Journalism Toolkit for Claude Code

The most comprehensive journalism AI system available as a Claude Code plugin. 30 specialized agents, 2 commands, and 1 analytical skill covering investigative reporting, fact-checking, disinformation analysis, AI content detection, foreign news de-biasing, bot & troll detection, multimedia production, and content distribution.

One command. Five journalism modes. From tip to publication.

---

Table of Contents

• Why This Plugin?
• Domains Covered
• Installation
• Quick Start Commands
• How It Works
• What’s Inside
• Architecture
• Repository Structure
• Use Cases
• Contributing
• FAQ
• Legal Notice & Disclaimer
• License & Author

---

Why This Plugin?

Modern journalism demands speed without sacrificing accuracy — verifying viral claims, detecting deepfakes, tracing disinformation networks, and producing multimedia content across platforms. Manual workflows can’t keep up.

Four Key Components:

• 30 specialist agents across investigative, forensic, multimedia, and distribution domains
• 19 verification agents organized in 5 tiers for systematic disinformation analysis
• 2 slash commands for full journalism workflows and dedicated disinformation investigations
• Master orchestrator routing queries automatically to the right specialist

Target Users: Investigative journalists, fact-checkers, newsroom editors, media analysts, and disinformation researchers.

---

Domains Covered

---

Installation

Requirements: Claude Code v1.0.33 or later

# Clone the repository
git clone https://github.com/swarochish/journalism-toolkit.git

# Load the plugin
claude --plugin-dir ./journalism-toolkit

Alternative: Add to .claude/settings.json:

{
  "plugins": ["path/to/journalism-toolkit"]
}

---

Quick Start Commands

Master Command:

/journalism-toolkit:ultimate-multimedia-journalist

Story Development:

/journalism-toolkit:ultimate-multimedia-journalist

Plan and develop stories with investigative reporting, fact-checking, editing, and multimedia.

Verification & Fact-Checking:

/journalism-toolkit:ultimate-multimedia-journalist verify [topic or URL]

Forensic analysis of viral content with interactive dashboards and risk assessment.

Investigative Reporting:

/journalism-toolkit:ultimate-multimedia-journalist investigate [topic]

Multi-week investigations with document analysis, source development, and legal review.

Multimedia Production:

/journalism-toolkit:ultimate-multimedia-journalist create [content type]

Video production, data visualization, and cross-platform content with Python automation.

Real-Time Monitoring:

/journalism-toolkit:ultimate-multimedia-journalist monitor [topic]

Ongoing narrative monitoring with automated alerts.

Disinformation Analysis:

/journalism-toolkit:disinformation-brain

Dedicated interface coordinating 19 forensic sub-agents for comprehensive disinformation detection.

Investigation Reports:

Use the journalism-toolkit:investigation-report-generator skill to produce self-contained HTML dossiers with a dark editorial theme, interactive navigation, timelines, and evidence cards.

---

How It Works

User Query
    |
journalism-master-orchestrator
    |-> Classifies query by journalism mode
    |-> Routes to specialist agent (1 of 30)
    |   |-> Agent applies domain-specific analysis
    |   |-> Collaborates with other agents as needed
    |   |-> Returns structured output
    |
Structured Response with:
  - Verified facts & source grades
  - Evidence assessment
  - Multimedia assets
  - Distribution-ready content
  - Actionable next steps

Multi-Agent Architecture: Each journalism domain has dedicated specialist agents. The master orchestrator handles routing; agents collaborate on cross-domain tasks (e.g., an investigation requiring fact-checking, forensic analysis, and multimedia production).

Tiered Verification Pipeline: Disinformation investigations use a 5-tier forensic pipeline — from initial narrative analysis through deep financial/technical investigation to real-time monitoring.

