Dispatch
Once you have a structured work tracker (Work Context Protocol) and a structured way to do work with an agent (Agent Pipeline), the missing piece is the worker loop that closes the two together — something that watches WCP for queued items and hands them off to a Claude Code session running with the right context.
Dispatch is that worker. It’s a Rails 8 app: Solid Queue for the polling, a small web UI for managing workers and watching run history, Kamal for deploy. Workers can run in prompt mode (one-shot completion) or skill mode (multi-step pipeline). Each run gets a structured record so the next session can pick up where it left off.
The three together — WCP for memory, Agent Pipeline for shape, Dispatch for the loop — make up the AI-augmentation stack I actually use day to day.