Skip to content

ThomasMoreAI/legal-skills-open / br/litigation/skills/juriscan-brunobracaioli

juriscan-brunobracaioli

Verified

Análise 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.

FileKindSize
.claude/agents/juriscan-advogado-autor.mdmarkdown4.3 KB
.claude/agents/juriscan-advogado-reu.mdmarkdown3.6 KB
.claude/agents/juriscan-auditor-processual.mdmarkdown7.0 KB
.claude/agents/juriscan-echo.mdmarkdown2.0 KB
.claude/agents/juriscan-parser.mdmarkdown4.7 KB
.claude/agents/juriscan-segmenter.mdmarkdown4.9 KB
.claude/agents/juriscan-sintetizador.mdmarkdown3.1 KB
.claude/agents/juriscan-verificador.mdmarkdown4.8 KB
.github/FUNDING.ymlconfig77 B
.github/workflows/ci.ymlconfig1.0 KB
.gitignoreother121 B
CHANGELOG.mdmarkdown20.6 KB
CLAUDE.mdmarkdown10.6 KB
LICENSEother1.0 KB
README.mdmarkdown16.9 KB
SKILL.mdskill md26.3 KB
assets/banner.svgother4.7 KB
docs/architecture.mdmarkdown10.3 KB
docs/index.htmlother4.3 KB
docs/operations.mdmarkdown6.7 KB
docs/subagent_contract.mdmarkdown6.0 KB
docs/subagents.mdmarkdown5.5 KB
install.shscript3.2 KB
pyproject.tomlmanifest272 B
references/agent_schemas/advogado_output.jsonconfig2.7 KB
references/agent_schemas/auditor_output.jsonconfig4.8 KB
references/agent_schemas/echo_output.jsonconfig666 B
references/agent_schemas/parser_output.jsonconfig4.4 KB
references/agent_schemas/recommendations_output.jsonconfig2.5 KB
references/agent_schemas/segmenter_output.jsonconfig2.9 KB
references/agent_schemas/sintetizador_output.jsonconfig2.7 KB
references/agent_schemas/verificador_output.jsonconfig2.3 KB
references/appellate_structure.mdmarkdown2.0 KB
references/binding_precedents.mdmarkdown2.2 KB
references/brazilian_legal_entities.mdmarkdown3.2 KB
references/chunk_analysis_schema.jsonconfig6.7 KB
references/cpc_prazos.jsonconfig4.4 KB
references/feriados_forenses.jsonconfig2.1 KB
references/output_schema.jsonconfig11.0 KB
references/output_schema_v2.jsonconfig11.0 KB
references/output_schema_v3.jsonconfig5.3 KB
references/piece_type_taxonomy.jsonconfig7.5 KB
references/prompt_templates.mdmarkdown11.8 KB
references/risk_scoring_rubric.mdmarkdown2.4 KB
references/whitelist_fontes.jsonconfig1.8 KB
requirements.txtmanifest250 B
scripts/__init__.pyscript0 B
scripts/agent_io.pyscript12.2 KB
scripts/analyzed_init.pyscript3.3 KB
scripts/audit.pyscript4.8 KB
scripts/cleanup_audit.pyscript2.0 KB
scripts/confidence_rules.pyscript6.4 KB
scripts/content_quality_check.pyscript11.1 KB
scripts/enrich_deterministic.pyscript10.3 KB
scripts/extract_and_chunk.pyscript14.3 KB
scripts/finalize.pyscript10.2 KB
scripts/finalize_legacy.pyscript7.4 KB
scripts/generate_report.pyscript29.8 KB
scripts/instance_tracker.pyscript6.7 KB
scripts/integrity_check.pyscript11.0 KB