---

What’s Inside

Summary

Key Agents

Key Skills

---

Architecture

journalism-master-orchestrator
|-- Core Agents
|   |-- investigative-reporter
|   |-- fact-checker
|   |-- interview-specialist
|   |-- ethics-advisor
|   |-- story-editor
|-- Verification Agents (Tier 1-5)
|   |-- Tier 1: narrative-timeline-analyst, source-ecosystem-mapper, attribution-tracking-specialist
|   |-- Tier 2: deepfake-forensics-specialist, ai-content-detector, bot-network-detector, grassroots-sincerity-analyst, psychological-manipulation-detector, narrative-cluster-analyst, cross-platform-tracker
|   |-- Tier 3: bias-assessment-analyst, foreign-news-analyst, financial-trail-investigator, author-history-profiler, agency-bias-profiler, technical-infrastructure-analyst
|   |-- Tier 4: impact-measurement-analyst, visualization-dashboard-generator
|   |-- Tier 5: real-time-monitoring-coordinator
|-- Multimedia Agents
|   |-- video-journalist
|   |-- data-visualization-specialist
|-- Distribution Agents
    |-- multi-platform-distributor
    |-- seo-optimization-specialist

disinformation-investigation-orchestrator
|-- Coordinates same verification agents for dedicated disinformation workflows

---

Repository Structure

journalism-toolkit/
|-- .claude-plugin/plugin.json    Plugin manifest
|-- LICENSE                       MIT License
|-- README.md                     This file
|-- agents/
|   |-- journalism/               29 specialist agents
|   |-- disinformation/           1 orchestrator agent
|-- commands/                     2 slash commands
|-- skills/
    |-- investigation-report-generator/
        |-- SKILL.md
        |-- assets/               HTML report template
        |-- references/           Component library

---

Use Cases

Investigative Journalists

• Plan multi-month investigations with document analysis, source development, and legal review
• Trace financial trails, dark money, and ownership networks
• Generate evidence packages with interactive dashboards

Fact-Checkers

• Verify viral claims using 5-tier forensic pipeline
• Detect deepfakes, AI-generated text, and manipulated media
• Assess bot amplification and coordinated inauthentic behavior

Newsroom Editors

• Evaluate source credibility and institutional bias
• Identify AI-generated or click-farm content before publication
• Cross-check foreign-language sources with PR-sanitization detection

Disinformation Researchers

• Map influence networks and narrative evolution across platforms
• Separate genuine grassroots voices from troll armies and paid shills
• Analyze state media propaganda patterns (RT, CGTN, IRNA, KCNA)

Media Analysts

• Quantify reach, engagement, and real-world impact of campaigns
• Track narrative clusters and echo chamber dynamics
• Monitor evolving campaigns with automated alerts

---

Contributing

Contributions welcome! Here’s how to get involved:

Adding a New Agent

Use the short-form pattern (YAML frontmatter + structured sections):

---
name: your-agent-name
description: One-line description
---

# Your Agent Name

**Purpose**: What it does
**Outputs**: What it produces
**When to Use**: When to deploy it (Tier level)
**Key Tools**: Tools and APIs used

## Core Functions

- Function 1
- Function 2

PR Workflow

1. Fork the repository
2. Create a feature branch (git checkout -b add-agent-name)
3. Add or update agent files following existing naming convention (domain-function-role.md)
4. Update orchestrators and commands to reference new agents
5. Update README agent counts and architecture tree
6. Submit a PR with a clear description of what the agent does and which issue it addresses

Reporting Issues

• Include the journalism domain and specific use case
• Describe what capability is missing or what isn’t working
• Reference relevant existing agents if applicable

---

FAQ

Q: Do I need API keys to use this plugin? A: The plugin works as a Claude Code agent system. Some agents reference external APIs (GPTZero, DeepL, Sensity AI, etc.) for tool suggestions, but the agents themselves provide analysis frameworks and workflows without requiring API keys.

Q: How do I choose between /ultimate-multimedia-journalist and /disinformation-brain? A: Use /ultimate-multimedia-journalist for general journalism workflows (story development, verification, investigation, multimedia production, monitoring). Use /disinformation-brain for dedicated disinformation investigations where you want the full forensic pipeline.

