agent-ops
A small Python library with the three things every agent that spends money needs.
Who has this problem
Every agent that runs in public needs the same three things, and they usually get built after something goes wrong: a spend cap that refuses when it cannot read its own counter, a record that answers "why did it decide that" with a rule code instead of generated prose, and a queue that does not lose work across a redeploy.
My role
Python package design, fail-closed spend controls, decision records and durable queue.
How to try it
- Open the repository to review the package API and installation workflow.
Decisions I made
- The spend counter fails closed when storage is unavailable. Decision records keep rule codes with the operation that produced them, and queued work is stored durably so it can resume after a process restart.
- The package is installable independently and can be composed with an application’s existing agent and persistence layers.
Limits that remain
- It covers spend caps, decision records and a durable queue. It is not a framework for building agents.