News Pipeline

Overview

The News pipeline is a multi-stage content enrichment workflow. Articles enter as raw tweets/links and are progressively enriched by agents through 6 stages.

NEW → REVIEWED → PROCESSED → REFINED → CONCLUDED → ARCHIVED
 ↑       ↑          ↑          ↑           ↑          ↑
Intake  Mason      Mack       Alex    Turf Monster   Alex

Stages & Agent Assignments

Services

All services live in app/services/news/. They reopen the News class (not a module — class News, not module News).

News::Intake

Fetches the latest Adam Schefter tweet not already in the DB.

news = News::Intake.new.call
# => News record or nil

• Uses X API v2 (GET /2/users/:id/tweets)
• Resolves AdamSchefter → user ID (cached per instance)
• Deduplicates by x_post_id
• Extracts first URL from tweet text
• Requires X_BEARER_TOKEN env var

News::Review (Mason)

News::Review.new(news).call(
  primary_person: "Patrick Mahomes",
  primary_team: "Kansas City Chiefs",
  primary_action: "extended",
  secondary_person: "Dak Prescott",
  secondary_team: "Dallas Cowboys",
  article_image_url: "https://example.com/image.jpg"
)

News::Process (Mack) — fully automated

News::Process.new(news).call
# No input needed — derives slugs from existing person/team fields
# "Patrick Mahomes" → "patrick-mahomes"

Also available via the UI: Process (auto) button on the show page when stage is reviewed.

News::Refine (Alex)

News::Refine.new(news).call(
  title_short: "Mahomes extends with Chiefs",
  summary: "Patrick Mahomes agrees to a record extension.",
  feeling: "excited",
  feeling_emoji: "🔥",
  what_happened: "Chiefs locked up their franchise QB."
)

News::Conclude (Turf Monster)

News::Conclude.new(news).call(
  opinion: "Smart move by KC — stability at QB is everything.",
  callback: "Watch for contract details and cap implications."
)

Rake Task

bin/rails news:intake    # Fetch latest Schefter tweet

Credentials Setup

X (Twitter) API — Bearer Token

1. 1Password: Stored in agent.turf.x in the agents vault
   bash
op item get zz3uigmkrwjlnnksst33butc4e --vault txqp6ijdo3ujsfhsfzdj5h5dzq --field "Bearer Token" --reveal


2. Local .env: Add to /Users/alex/projects/.env (symlinked into all apps):
   
X_BEARER_TOKEN=<bearer-token-from-1password>


3. Heroku (production): Set via config var:
   bash
heroku config:set X_BEARER_TOKEN=<token> --app mcritchie-studio


X Developer Portal

• Login: alex@turfmonster.com (credentials in 1Password agent.turf.x)
• Portal: https://developer.x.com/en/portal/dashboard
• Tier: Free (read-only, 1 app, 1500 tweets/month read)
• Rate limits: 1 request per 15 minutes for user tweets endpoint

Onboarding Checklist

When starting fresh (new machine, new DB, or new agents suite):

1. Database

bin/rails db:migrate
bin/rails db:seed        # Creates 7 sample news articles across all stages

2. X API Credentials

# Fetch bearer token from 1Password
export OP_SERVICE_ACCOUNT_TOKEN='<token-from-zshrc>'
X_TOKEN=$(op item get zz3uigmkrwjlnnksst33butc4e \
  --vault txqp6ijdo3ujsfhsfzdj5h5dzq \
  --field "Bearer Token" --reveal)

# Add to shared .env
echo "X_BEARER_TOKEN=$X_TOKEN" >> /Users/alex/projects/.env

Or manually: copy the Bearer Token from 1Password and paste into .env.

3. Verify Intake

bin/rails news:intake
# Expected: "Created: <tweet text> (news-xxxxxxxxxxxx)"

4. Walk the Pipeline (console)

n = News.last

# Step 1: Review (Mason)
News::Review.new(n).call(
  primary_person: "...", primary_team: "...", primary_action: "...",
  secondary_person: nil, secondary_team: nil, article_image_url: nil
)

# Step 2: Process (Mack) — automated
News::Process.new(n).call

# Step 3: Refine (Alex)
News::Refine.new(n).call(
  title_short: "...", summary: "...",
  feeling: "...", feeling_emoji: "...", what_happened: "..."
)

# Step 4: Conclude (Turf Monster)
News::Conclude.new(n).call(
  opinion: "...", callback: "..."
)

5. Verify UI

• Kanban board: http://localhost:3000/news
• Article detail: click any card
• Workflow docs: http://localhost:3000/news/workflow

Architecture Notes

• Free movement: Articles can move backward/forward between any stages (unlike Tasks which enforce transitions)
• Stage timestamps: Each stage records when it was reached; moving backward preserves prior timestamps
• Position system: Cards are ordered within each Kanban column; position resets on stage change
• No model associations: News stores person/team names and slugs as strings — no FK relationships to Player/Team models (yet)
• Service namespace: Services use class News (reopening the AR model class), not module News, to avoid TypeError collision