interactive course · 9 lessons · ~1.8h

Learn ruflo by building a Gomoku AI.

ruflo is an agent meta-harness for Claude — swarms, memory, hooks, and more. You'll master it the only way that sticks: by using each capability to build a real five-in-a-row game with a five-level AI opponent.

Start learning →Play the game

the finished game — light wins on row 8

What you'll build

A production-grade Gomoku game: animated stone drops, a glowing win-line, full keyboard play, and an AI you can dial from clueless to near-unbeatable.

15×15 boardWeb Worker AIVCF/VCT searchreduced-motion

9lessons, each shipping a real feature

5AI difficulty levels you build by hand

Real ruflo, documented

Every lesson reflects what actually happened when we drove the build with ruflo — memory, SPARC, swarms, hooks — including when not to use it.

Four chapters

1Getting oriented
2Building the game
3Teaching the machine
4Automating the work

The curriculum

Each lesson pairs one ruflo capability with one piece of the game.

Lesson 008 min

The mental model & setup

What ruflo actually is, how its three surfaces fit together, and why we scaffolded the app by hand instead of with a swarm.

▹ init · doctor · Agent vs MCP vs CLI✦ Project scaffolded, board renders

Lesson 0110 min

Skills: the shortcuts

Invoke a ruflo skill to scaffold the interactive board. Learn the highest-value skills and when a skill beats a from-scratch prompt.

▹ Skills — what they are & which matter✦ Click-to-place stones + turns

Lesson 0212 min

Agents: delegating work

Spawn a named coder agent to implement and verify five-in-a-row detection. Learn agent types and how ruflo routes a task to the right one.

▹ Agent types, routing, the Agent tool✦ Win detection → 2-player game

Lesson 0311 min

Memory & learning

Search memory before a task and store what worked after. See the vector recall that lets ruflo remember decisions across sessions.

▹ store · search · namespaces · vectors✦ History, undo, last-move highlight

Lesson 0414 min

SPARC: design before code

Run the SPARC methodology to design the board-evaluation function before writing it, then build the first two AI levels on top of it.

▹ Specification → Pseudocode → … → Completion✦ AI L1 (random) + L2 (heuristic)

Lesson 0516 min

Swarms: coordinated teams

Coordinate an architect → coder → tester → reviewer team to build the search engine, and move it into a Web Worker so the UI never freezes.

▹ topology · SendMessage · pipeline vs fan-out✦ AI L3 minimax + L4 alpha-beta + Worker

Lesson 0612 min

Hooks & background workers

The automation layer: hooks that fire on your tool calls and background workers that sweep the codebase while you build.

▹ pre/post hooks · audit/optimize/testgaps · daemon✦ Tests, animations, difficulty selector

Lesson 0715 min

Hive-mind, MCP & judgment

Reach a hive-mind consensus on the capstone AI, discover MCP tools on demand, and learn — honestly — when the harness helps and when it's overhead.

▹ consensus · ToolSearch · when (not) to use ruflo✦ AI L5 threat-space search, final polish

Lesson 0812 min

GitHub automation

Put ruflo's GitHub agents to work: an automated PR, a code-review swarm, and a release — then deploy the finished game.

▹ PR manager · code-review swarm · releases✦ Ship it: PR, review, deploy to Vercel