scripts/integrity_gate.pyscript4.6 KB
scripts/legacy/__init__.pyscript0 B
scripts/legacy/contradiction_report.pyscript11.9 KB
scripts/legacy/risk_scorer.pyscript10.2 KB
scripts/merge_chunk_analysis.pyscript14.8 KB
scripts/migrate_v2_to_v3.pyscript3.1 KB
scripts/obsidian_export.pyscript28.0 KB
scripts/persist_chunks.pyscript6.2 KB
scripts/prazo_calculator.pyscript14.2 KB
scripts/report_metrics.pyscript8.4 KB
scripts/schema_validator.pyscript5.2 KB
scripts/utils/__init__.pyscript0 B
scripts/utils/cnj.pyscript4.4 KB
scripts/utils/dates.pyscript7.8 KB
scripts/utils/filenames.pyscript2.6 KB
scripts/utils/monetary.pyscript5.2 KB
tests/__init__.pyscript0 B
tests/conftest.pyscript1.1 KB
tests/fixtures/agent_io/echo_invalid_missing_field.jsonconfig45 B
tests/fixtures/agent_io/echo_invalid_wrong_const.jsonconfig70 B
tests/fixtures/agent_io/echo_valid.jsonconfig103 B
tests/fixtures/agent_io/segmenter_valid.jsonconfig359 B
tests/fixtures/sample_analyzed.jsonconfig10.0 KB
tests/fixtures/sample_analyzed_with_prazos.jsonconfig470 B
tests/fixtures/sample_chunk_acordao.txtmarkdown3.4 KB
tests/fixtures/sample_chunk_peticao.txtmarkdown3.4 KB
tests/fixtures/sample_chunk_sentenca.txtmarkdown2.9 KB
tests/fixtures/sample_index.jsonconfig1.6 KB
tests/golden/_pdf_builder.pyscript6.1 KB
tests/golden/processo_01_sintetico_simples/expected_invariants.jsonconfig993 B
tests/golden/processo_01_sintetico_simples/source.txtmarkdown3.3 KB
tests/golden/processo_02_sintetico_art942/expected_art_942.jsonconfig898 B
tests/golden/processo_02_sintetico_art942/expected_invariants.jsonconfig1.3 KB
tests/golden/processo_02_sintetico_art942/source.txtmarkdown7.8 KB
tests/test_agent_io.pyscript4.4 KB
tests/test_analyzed_init.pyscript5.6 KB
tests/test_audit.pyscript4.4 KB
tests/test_cnj.pyscript2.9 KB
tests/test_confidence_rules.pyscript4.8 KB
tests/test_content_quality_check.pyscript14.5 KB
tests/test_contradiction_report.pyscript3.5 KB
tests/test_dates.pyscript5.0 KB
tests/test_enrich_deterministic.pyscript6.3 KB
tests/test_filenames.pyscript2.1 KB
tests/test_finalize.pyscript5.8 KB
tests/test_finalize_legacy.pyscript6.5 KB
tests/test_generate_report.pyscript24.0 KB
tests/test_golden_regression.pyscript6.3 KB
tests/test_instance_tracker.pyscript4.2 KB
tests/test_integrity_check.pyscript4.8 KB
tests/test_integrity_gate.pyscript3.6 KB
tests/test_merge_chunk_analysis.pyscript20.9 KB
tests/test_migrate_v2_to_v3.pyscript4.3 KB
tests/test_monetary.pyscript3.4 KB
tests/test_obsidian_export_v3.pyscript6.9 KB
tests/test_persist_chunks.pyscript6.1 KB
tests/test_prazo_calculator.pyscript14.1 KB
tests/test_report_metrics.pyscript5.3 KB
tests/test_risk_scorer.pyscript5.7 KB
tests/test_whitelist_enforcement.pyscript6.1 KB

