back

toolforge

a self-extending agent system. when it hits a task it has no tool for, it writes one, tests it in a sandbox, and files it away so the next run can just pick it up. 2nd place at the snowflake hackathon.

Toolforge pipeline: a task passes through a planner, a codegen agent, and a sandbox, then a validator files the tool in a vector registry that the planner reads from on later runs. task planner agent codegen agent sandbox run validator + fuzz vector registry snowflake, 1536-d reuse on later runs
one pass through the loop. tools only leave the sandbox if they survive the validator.

the problem

an agent is only as good as the tools you hand it. give it a fixed toolbox and it fails the moment a task falls outside that box. the usual fix is a human noticing the gap and writing another tool, which puts a person in the loop of every new capability.

how it works

what it means

the toolbox grows on its own. a task that took four steps and a fresh tool on monday takes one registry lookup on tuesday, and the cost of that capability is paid once instead of on every run.