🛳 BG Shipping Log #35

Hey builders! 100 days into 2026 and a clear theme is showing up across our tools: we’re building for a world where developers and agents work together. That shift is already changing how people learn Ethereum, build dapps, and use Scaffold-ETH 2.

That's why this issue is more packed than usual: ETHSkills, agent-ready challenges, a tutor mode, a grumpy reviewer, and an agent we built from scratch. Full thesis here.

⛵️ Let’s dive into what we shipped so far:

AI agents are stale on Ethereum.

They overestimate gas, hallucinate contract addresses, miss shipped standards like EIP-7702 and ERC-8004, confuse L2 tradeoffs, mishandle token decimals, and sometimes commit private keys.

EthSkills gives agents current Ethereum context in the format they can actually use. No install required: just give your agent the URL.

🌐 20+ skills, one URL away. Gas, L2s, wallets, standards, security, frontend UX, audits, Noir, indexing, verified contract addresses, and production deployment.

📡 Works wherever your agent works. Add it to Claude, Cursor, Codex, install it as a Claude Code plugin, or grab it via ClawHub.

📝 Built from real agent failures. We test agents without EthSkills, find what they get wrong, and turn those failures into reusable skills.

We are making Speedrun Ethereum easier to use with an agent in the loop, not just for devs, but for builders who have a product idea and want help building it.

🤖 AI-ready challenges. Every challenge now ships with an AGENTS.md file, so your assistant starts with the project context, commands, gotchas, and conceptual background, instead of guessing. (example)

🎓 AI Tutor Mode. Type /start in your IDE/CLI and an agent turns the challenge into a guided learning flow: it asks questions, checks your understanding, gives hints, and reviews your code as you go. Works in Claude, Cursor, and other CLI/IDEs that support skills.

🛠️ Curated Build Prompts. A collection of ready-to-use AI prompts with project specs so builders can go from idea to working Scaffold-ETH app faster.
Skip the blank canvas paralysis. Grab a prompt, tweak it with your needs, feed it to your agent, and get a working prototype in minutes instead of days. 🔗 launch tweet

🔨 Foundry support is live. Foundry versions of the top 5 challenges (tokenization, crowdfunding, token-vendor, dice game, DEX) shipped, with a Hardhat/Foundry toggle on the site so builders can pick their stack per challenge.

Scaffold-ETH 2 picked up the same agent-first treatment.

📄 Agent context. Added an AGENTS.md so any agent harness (Claude Code, Cursor, Windsurf, etc.) starts with the right project context. That means less prompt babysitting and a better first pass when you ask your assistant to work in a SE2 codebase.

🧩 AI Skills replace extensions. We migrated Extensions over to AI Skills files: focused context the agent loads before implementing functionality. The AI now does the heavy lifting of integration, giving you much more flexibility than our legacy system and letting you seamlessly stack multiple extensions at any time
Coverage now includes x402, SIWE, OpenZeppelin contracts, EIP-5792, drizzle-neon, subgraph, and more.

📚 AI-friendly docs. docs.scaffoldeth.io/build-with-ai ships with AI metadata baked in (AGENTS.md, skills list, orchestrator skill).
So if you build with SE2 using an agent, the docs are now much more likely to produce useful output on the first try

🧙 Grumpy Carlos Reviewer. A code-review subagent with a personality scraped from BuidlGuidl PR history. Strict, thorough, occasionally grumpy, exactly the way Carlos would do it 😂.
Human review is still critical, but this is a great first pass.

🧶 Yarn v4. Updated to the latest Yarn, now featuring npmMinimalAgeGate to keep your supply chain safe from newly deployed malicious packages.

Alongside the product work, we’ve been building from first principles so we understand the layer underneath the tools.

🔧 raked: a minimal TypeScript agent (sessions, memory, tools, skills) with the core under 100 lines. Built to be read, forked, and tweaked.
Walkthrough: Building an AI agent from scratch.

🔍 Custom RAG. We implemented retrieval-augmented generation from scratch on Arbitrum DAO governance data, mainly to understand how retrieval and evaluation actually work under the hood. Writeup: RAG from the ground up.

Live Workshops aren’t a core focus for BuidlGuidl this year, but a few of us showed up at events around the world to onboard new builders:

• Austin at ETH Denver

• Shiv at ETH Kuala Lumpur

• Shiv at ETH Mumbai

• Damu at Crecimiento AR

• Carlos at Malaga University

If you want to support our work, we’re participating in the @thedaofund x @Giveth Ethereum Security QF round.

It’s quadratic funding, so every dollar gets amplified by the matching pool.

👉 Donate to BuidlGuidl on Giveth

That’s it for the first 100 days of 2026.

If you want to try any of this, all the code, skills, and writeups are open.
Go build something.

🛠️ See you next quarter, builders!