convez.ai
MCP ServerBusca & PesquisaAvançado
ehurrn

Newsroom Extension

ehurrn/newsroom-extension

Toolkit de jornalismo investigativo com OSINT, FOIA e defesa legal

Ver no GitHub
InicialPluginScore · de 0 a 100

Kit de jornalismo investigativo alimentado por IA com 18 skills para OSINT, engenharia FOIA, rastreamento corporativo, preservação de evidências e fluxo editorial completo. Funciona com Claude, Gemini CLI e agentes compatíveis com AGENTS.md, capacitando jornalistas humanos em cada etapa da investigação — do lead inicial até publicação segura.

Instalação

$npx convez-ai ehurrn/newsroom-extension

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
10
Forks
3
Acompanhando
1
Saúde
42%
Último commit
há 5 meses
Versão
3.0.1
Issues abertas
0
Contribuidores
1

Casos de uso

  • Conduzir investigações estruturadas com mapeamento de entidades e rastreamento de dependências
  • Preservar e gradeuar evidências com padrão único (Admiralty A-F) para publicação defensável
  • Piercear estruturas corporativas e rastrear propriedade beneficiária de shell companies
  • Gerar requisições FOIA estratégicas e verificar claims com auditoria zero-erro
  • Revisar cópia, distribuir em redes sociais e arquivar documentos em banco de dados consultável

Veredito

Ferramenta profissional de nível enterprise para jornalismo investigativo. Oferece workflows completos de OSINT, preservação legal de evidências e defesa contra libelo — essencial para redações que trabalham com investigações complexas. Requer expertise em jornalismo para aproveitar todo o potencial.

README

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

Newsroom

Version License Gemini Cowork

An AI-powered investigative journalism toolkit. 18 skills covering subject dossiers, OSINT, FOIA engineering, corporate veil piercing, evidence preservation, libel defense, and full editorial workflow — from first lead to published story.

Works with Claude (Cowork, Claude Code), Gemini CLI, and any AGENTS.md-aware agent (OpenAI Codex, Cursor — see AGENTS.md).

Installation

Gemini CLI

gemini extensions install https://github.com/ehurrn/newsroom-extension

Cowork / Claude Code

Add ehurrn/newsroom-extension as a marketplace in Cowork settings, or:

claude plugins add ehurrn/newsroom-extension

Commands (Cowork / Claude Code)

Slash commands for common workflows:

CommandDescription
/investigateLaunch an investigation — entity mapping, dependency tracing, OSINT, FOIA, libel-proof documentation
/copy-reviewLine-edit for readability, accessibility, and SEO without touching facts
/editor-reviewFinal editorial review — fact verification, legal risk, sourcing, publication readiness
/publishSafe single-article deployment with pre-flight checklist
/publish-seriesCoordinated deployment of a multi-part investigative series
/distributeGenerate platform-optimized social posts from a published article
/archiveIngest raw OSINT into a queryable database with chain of custody

Skills

Domain knowledge used automatically when relevant. On Gemini CLI, ask the agent to load a skill by name. On Cowork/Claude Code, skills trigger automatically or via slash commands.

Investigative Desk

SkillWhat it does
investigative-journalistScope your investigation, define claims, map what you need to prove
muckraker-master-fileBuild your central intelligence file — structured master-file.json of entities, evidence, claims, gaps
evidence-preservation-protocolMandatory: write-once evidence store, append-only collection log, legal hold — delete nothing
evidence-gradingOne shared standard (Admiralty A–F × 1–6) and publication rule used by every sourcing desk
subject-dossier-constructionPI-grade public records sweep and structured dossier per subject
structural-dependency-mappingUncover hidden connections between people, organizations, and money
corporate-veil-piercingTrace corporate structures, shell companies, and beneficial ownership
zero-error-defensive-auditFact-check every claim with source attribution and confidence scoring
osint-source-inversionAssess what's publicly findable and evaluate publication risk
temporal-anomaly-sequencingSpot timeline inconsistencies and suspicious chronological patterns
precision-foia-engineeringDraft legally rigorous public records requests with strategic sequencing