docs/architecture.md

View on GitHub
# Juriscan — Architecture

> **v3.1.x note:** the legacy pipeline was significantly upgraded in
> v3.1.0-legacy / v3.1.1 / v3.1.2 with a per-chunk file pattern
> (`analyzed_init.py` + `merge_chunk_analysis.py`), an executive report
> generator (`generate_report.py`), strategic recommendations with
> verbatim evidence, and Lei 14.905/2024 monetary recalculation. See
> `CHANGELOG.md` and the updated legacy pipeline flow in `CLAUDE.md`.
> This document describes primarily the agents pipeline architecture,
> which remains the long-term target but has not been validated end-to-end
> against real PDFs yet.

## TL;DR

Juriscan is a Claude Code skill that analyzes Brazilian legal proceedings.
It has two pipelines during the v2→v3 transition, and the agents pipeline
is the long-term target.

- **Python owns every deterministic computation.** Dates, BRL normalization,
  CNJ parsing, CPC deadline math, schema validation, Decimal money arithmetic,
  Obsidian export.
- **Native Claude Code subagents own every semantic judgment.** Segmentation,
  parsing, adversarial dialectic (advogado-autor × advogado-reu × auditor),
  jurisprudence verification, synthesis.
- **Zero ANTHROPIC_API_KEY.** The skill uses the user's Claude Code
  subscription; subagents are invoked via the `Task` tool from `SKILL.md`.
- **Every LLM output is JSON-schema-validated and logged** before any
  Python script consumes it.

## Pipeline overview

```
                           ┌─────────────────────┐
                           │        PDF          │
                           └─────────┬───────────┘
                                     │
                          ┌──────────▼──────────┐
                          │ scripts/extract_pdf │  (legacy reuses extract_and_chunk.py)
                          └──────────┬──────────┘
                                     │ raw_text.txt + page_map.json
                                     │
                          ┌──────────▼──────────┐
                          │ juriscan-segmenter  │  [Task, model=haiku]
                          └──────────┬──────────┘
                                     │ /tmp/<run_id>-segmenter.json
                                     │     ▲ agent_io.py validate + log
                                     │
                          ┌──────────▼──────────┐
                          │ persist_chunks.py   │  — writes chunks/*.txt + index.json
                          └──────────┬──────────┘   with integrity gate
                                     │
                   ┌─────────────────┴─────────────────┐
                   │   Task × N (parallel, one message)│
                   │    juriscan-parser per chunk      │
                   └─────────────────┬─────────────────┘
                                     │
                          ┌──────────▼──────────┐
                          │ enrich_deterministic│  — utils.dates, utils.monetary, utils.cnj
                          └──────────┬──────────┘   raises if mismatch rate > 10%
                                     │
          ┌──────────────────────────┼──────────────────────────┐
          │           Task × 3 (parallel, one message)          │
          │  advogado-autor    advogado-reu   auditor-processual │
          │  (sonnet)          (sonnet)       (sonnet)           │
          └──────────────────────────┬──────────────────────────┘
                                     │
                          ┌──────────▼──────────┐
                          │ juriscan-verificador │  [Task + WebFetch, whitelist]
                          └──────────┬──────────┘
                                     │
                          ┌──────────▼──────────┐
                          │ juriscan-sintetizador│  (sonnet)
                          └──────────┬──────────┘
                                     │
                          ┌──────────▼──────────┐
                          │ confidence_rules.py │  — preservation invariant
                          └──────────┬──────────┘     + verification downgrade
                                     │
                          ┌──────────▼──────────┐
                          │   finalize.py       │  — Lei 14.905 split, honorarios
                          └──────────┬──────────┘     (Decimal arithmetic)
                                     │
                          ┌──────────▼──────────┐
                          │ obsidian_export.py  │  — _AUDITORIA, _VERIFICAÇÕES,
                          └─────────────────────┘     _PERSPECTIVAS added in v3
```

