convez.ai
SkillDados & AnalyticsIntermediário
BaseInfinity

Claude RDLC Wizard

BaseInfinity/claude-rdlc-wizard

Ciclo de vida de pesquisa com verificação de fontes e confiança

Ver no GitHub
InicialPluginScore · de 0 a 100

Framework de ciclo de vida para pesquisa em repositórios Claude Code, garantindo que claims sejam verificadas com fontes e níveis de confiança documentados. Implementa hooks, skills e templates para pesquisa acadêmica, análise forense e relatórios técnicos que exigem rastreabilidade completa.

Instalação

$npx convez-ai BaseInfinity/claude-rdlc-wizard

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 3 meses
Versão
—
Issues abertas
6
Contribuidores
1

Casos de uso

  • Preparação de entrevistas e dossiers de evidências
  • Análises forenses e investigações técnicas
  • Relatórios médicos, legais e de mercado com rastreabilidade
  • Validação cruzada de claims com scoring de confiança

Veredito

Ferramenta especializada e madura para pesquisa rigorosa em contextos onde credibilidade e rastreabilidade são críticas. Bem documentada com case studies reais; exige disciplina do usuário para máximo valor.

README

A tradução para o português deste README ainda está na fila.

claude-rdlc-wizard

RDLC enforcement for Claude Code — hooks, skills, and wizard setup for research repos. The research-domain sibling of claude-sdlc-wizard.

Status: v0.1.0 (bootstrap, 2026-05-04). Three case studies cleared the xdlc "two case studies → extract" threshold: anticheat, states-project-research, tucson-investigation. Pattern catalog lives in PATTERNS.md; proof-points in CASE_STUDIES.md; build journal in EXTRACTION_NOTES.md. The standalone ~/rdlc/ repo was consolidated into this wizard on 2026-05-04 (mirrors GDLC retirement pattern).

What this installs

A research lifecycle on top of any Claude Code project that produces sourced research deliverables — interview prep, medical/legal evidence dossiers, automotive diagnostics, market analyses, anything where claims need sources and confidence levels.

LayerWhat it does
RDLC.mdConsumer-installable canonical doc — confidence vocabulary, source hierarchy, slop gate, audience-firewall rule
/rdlc skillFull lifecycle workflow (Plan → Verify → Review → Ship → Improve)
/setup-rdlc skillConfidence-driven setup — scans repo, asks only what it can't detect
/update-rdlc skillDrift reconciliation when wizard upgrades
/feedback skillPrivacy-first contribution loop
HooksSlop scan, confidence-required, source-required, prompt-check, audience-firewall
Templatesregression_test.sh, slop_scan.sh, multi-deliverable generator scaffold

Install

curl -fsSL https://raw.githubusercontent.com/BaseInfinity/claude-rdlc-wizard/main/install.sh | bash

Or scaffold manually:

npx -y claude-rdlc-wizard init

Run inside a research repo. The wizard scans for evidence/sources/confidence-marker patterns and only asks what it can't infer.

Relationship to claude-sdlc-wizard

Both run in parallel. SDLC handles code quality; RDLC handles research correctness. From ~/xdlc/README.md:

SDLC without the domain DLC = clean code that produces wrong output. Domain DLC without SDLC = correct requirements built with sloppy code.

A research repo installs both:

curl -fsSL .../claude-sdlc-wizard/install.sh | bash   # code layer
curl -fsSL .../claude-rdlc-wizard/install.sh | bash   # research layer

Vocabulary mapping (vs SDLC)

SDLCRDLC
Write failing testForm hypothesis / define claim
Implement featureGather evidence / research
Run tests (green)Cross-validate sources
Lint / typecheckConfidence scoring
Regression testsFact regression tests
Code reviewCross-model adversarial review
TDD guard hooksSource verification hooks

What's in v0.1.0 vs deferred

In v0.1.0:

  • Skill triple (rdlc, setup, update, feedback)
  • Five hooks (prompt-check, slop-scan, confidence-required, source-required, audience-firewall)
  • Three templates (regression_test.sh, slop_scan.sh, generate_deliverable.py skeleton)
  • RDLC.md consumer canonical
  • install.sh + package.json

Deferred to later versions (per skills-first rule, earned through use):

  • Setup wizard scan logic refinement (starts with the 5-row signal table from xdlc cross-domain-concerns.md)
  • Per-domain RDLC presets (medical/legal, political, automotive, journalism)
  • Codex adapter (codex-rdlc-wizard)
  • L-code enumeration (anticheat A–G, tucson L1–L15 — earn RDLC's own through use)
  • npm registry publish (currently local install via npx --prefix against the cloned repo)

Documentation

  • RDLC.md — the consumer canonical that gets installed into target repos
  • PATTERNS.md — pattern catalog and lessons learned across case studies
  • CASE_STUDIES.md — proof-point cross-index + Contributions Inventory (artifact → wizard slot)
  • EXTRACTION_NOTES.md — build journal from the v0.1.0 extraction
  • WIZARD_PLAN.md — historical implementation plan (superseded by v0.1.0 ship)
  • HANDOFF.md — historical pre-build handoff (superseded)
  • CLAUDE.md — wizard self-instructions (this repo)
  • ARCHITECTURE.md — how the pieces fit
  • CHANGELOG.md — version history
  • ROADMAP.md — what's queued for v0.2 and beyond

Case studies that proved the patterns

See CASE_STUDIES.md for the full cross-index. Short version:

  • anticheat (160 commits, 316 tests) — medical/legal evidence aggregation. GRADE-aligned labels, certification queue, compound mechanism verification.
  • states-project-research (43 commits, 257 bash asserts) — political/interview prep. VERIFIED/SUPPORTED/INFERRED/UNVERIFIED, fact regression suite, multi-deliverable generator with audience firewall.
  • tucson-investigation (36 commits, 111 tests) — automotive diagnostic + dealer audit. DIRECT/SUPPORTED/INFERRED/GAP, Playwright persona tests, methodology leak gate, 9-round Codex xhigh loop.

License

MIT

AI Setup Lanes

Three recommended setups in AI_SETUP_LANES.md:

LanePlannerDriverReviewerWhen
A — Research PremiumOpus 4.6 maxOpus 4.6 maxGPT-5.5 xhighResearch drafting, confidence-critical, source evaluation
B — Research SaverOpus 4.6 maxSonnet (latest)GPT-5.5 xhighRoutine drafting, formatting, docs
C — Research LiteYouHaiku 4.5NoneData formatting, bibliography, file management, grunt work
Ver o original no GitHub

Bom para

Relacionados