Newsroom Operations

SkillWhat it does
copy-reviewLine-edit for readability, SEO, accessibility, and ad compliance
data-archivistTurn raw documents into structured, queryable databases
social-distributorGenerate platform-ready social posts with legally defensible copy
final-editor-reviewAdversarial pre-publication review — libel, ethics, editorial sign-off
publish-articleSafe single-article deployment with pre- and post-publish checks
publish-seriesCoordinate multi-part series with sequential editorial gates
managing-editorTrack assignments, deadlines, blockers, and hand off tasks that need a human

How It Works

Skills are markdown files in the skills/ directory. Each desk has a SKILL.md with operational guidelines. The investigative-journalist desk has additional sub-skill files for specialized techniques.

Skills are designed to run in sequence — each one builds on the previous:

Define investigation → Gather intelligence → Map connections
  → Verify claims → Edit and review → Publish → Track what's next

See GEMINI.md for the full sub-agent delegation protocol and desk ordering. AGENTS.md-aware tools (OpenAI Codex, Cursor) get the same operating model via AGENTS.md.

Data Schema & Evidence Discipline

All investigation state lives in a single master-file.json conforming to skills/investigative-journalist/schemas/master-file.schema.json — entities, typed relationships, graded evidence, claims, timeline, gaps, leads, and an append-only collection log. Markdown reports are rendered from it, never maintained separately.

Core invariants enforced across every desk:

  • Grade at intake — every evidence item carries an Admiralty grade (source reliability A–F × information credibility 1–6) and a chain-of-custody block (where obtained, when, how, SHA-256 hash).
  • Preserve everything, delete nothing — collected items go into a write-once evidence/ directory; web sources are archived to a third party (Wayback/archive.today) at collection time. Notes, negative search results, dead-end leads, and drafts are retained under legal hold (on by default).
  • Append-only audit trail — every collect/grade/cite/comment/publish action is a collection_log[] entry with a monotonic sequence number; the log is never edited, only amended.
  • Publication rule — a claim is publishable only when corroborated by two independent sources, or single-sourced to an A1/A2 official record.
  • Right of reply — high defamation-risk claims require a documented comment request before drafting.
  • Legal boundary — public records and lawful FOIA only; no pretexting, account access, impersonation, or non-consensual recording.

What's New in 3.2.0

  • One evidence-grading standard (evidence-grading.md) — the Admiralty scale (source reliability A–F × information credibility 1–6) and a single publication rule, replacing the divergent HIGH/MEDIUM/LOW and PRIMARY/SECONDARY/TERTIARY scales that the audit and final-review desks used separately.
  • Reconciled workspace layout — managing-editor and evidence-preservation-protocol now share one canonical session structure (master-file.json + evidence/ + working/ + drafts/); managing-editor is now the documented legal-hold authority that deletion requests escalate to, with a hash-mismatch tamper check.
  • Jurisdiction-aware libel defense — final-editor-review now flags that U.S. actual-malice protections don't apply abroad (UK/Commonwealth reverse the truth burden; libel tourism) and defaults to the strictest plausible regime.
  • Legal boundary on corporate-veil-piercing — explicit no-pretexting/public-records-only guardrail (parity with subject-dossier-construction), plus schema hooks (ENT-/REL- edges) and A1/A2 grading before naming a beneficial owner.

What's New in 3.1.0

  • Fixed Claude Code installability — marketplace.json now registers one plugin sourced from the repo root (it previously listed skill markdown files as plugins, which could not install); removed the unsupported inline commands array from plugin.json (commands are auto-discovered from commands/); unified name and version across all manifests.
  • New: machine-readable Master File schema (schemas/master-file.schema.json) replacing markdown-only tables.
  • New skill: subject-dossier-construction — PI-grade identity resolution and a 17-row public records sweep per subject.
  • New skill: evidence-preservation-protocol — mandatory spoliation-proof record keeping with desk-handoff checklists.
  • New: AGENTS.md — compatibility with OpenAI Codex, Cursor, and other AGENTS.md-aware agents.

License

Unlicense — free to use, modify, and share. No restrictions.

Important: You are responsible for all legal, ethical, and professional standards in your jurisdiction. Consult legal counsel before publishing sensitive investigations.

Ver o original no GitHub

Bom para

Relacionados