Talking to Claude

One command starts everything. The rest happens automatically.

#The main command

/buidl "your idea here"

That's how you kick off a full build. Claude takes your idea, breaks it into a spec, spins up the right agents, builds the contracts and frontend, audits the code, and opens a PR — all without you doing anything else.

Example:

/buidl "a simple token faucet where anyone can claim 100 tokens once per day"

Claude will:

1. Clarify the idea (challenges you to think through edge cases)
2. Write a spec
3. Build the smart contract
4. Build the frontend
5. Run the auditor
6. Deploy to testnet
7. Test everything
8. Mark it done

You watch it happen. You don't prompt each step.

#The eight agents

When you run /buidl, you're not just talking to one Claude. You're activating a pipeline of 8 specialized agents. Each one has deep knowledge of their domain.

opnet-contract-dev goes first. It writes the AssemblyScript smart contract. It knows OPNet's token standards, storage patterns, SafeMath requirements, and all the things that make contracts work on Bitcoin L1 vs. other chains.

opnet-frontend-dev and opnet-backend-dev go next, in parallel. The frontend dev builds a React UI with proper wallet connection and transaction flows. The backend dev handles any API services or indexer integrations.

opnet-auditor reviews everything. It checks your code against 27 known bug patterns — 9 of them critical — pulled from real issues found in production OPNet codebases. It catches things a generic code reviewer would miss.

opnet-deployer handles getting the contract onto the network. It knows the deployment flow, handles the Bitcoin transaction construction, and confirms the contract is live.

opnet-e2e-tester and opnet-ui-tester verify the whole thing actually works. End-to-end tests for the contract, interface tests for the frontend.

loop-reviewer is the final check. It looks at the complete build and either signs off or sends things back for fixes.

#Autonomous mode vs. interactive mode

Autonomous mode (claudeyproj) — the pipeline runs without stopping to ask you things. This is how /buidl is meant to be used. Agents hand off to each other. The whole thing completes while you do something else.

Interactive mode (claudey) — Claude pauses and asks for approval before running commands. Use this when you're learning how the pipeline works, or when you want to review each decision as it happens.

For building apps, use claudeyproj. The agents are well-tested. The pipeline is reliable. Micromanaging each step just slows you down.

#Other commands

/buidl "idea" — full pipeline from scratch

/buidl-spec "idea" — just write the spec, don't build yet. Useful when you want to review the plan before committing to a full build.

/buidl-review 42 — review a specific PR number

/buidl-status — see where the pipeline is right now. Did it finish? Is it mid-build? Did something stall?

/buidl-resume — if the pipeline got interrupted (network issue, you closed the terminal, hit a timeout), this picks up where it left off instead of starting over.

#When you want to have a conversation

Not everything needs the full /buidl pipeline. Sometimes you just want to ask a question or make a small change.

Just talk to Claude normally:

How does the NativeSwap AMM calculate price impact?

Add a loading spinner to the claim button

The connect wallet button is too small on mobile, fix it

Claude has the OPNet knowledge loaded via the plugin. You don't need to explain what OPNet is or how Bitcoin transactions work. Just describe what you want.

#A note on iteration

The /buidl pipeline produces a working first version, not a finished product. After the pipeline completes, you'll typically want to:

• Adjust the frontend design (Chapter 6 covers this)
• Test edge cases with your own wallet
• Add features the initial spec didn't include
• Polish things before shipping

That's normal. Use direct prompts for iteration after the initial build. Save /buidl for starting new projects or major rebuilds.