journalism-tools

Journalism Tools for Claude Code

A plugin marketplace with tools for investigative journalism: Python script execution with automatic dependency management, data preprocessing with provenance tracking, and transparent data analysis designed to be defensible under scrutiny.

Installation

Add the marketplace and install the plugin:

/plugin marketplace add nhagar/claude-plugins-journalism
/plugin install journalism-tools@journalism-tools

Skills

Once installed, you'll have access to four skills:

/journalism-tools:python-runner

Run Python scripts with automatic dependency management using uv. No manual environment setup required—dependencies are installed automatically in isolated environments.

/journalism-tools:journalistic-data-preprocessing

Preprocessing workflow for journalistic data analysis emphasizing transparency, provenance, and human oversight. Core principles:

• Provenance first: Every row traces to source file, sheet, and row number
• No silent transformations: Every change is documented and approved
• Human-in-the-loop: Present findings and get approval before transformations
• Transparency artifacts: Generate documentation suitable for reporters and editors

/journalism-tools:structured-data-analysis-journalism

Analyze preprocessed data for investigative journalism with full transparency. Emphasizes simple, legible analyses over complex methods—every finding must be explainable to editors and defensible under scrutiny. Core principles:

• Simple beats clever: Analyses must be explainable in plain language
• Every number needs a source: Statistics trace back to verifiable records
• Findings are hypotheses: Analysis surfaces patterns, not proof of wrongdoing
• Defensibility over sophistication: Simple analyses that hold up under scrutiny

/journalism-tools:document-extractor

Extract structured data from documents that resist standard parsing—scanned PDFs, redacted FOIA responses, inconsistent government forms, and OCR artifacts. Follows a five-step workflow:

1. Convert documents to page images
2. Transcribe each page to markdown, preserving layout and marking redactions, illegible text, and handwriting
3. Stitch page transcripts into a single document
4. Schema proposal with journalist approval before extraction
5. Extract to JSON with provenance tracking and a browser-based review interface

Core principles:

• Provenance: Every extracted record traces to source page and document
• Human-in-the-loop: Schema must be approved before extraction proceeds
• Transparent ambiguity: Redactions, illegible text, and low-confidence values are explicitly marked
• Zero-install review: Generates a self-contained HTML interface for journalist verification

Local Development

Test the plugin locally without installing:

claude --plugin-dir ./plugins/journalism-tools

License

MIT