nwsrm

AI-powered content pipeline. 6 agents. Zero infrastructure. Feed it an idea, get publication-ready content.

Idea → Scout → Researcher → Writer → Adapter + Marketer → Editor → Done

Install

One command. Pick your method:

npx nwsrm                    # npm (recommended)

curl -sSL https://raw.githubusercontent.com/MatthewDC/nwsrm/main/install.sh | bash

git clone https://github.com/MatthewDC/nwsrm.git

Then open a Claude Code, OpenClaw, or any AI session in the directory. The AI reads CLAUDE.md, detects a fresh install, and walks you through setup.

How It Works

nwsrm is a config-driven, multi-agent content production system. You define your brand's voice, audience, and quality standards through a conversational setup. Six specialized agents handle the rest.

Agents define the PROCESS. Config defines the PERSONALITY. Same agents, different config = different publication.

The 6 Agents

| # | Agent | ID | Job | |---|-------|----|-----| | 1 | Scout | nwsrm.scout | Monitors sources, finds trending topics, writes briefs | | 2 | Researcher | nwsrm.research | Deep-dives topics, verifies claims, gathers sources | | 3 | Writer | nwsrm.write | Drafts content following voice guide and content playbook | | 4 | Editor | nwsrm.edit | Quality gate — AI-ism detection, voice compliance, fact check | | 5 | Adapter | nwsrm.adapt | Adapts content into platform-specific formats | | 6 | Marketer | nwsrm.market | Drafts engagement content and distribution strategy |

Content Types

Configure which types to produce during setup. The Writer drafts your primary type; the Adapter creates derivatives.

| Type | ID | Length | |------|----|--------| | Newsletter Article | article | 500-700 words | | Blog Post | blog | 800-1500 words | | X/Twitter Thread | thread | 8-12 tweets | | X/Twitter Post | tweet | 280 chars | | LinkedIn Post | linkedin | 800-1200 chars | | YouTube Shorts Script | short | 60 seconds | | Long-Form Video Script | video | 10-15 minutes | | Podcast Outline | podcast | Episode structure | | Podcast Script | podcast-script | Full spoken-word | | Quick Take | quick | 200-300 words |

Supported Platforms

| Platform | How Agents Run | |----------|----------------| | Claude Code | Each agent is a scheduled task with cron | | OpenClaw | Each agent is a skill; manager orchestrates | | Any AI platform | SKILL files work as prompts on any agent framework |

Setup Flow

After install, the AI drives an interactive onboarding conversation:

1. Identity — Publication name, niche, author background
2. Audience — Target reader, what they want from your content
3. Voice — Tone style (4 presets), patterns to avoid
4. Content Types — Which formats to produce, primary type
5. Operations — Publishing cadence, approval mode, sources to monitor
6. Config Generation — All config files created from your answers
7. First Run — Give it a topic or let the Scout find one

No manual config editing required. The AI generates everything from the conversation.

Pipeline Stages

topics/ → approved/ → research/ → drafts/ → reviewed/ → ready/ → published/
                                                            ↓
                                                  media/ + marketing/

| Directory | Contents | Written By | Read By | |-----------|----------|------------|---------| | topics/ | Topic briefs | Scout | You (Mode A) or auto | | approved/ | Approved topics | You or auto | Researcher | | research/ | Research packets | Researcher | Writer | | drafts/ | Content drafts | Writer | Editor | | reviewed/ | Editor-passed content | Editor | You (Mode A) or auto | | ready/ | Approved for publish | You or auto | Adapter, Marketer | | published/ | Published with URLs | You | Archive | | media/ | Derivative content | Adapter | You | | marketing/ | Engagement drafts | Marketer | You |

Approval Modes

Mode A (default) — You approve topics and publishing. Everything else is automatic.

Mode B (full auto) — Agents run without human gates. You still publish manually.

The Researcher (Optional)

For news-reaction niches, skip it:

agents:
  researcher:
    enabled: false
  writer:
    reads_from: "approved"

For research-heavy niches (product reviews, finance, science), the Researcher is the difference between surface-level and deeply sourced content.

Example Configs

The examples/ directory includes complete configurations:

• news-analysis/ — Analysis publication covering trending topics
• fitness-coaching/ — Health and fitness content
• saas-product/ — B2B SaaS thought leadership

Copy any example to get started: cp -r examples/news-analysis/config/* config/

Documentation

• Setup Guide — Step-by-step walkthrough
• Config Guide — How to fill out each config file
• Agent Guide — How agents work, how to customize
• Architecture — Full technical design
• Handoff — Session continuity for AI assistants

License

MIT