Meet the Agents 🦞

News Workflow

News items flow through 6 stages, each owned by a specific agent. Articles are progressively enriched with metadata, summaries, and opinions as they move through the pipeline.

New Reviewed Processed Refined Concluded Archived
New

Manual Intake

No agent

Raw news enters the system. Articles are added manually or via API with basic source information.

Fields Populated

title url x_post_id x_post_url author published_at
Reviewed

Mason

Infrastructure Specialist

Reads the article, identifies key people and teams involved, and labels the primary action. Extracts the article image if available.

Fields Populated

primary_person primary_team primary_action secondary_person secondary_team article_image_url
Processed

Mack

General Worker

Links the people and teams identified in the review stage to their matching database records using slug lookups. This connects the news to the rest of the system.

Fields Populated

primary_person_slug primary_team_slug secondary_person_slug secondary_team_slug
Refined

Alex

Lead Orchestrator

Summarizes the article, writes a concise short title, and infers the emotional tone and significance. Polishes the raw primary_action into a cleaner what_happened label.

Fields Populated

title_short summary feeling feeling_emoji what_happened
Concluded

Turf Monster

Sports Domain Specialist

Forms an opinion on the article's impact, especially as it relates to sports props, picks, and content strategy. Creates a callback with a content idea or follow-up action.

Fields Populated

opinion callback
Archived

Alex

Lead Orchestrator

Auto-archived after 48 hours, or manually archived when no longer relevant. No new fields are populated. Articles remain searchable.

Fields Populated

None (terminal stage)

How It Works

  • Free movement — Articles can be dragged to any stage on the Kanban board. Agents are expected to follow the pipeline logically, but there are no enforced restrictions.
  • Stage timestamps — Each stage transition records a timestamp (reviewed_at, processed_at, etc.). Moving backwards does not clear previous timestamps.
  • Progressive enrichment — Each stage adds new metadata. Later stages build on earlier data (e.g. primary_action from Review becomes the basis for what_happened in Refine).
  • Inspired by Boodle Scraper — This pipeline adapts the news workflow from Boodle Scraper for the McRitchie Studio agent system.