintas (ex: sentença de mérito + sentença de embargos)',
})
# Process number changes
numeros = set()
for c in chunks:
n = c.get('processo_number') or c.get('numero_processo')
if n:
numeros.add(n)
if len(numeros) > 1:
anomalies.append({
'tipo': 'NUMERO_PROCESSO_DIVERGENTE',
'descricao': f"Múltiplos números de processo detectados: {', '.join(numeros)}",
'impacto': 'ALTO',
'recomendacao': 'Verificar se o PDF contém processos distintos ou se há erro de OCR nos números',
})
return anomalies
def detect_page_gaps(full_text: str, pdf_info: dict) -> list[dict]:
"""Flag potential missing pages based on page markers in text."""
gaps = []
# Extract page numbers from "fls." markers
page_numbers: list[int] = []
for pattern in _PAGE_MARKERS:
for m in pattern.finditer(full_text):
try:
page_numbers.append(int(m.group(1)))
except (ValueError, IndexError):
continue
if len(page_numbers) < 3:
return gaps
page_numbers = sorted(set(page_numbers))
# Find gaps > 5 pages
for i in range(len(page_numbers) - 1):
gap = page_numbers[i + 1] - page_numbers[i]
if gap > 5:
gaps.append({
'tipo': 'LACUNA_PAGINAS',
'descricao': f"Possível lacuna de {gap} páginas entre fls. {page_numbers[i]} e fls. {page_numbers[i+1]}",
'impacto': 'MÉDIO',
'recomendacao': 'Verificar se há páginas faltantes ou se a numeração é descontínua',
})
# Check against total pages from PDF info
total_pages = pdf_info.get('pages', 0)
if total_pages > 0 and page_numbers:
max_fls = max(page_numbers)
if max_fls > total_pages * 1.5:
gaps.append({
'tipo': 'NUMERACAO_INCOMPATIVEL',
'descricao': f"Maior número de fls. ({max_fls}) excede total de páginas do PDF ({total_pages})",
'impacto': 'BAIXO',
'recomendacao': 'Numeração de fls. dos autos ≠ páginas do PDF — normal em processos digitalizados',
})
return gaps
def generate_integrity_report(
chunks: list[dict],
full_text: str = '',
pdf_info: dict | None = None,
) -> dict:
"""Aggregate all integrity checks into a single report."""
pdf_info = pdf_info or {}
# OCR confidence per chunk
ocr_scores = []
for chunk in chunks:
text = chunk.get('text', '')
score = calculate_ocr_confidence(text)
ocr_scores.append({
'chunk_index': chunk.get('index', 0),
'label': chunk.get('label', '?'),
'score': score,
})
overall_confidence = (
sum(s['score'] for s in ocr_scores) / len(ocr_scores) if ocr_scores else 0.0
)
# Metadata anomalies
anomalies = detect_metadata_anomalies(chunks)
# Page gaps
page_gaps = detect_page_gaps(full_text, pdf_info) if full_text else []
# Low confidence warnings
low_confidence = [s for s in ocr_scores if s['score'] < 0.7]
recommendations = []
if low_confidence:
recommendations.append(
f"{len(low_confidence)} chunk(s) com OCR confidence < 0.7 — "
"considerar re-extração com vision ou OCR de maior qualidade"
)
high_impact = [a for a in anomalies if a.get('impacto') == 'ALTO']
if high_impact:
recommendations.append(
f"{len(high_impact)} anomalia(s) de alto impacto detectada(s) — revisar antes de prosseguir"
)
return {
'overall_confidence': round(overall_confidence, 3),
'ocr_scores': ocr_scores,
'anomalies': anomalies,
'page_gaps': page_gaps,
'recommendations': recommendations,
}
def main():
parser = argparse.ArgumentParser(description='Check document integrity and OCR quality')
parser.add_argument('--input', '-i', required=True, help='Analysis directory (with index.json and chunks/)')
args = parser.parse_args()
input_dir = args.input
index_path = os.path.join(input_dir, 'index.json')
if not os.path.exists(index_path):
print(f"[ERROR] index.json not found in {input_dir}", file=sys.stderr)
sys.exit(1)
with open(index_path, 'r', encoding='utf-8') as f:
index = json.load(f)
# Load chunks with text
chunks = []
for chunk_meta in index.get('chunks', []):
chunk_file = os.path.join(input_dir, chunk_meta.get('chunk_file', ''))
text = ''
if os.path.exists(chunk_file):
with open(chunk_file, 'r', encoding='utf-8') as f:
text = f.read()
chunk = {**chunk_meta, 'text': text}
chunks.append(chunk)
# Load full text if available
full_text_path = os.path.join(input_dir, 'full_text.txt')
full_text = ''
if os.path.exists(full_text_path):
with open(full_text_path, 'r', encoding='utf-8') as f:
full_text = f.read()
report = generate_integrity_report(
chunks,
full_text=full_text,
pdf_info=index.get('pdf_info', {}),
)
# Save report
output_path = os.path.join(input_dir, 'integrity_report.json')
with open(output_path, 'w', encoding='utf-8') as f:
json.dump(report, f, ensure_ascii=False, indent=2)
print(f"\nIntegrity Report:")
print(f" Overall confidence: {report['overall_confidence']:.1%}")
print(f" Anomalies: {len(report['anomalies'])}")
print(f" Page gaps: {len(report['page_gaps'])}")
for rec in report['recommendations']:
print(f" [!] {rec}")
print(f" Output: {output_path}")
if __name__ == '__main__':
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |