Umpire

Teach your agent to program like you

Umpire's review UI: a list of commits on the left, the selected commit's message in the middle, and the diff on the right with a comment being written against a line.

A developer tool providing a local, GitHub PR review-like interface for giving feedback to your coding agent. Feedback is aggregated over time and used to suggest feed-forward improvements to AGENTS.md.

GitHub ↗

Streams

Agent orchestration based on stateful, iterative looping

Streams' terminal UI: a list of plan and coding iterations down the left, and a snapshot of the selected iteration on the right with the agent's report, the issues it closed and opened, and a diffstat.

Streams implements agentic loops as progressive state machines with configurable stages (ex. research, planning, coding, review). Within each stage, agents take turns implementing and reviewing. They pass context to one another using Beads.

GitHub ↗

Signet

A framework for automated contract testing in CI/CD

Service A talks to a test double through the Signet Proxy while Signet's API test exercises Service B; both report results to the Signet Broker, which holds the OpenAPI specification.
Written in Go, Node.js, and React. Deploys to AWS.

Signet tests an HTTP API and the services that consume it independently. When deployed in CI/CD, it identifies incompatibilities between services before they reach prod.

Read the case study ↗