Every subagent run on our own codebase since June 2026, measured monthly in production: runs, agent hours, tokens read and written. Open data, CSV export.
Our actual configuration: the three rule files, the hooks that block, the 5 agents, the skills, session continuity and memory. Not a tutorial, the config we run.
How a plain-language client request becomes a ticket, a verified code analysis, then a pull request, across three agents with strictly separated permissions.
An exhaustive functional specification for internal tools: every process, every screen field, every formula and the data model, defined before a line of code.