Pasé las últimas semanas certificándome en el stack completo de
Anthropic: Claude, Claude Code, Agent Skills y Subagents.
Porque la próxima ola del dev workflow no es 1 dev + 1 IA.
Es 1 dev + N agentes coordinados.
Y en Web3, esto cambia todo. Esta semana publico cómo lo aplico
The biggest Starknet hack ($9.5M zkLend) wasn't a Cairo language bug.
It was rounding + accumulator manipulation — impossible to catch with static tools.
My multi-agent swarm found 31 similar patterns in 5 min.
PAT-001 docs coming this summer 🛠️
https://t.co/odE06776wM
@base The "agents will use crypto" thesis is right, but the gating factor isn't the crypto — it's predictable settlement and identity for AI actors. Account abstraction + gasless flows are the actual primitive that unlocks agent-to-agent payments at scale.
@EliBenSasson This is exactly why the "DAO without gas" pattern matters. Replacing centralized governance with one that still requires capital to participate is not replacement — it's gatekeeping in new clothes. Engineering should remove friction, not preserve it.
@smpalladino Same workflow shift here. I run a multi-agent swarm (Claude Code + RuFlo) for Cairo audits — Codex now handles the "side errands" while the main agent works. The bottleneck moved from compute to orchestration design.
@rauchg The "any GitHub repo as a registry" pattern is huge. npm-level distribution without the dependency hell. Curious if this extends naturally to language tools Solidity components, Cairo modules, even Anthropic Agent Skills. Would unlock real composability across Web3 dev tooling
@gakonst The parallel with security testing is sharp — same primitive, different objective ("find the bug" vs "find the alpha").
Building this for Cairo audits (multi-agent swarm). For MEV the spec IS the money, while audits keep it implicit.
🗳️ Just shipped: DaoVotacionGasless
A DAO where voting doesn't cost gas — solving the UX problem that kills decentralized governance.
Module 7/19 of my Master's at CodeCrypto 🎓
How it works 🧵👇
Governance design:
- 1 person = 1 vote (deposit amount doesn't matter)
- Proposals require ≥10% of real treasury (anti-spam)
- "Real Treasury" model: governance weight doesn't decay on execute, but quorum is calculated on actual balance
First multi-agent swarm on a real project 🐝
Mission: map Cairo smart contract vulns on Starknet.
5 min → 31 patterns, prioritized.
Surprise: the $9.5M zkLend hack wasn't a Cairo quirk — it was rounding + accumulator manipulation.
PAT-001 docs first 🛠️
https://t.co/odE0677Emk
Imagine an AI Queen orchestrating a swarm of architects, devs and auditors — all coordinated, all working in parallel, all living inside your terminal. 🐝🤯
Meet RuFlo V3, the open-source multi-agent framework transforming my workflow in Claude Code. 🧵
...and I grab a coffee while the swarm runs the terminal. ☕💻
We're entering a stage where dev work isn't 1 dev + 1 AI — it's 1 dev + N coordinated agents.
The devs who learn to orchestrate this today will have a brutal edge tomorrow.
Already using MCP servers daily? 👇
@gakonst This is the multi-agent infra direction that makes sense to me — the Kubernetes layer addresses real production concerns that CLI-based swarms hand-wave away. Any plans for MCP server support so teams can plug in their own harnesses (e.g., Foundry test runners)?
@ericng39 Tried Claude Code + MCP + subagents a couple months ago and never went back — the orchestration is the real unlock. Running parallel agents on Solidity & Cairo projects (audit / tests / review). Any plans to explore multi-agent setups for OSS Cairo at OZ?
@PatrickAlphaC This is the kind of attack that scares me most as a Web3 dev — your code isn't even the vulnerability, your dependencies are. I now audit package.json & Cargo.toml with the same rigor as the main contract. Any specific Cyfrin tooling for this, or still manual + Socket?