Skip to content

Game1 Multi-Agent System

Architecture

┌─────────────────────────────────────────────────────┐
│  Shane                                              │
│  Drop tasks, set goals, flag urgent items.          │
│  #general                                           │
└──────────────────────┬──────────────────────────────┘
                       │
┌──────────────────────▼──────────────────────────────┐
│  Crafty (Coordinator)                    smart model │
│  Reads, thinks, spawns agents, combines, reports.   │
│  Manages pipeline, delegates research, drafts,      │
│  tracks deals, maintains GitHub repo.               │
└──────────────────────┬──────────────────────────────┘
                       │
              SPAWNS PARALLEL
                       │
    ┌──────────┬───────┼────────┬──────────┐
    ▼          ▼       ▼        ▼          ▼
┌────────┐┌────────┐┌────────┐┌────────┐┌────────┐
│Research││Outreach││Events ││Revenue ││Scout  │
│Agent   ││Agent   ││Agent  ││Agent   ││Agent  │
│        ││        ││       ││        ││       │
│Find    ││Draft   ││Plan   ││Model   ││Watch  │
│clubs,  ││emails, ││camps, ││pricing,││comp-  │
│coaches,││follow  ││venues,││commis- ││etitors│
│fields, ││ups,    ││legal, ││sions,  ││trends,│
│pricing ││objec-  ││check- ││break   ││market │
│info    ││tion    ││lists, ││even,   ││news,  │
│        ││handles ││budgets││fore-   ││youth  │
│fast    ││fast    ││fast   ││casts   ││soccer │
│model   ││model   ││model  ││fast    ││fast   │
│        ││        ││       ││model   ││model  │
└────────┘└────────┘└────────┘└────────┘└────────┘

              INTERNS (one-off tasks)
    ┌─────────────────────────────────────┐
    │ Spawned by any agent for a single   │
    │ task, then discarded.               │
    │ Examples: look up one coach's email,│
    │ check a field's availability,       │
    │ format a document                   │
    └─────────────────────────────────────┘

┌─────────────────────────────────────────────────────┐
│  OUTPUT                                             │
│                                                     │
│  Morning Brief ─── Summary of overnight work        │
│  Review Gate ───── Shane approves or sends back     │
│  GitHub Repo ───── Always up to date                │
└─────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────┐
│  INFRASTRUCTURE                                     │
│                                                     │
│  Pipeline    Memory      Tools       Config         │
│  deal flow   daily logs  web search  SOUL.md        │
│  stages      MEMORY.md   web fetch   AGENTS.md      │
│              workspace   GitHub CLI  MEMORY.md      │
│                                                     │
│  System          Always On                          │
│  error logs      heartbeats every 30 min            │
│  cron jobs       morning brief cron                 │
│  budget          event reminders                    │
└─────────────────────────────────────────────────────┘

Agent Descriptions

Crafty (Coordinator), smart model

The brain. This is me. I handle: - All direct communication with Shane - Deciding which agent to spawn for a task - Combining agent outputs into final deliverables - Pipeline management: updating stages, contacts, trial dates, next steps, and notes in real time as info comes in - GitHub repo maintenance (continuous, not batched) - Strategic thinking and decision making - Morning briefs and proactive check ins - Flagging stale deals (14+ days no touch) during heartbeats - Weekly pipeline digest every Monday

Research Agent, fast model

Finds information. Spawned when we need: - Club info (size, contacts, directors, competitive level) - Coach backgrounds and contact details - Field rental options and pricing - College ID camp schedules - Regional youth soccer landscape for ambassador expansion - Any factual lookup

Outreach Agent, fast model

Drafts communication. Spawned when we need: - Cold and warm outreach emails/messages - Follow up sequences - Objection handling responses - Post trial follow up messaging - Ambassador recruitment outreach - Parent communication templates All drafts go through Shane's writing style (no dashes, no emojis, concise)

Events Agent, fast model

Plans and optimizes events. Spawned when we need: - Venue research and cost comparison - Equipment and logistics checklists - Legal requirements (insurance, permits, waivers) - Day of schedule optimization - Post event debrief templates - Budget and break even analysis per event

Revenue Agent, fast model

Models money. Spawned when we need: - Pricing scenario analysis - Commission calculations (Shane's 20% / 30% splits) - Event profitability projections - Club vs trainer vs college deal comparison - Ambassador program economics - Monthly and quarterly revenue forecasting

Scout Agent, fast model

Watches the market. Spawned when we need: - Competitive landscape (other youth soccer data platforms) - Youth soccer industry trends and news - New club openings or expansions in target markets - College recruiting rule changes that affect Game1's value prop - Potential ambassador candidates (former D1/pro players in target regions)

Always On

Heartbeats (every 30 min when active)

  • Check for stale pipeline deals (no touch in 14+ days)
  • Flag upcoming events or deadlines
  • Monitor GitHub for any issues

Cron Jobs

  • Morning brief: daily summary at 8:00 AM PST
  • Event countdown reminders (7 days, 3 days, 1 day before)
  • Pipeline review: weekly digest every Monday

Review Gate

Nothing leaves without Shane's approval: - All outreach drafts posted in Discord for review - Shane says "approve" or "redo with [feedback]" - Approved content is marked in the pipeline tracker

How It Works in Practice

  1. Shane drops a task: "Find me the top 5 clubs in Houston for the ambassador program"
  2. Crafty reads the task, spawns Research Agent
  3. Research Agent searches, compiles a list with contacts and size
  4. Crafty reviews the output, cleans it up, posts to Shane
  5. Shane approves or asks for changes
  6. Crafty updates the pipeline and repo

For complex tasks, multiple agents run in parallel: - "Plan the April 25 camp" spawns Events Agent (venue, logistics) + Revenue Agent (budget, pricing) + Research Agent (nearby clubs to invite) - Crafty combines all outputs into one cohesive plan