Skip to main content

Agent Workflow — How Work Actually Flows

This is the real operational workflow. Every task, feature, and fix flows through this system. One human, one AI brain, two AI hands, connected by GitHub.

Step-by-Step Walkthrough​

1. Ryan — The Owner​

Ryan is the only human in the system. Everything starts and ends with him. He sets direction, makes final decisions, and approves what ships. He works from his phone or MacBook — often away from the computer — so everything must be viewable remotely via the wiki or GitHub.

What he does:

  • Gives Claude an idea, a problem, or a direction ("build X", "fix Y", "look into Z")
  • Reviews work via GitHub PRs, wiki pages, and Telegram updates
  • Approves or rejects before anything ships

What he doesn't do:

  • Write code directly
  • Manage individual tasks — that's Claude's job

2. Claude — The CEO​

Claude is the brain of the operation. Running as Claude Code CLI on Ryan's MacBook, Claude takes Ryan's intent and turns it into structured, actionable work. Claude doesn't write product code — Claude decides what needs to happen and delegates it to the right bot via GitHub.

What he does:

  • Listens to Ryan, asks clarifying questions, understands the goal
  • Breaks work into tasks and creates GitHub Issues with clear acceptance criteria
  • Assigns issues to Max (product) or Hank (engineering) based on the work type
  • Uses BMAD agents for planning and architecture when needed
  • Uses the Pulp Fiction Crew workflow for focused engineering tasks
  • Maintains persistent memory across sessions so context is never lost
  • Pushes documentation to the wiki — single source of truth

What he doesn't do:

  • Run in production — Claude is strategy/orchestration only
  • Touch the product runtime — OpenAI powers the product, not Anthropic
  • Execute tasks that Max or Hank should handle

3. Max Power — Product / Design Lead​

Max is an OpenClaw bot running on a dedicated iMac mini. He handles product and design work — visual QA, UI implementation, screenshots, and product-level issue triage. He picks up issues from GitHub and delivers PRs with artifacts attached.

What he does:

  • Picks up product/design issues from GitHub
  • Creates PRs with screenshots, design artifacts, and visual proof
  • Runs mobile-first QA and visual lock checks
  • Triages product issues and ships incrementally

What he doesn't do:

  • Engineering or infrastructure work — that's Hank
  • Make strategic decisions — that's Claude
  • Talk to Hank directly — they communicate through GitHub only

Best input for Max: Objective + deadline + must-haves + approval workflow

4. Hank Scorpio — Engineering Lead​

Hank is an OpenClaw bot running on localhost, powered by OpenAI. He's the engineering muscle — code, pipelines, infrastructure, deploys. When Claude creates an engineering issue on GitHub, Hank picks it up and delivers working code as a PR.

What he does:

  • Picks up engineering issues from GitHub
  • Writes code, builds pipelines, fixes infrastructure
  • Creates PRs with clear commit history and audit trail
  • Handles deploys, observability, and pipeline throughput
  • Prototypes fast and ships to production

What he doesn't do:

  • Product or design work — that's Max
  • Make strategic decisions — that's Claude
  • Talk to Max directly — they communicate through GitHub only

Best input for Hank: Concrete outcome + hard constraints + authority boundaries + decision owner

5. GitHub — The Glue​

GitHub is the central nervous system. It's the only place where Claude, Max, and Hank intersect. There are no side channels — no direct bot-to-bot communication. Everything is tracked, reviewable, and auditable.

How it works:

  • Issues = task assignment. Claude creates them, Max and Hank pick them up.
  • PRs = delivered work. Max and Hank submit PRs, Claude or Ryan reviews.
  • Actions = CI/CD automation. Tests, builds, and checks run automatically.
  • Projects = tracking across 4 boards: Product, Engineering, UI-UX, DevOps.

6. Vercel — Delivery​

Vercel is the last mile. When code merges to main on GitHub, Vercel automatically builds and deploys. The wiki and the app both deploy this way. Push to publish.

What it deploys:

  • The Bitcoinology Wiki (Docusaurus) — the single source of truth
  • The Bitcoinology App — the product itself

Key Rules​

  1. GitHub is the glue — Claude creates Issues, Max and Hank pick them up, deliver PRs. No side channels.
  2. Claude never touches product runtime — strategy and orchestration only. OpenAI powers the product via Hank.
  3. Max and Hank are independent — separate machines, no direct connection. Both talk to GitHub.
  4. Push to publish — if it's not in the wiki, it doesn't exist.
  5. One brain, two hands — Claude decides; Max and Hank execute. They don't make strategic decisions.