Q: Can I use individual agents without the orchestrator? A: Yes. Each agent is a standalone markdown file that can be referenced directly. The orchestrators coordinate multiple agents for complex workflows, but individual agents work independently for focused tasks.

Q: How does the bot-to-grassroots pipeline work? A: The bot-network-detector first classifies accounts as bot or human. Human accounts that show suspicious patterns are then passed to grassroots-sincerity-analyst for sincerity scoring (0-100), troll-army detection, and information value assessment.

---

Legal Notice & Disclaimer

⚠️ READ BEFORE USE. Three documents govern this repository: LICENSE (MIT), DISCLAIMER.md (14-clause legal disclaimer), and NOTICE (hazard notice). The summary below is not a substitute for those documents. By cloning, forking, installing, or using this software you accept all three in full.

Nature of the software. This repository is an educational and analytical framework of prompt templates, agent definitions, and orchestration logic for use inside Claude Code. It is not a news publication, journalism product, fact-checking service, forensic laboratory, or regulated media offering. The Author is not a “journalist,” “news publisher,” “intermediary,” or “Data Fiduciary” by reason of distributing this software.

AI / LLM hallucination warning. All content referencing identifiable persons, entities, financial links, bot networks, sincerity scores, deepfake verdicts, or translations is generated at runtime by a large language model under the User’s sole direction. Such output may be fabricated, hallucinated, or materially false — including confident-sounding but fictitious allegations, non-existent sources, invented quotations, imaginary financial relationships, and wrongly-classified media. Example: the tool may generate a “financial link” between two named persons that does not exist in reality. Confident language, numerical precision, and source-grade labels (e.g., “Tier 1 verified”) in any output are stylistic and carry no probative value. Every identifying factual claim must be independently corroborated against at least two contemporaneous, independent, human-sourced, non-LLM references before any use.

User is sole publisher. The User is the sole commissioner, originator, and publisher of any output. The User bears full legal, editorial, ethical, reputational, and financial responsibility and indemnifies the Author, contributors, and maintainers against all claims, FIRs, notices, DPDP Board proceedings, damages, penalties, and legal costs arising from the User’s use (DISCLAIMER.md clause 11).

Prohibited uses (non-exhaustive — see DISCLAIMER.md clause 5 for the full enumerated list and residual catch-all):

In addition to the enumerated list, the User must comply with every applicable law, rule, regulation, court order, and platform term of service of India and of any jurisdiction from which the User operates or in which any output is published or consumed. Omission of any specific prohibition from the list above is not consent or licence.

No professional advice. Nothing produced by this software is legal advice, editorial judgement, forensic-expert opinion, or financial advice. No attorney–client, journalist–source, or fiduciary relationship is created. Consult a qualified human professional before acting.

Governing law and exclusive forum. This software, its use, and this notice are governed by the laws of the Republic of India. The competent courts at Hyderabad, Telangana have exclusive jurisdiction over all civil, contractual, tortious, and regulatory proceedings to the extent forum is electable. Criminal jurisdiction is fixed by statute (BNSS 2023) and is not subject to forum selection; the User acknowledges that any offence alleged to have been committed through this software is committed by the User, not by the Author.

Limitation of liability. Liability of the Author, contributors, and maintainers is excluded to the maximum extent permitted by law; where the law does not permit exclusion, aggregate liability is capped at INR 100.00. See DISCLAIMER.md clause 12 for the full terms and the MIT License for the underlying “AS IS / NO WARRANTY” provisions.

If you are a journalist, fact-checker, researcher, or editor: you are not the target of these provisions — they exist to keep the repository safe to open-source against motivated attack and LLM misuse. Use the software professionally, verify independently, obtain legal review on sensitive matters, and publish with care.

---

License & Author

License: MIT — Swarochish Chekuri

Author: Swarochish Chekuri

Repository: https://github.com/swarochish/journalism-toolkit