A runbook is a procedure that a coding agent session executes step by step through subagents. The steps are prompt files, the transitions are a few lines of Python, and the run leaves a record you can read.
- skills: the tools.
agent-runbook-authoringwrites and reviews runbooks,runbook-viewershows the progress of a run. - gallery: ready-made runbooks to run as they are or read as examples.
- throng-mcp: an MCP server that runs Claude Code, Codex, OpenCode and other harnesses as subagents of each other, so a runbook step can go to any model.
New here? Install the skills and ask your agent to turn a procedure you repeat into a runbook: agent-runbook-authoring writes it with you. Rather see a run before writing your own? Install runbook-review-loop from the gallery and try it on your next code change: its steps run on your session's own subagents, with nothing else to install.