## Hard invariants

1. **Chunks in analyzed.json have 1:1 correspondence with files on disk.**
   Enforced by `scripts/integrity_gate.py` and called from both
   `schema_validator.py` and `persist_chunks.py`.
2. **`likely_range.min ≤ likely_range.max`** in all risk outputs.
   Enforced by `scripts/legacy/risk_scorer.py` via sorted() + assert.
3. **Legitimate appellate reforms are not contradictions.**
   `scripts/legacy/contradiction_report.py` emits `REFORMA_PARCIAL` notes
   in `instance_tracking` when sentença→acórdão values differ.
4. **Sintetizador cannot drop auditor findings.**
   `scripts/confidence_rules.py` raises `ConfidenceRuleError` if violated.
5. **Every Task call is logged to the audit trail.** Even failures
   (with `error=...` and `schema_valid=false`).
6. **Web lookup only in the whitelist.** `scripts/agent_io.py` rejects
   verificador outputs whose `source_url` host is not in
   `references/whitelist_fontes.json`, as a post-hoc hard gate.
7. **Monetary mismatch rate >10% aborts the pipeline** at the
   `enrich_deterministic.py` step. The dialectic layer never runs on
   unreliable parser output.
8. **Budget gate:** p95 latency per run ≤ 10 minutes and total Task
   invocations ≤ 30 for a 15-piece process. Enforced by
   `scripts/report_metrics.py --enforce` in CI.

## Schema versions

- **v2** — `references/output_schema_v2.json`. Output of the legacy
  pipeline. `schema_version` field absent or `"2.0"`.
- **v3** — `references/output_schema_v3.json`. Superset with
  `run_id`, `pipeline_mode`, `process_state`, `perspectives`,
  `auditor_findings`, `verifications`, `verification_summary`,
  `monetary_recalculations`, `dissensos`, `resumo_executivo`,
  `audit_trail_uri`. `schema_version == "3.0"`.

`scripts/schema_validator.py` auto-dispatches by the `schema_version`
field. `scripts/migrate_v2_to_v3.py` performs additive-only migration
(keys in the input are never overwritten).

## Directory map

```
.claude/agents/           7 subagent markdown files — juriscan-*
references/               schemas, taxonomies, whitelists, binding precedents
  agent_schemas/          8 JSON schemas for subagent outputs
  output_schema_v2.json   legacy schema (preserved)
  output_schema_v3.json   new schema, superset of v2
  whitelist_fontes.json   authoritative hosts for verificador
scripts/
  agent_io.py             glue CLI: new-run / validate / log / extract-field
  audit.py                append-only JSONL logger
  persist_chunks.py       writes chunks to disk from segmenter output
  enrich_deterministic.py normalizes dates/money/CNJ, makes dialectic digests
  confidence_rules.py     preservation invariant + verification downgrade
  finalize.py             Lei 14.905 + honorarios post-reform (Decimal)
  report_metrics.py       audit trail aggregation + budget gate
  migrate_v2_to_v3.py     v2 → v3 migration
  schema_validator.py     dispatches by schema_version
  integrity_gate.py       1:1 chunk-file ↔ JSON check
  prazo_calculator.py     CPC deadlines, state-aware
  obsidian_export.py      Obsidian vault with v3 views
  legacy/                 retired scripts still consumed by --pipeline=legacy
tests/                    435+ pytest unit/integration tests (v3.1.2)
docs/                     architecture.md, subagents.md, operations.md
```

## How the two pipelines coexist

During Phases 2–5 the agents pipeline is **opt-in** via `/juriscan --pipeline=agents`.
The legacy pipeline remains the default. Flip is the subject of Phase 6
Step 6.4 — requires the agents suite to pass the golden regression three
times in a row, then `SKILL.md` changes the default and the legacy mode
becomes `--pipeline=legacy` explicit for three more releases before removal.

Schema v2 never dies silently. `migrate_v2_to_v3.py` converts any v2
artifact into v3 additively, preserving the v2 value set unchanged.

## Prompt caching

System prompts for each subagent live in the body of
`.claude/agents/juriscan-*.md` and are stable across invocations — Claude
Code's automatic prompt cache retains them. Dynamic inputs come only
via the `prompt` parameter of each `Task` call, so the stable portion
stays at the beginning of the context window and cache hits are
maximized.

The `enrich_deterministic.make_dialectic_summary()` helper produces
compact digests of the parsed pieces for the dialectic layer, so the
three parallel advogados/auditor subagents receive structured facts
instead of raw chunk text — further reducing token count and
improving cache stability across their parallel invocations.
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
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: 196kUpdated Claude CodeCodexVerifiedInstall