agent-lab

A multi-agent research crew that reports its bill and its failures - and loses to one agent on cost. Agent Lab Vol 1 #04.

Screenshot to code, where the product is the loop: a comparison that can actually converge, best-so-far rather than last, and a generator that never sees the answer. Agent Lab Vol 2 #01.

AI agent for safe testnet transactions: inspects a wallet, checks allowances, dry-runs a tx, and explains risk in plain English before anything is signed — Sepolia only, safety by design.

An autonomous coding agent whose interesting part is how it stops: it cannot edit its own tests, it detects when it is looping, and it names which of five reasons ended the run. Agent Lab Vol 1 #05.