splitty

Pipes-and-filters orchestration for chunk-level analysis with Claude Code sub-agents. Splits a corpus into chunks, runs a deterministic DAG of map / reduce / gate / terminal filters in parallel, and unions the results. The orchestration loop runs in the main Claude conversation — no daemon, no MCP server — driven by an on-disk state machine that makes runs resumable and lost-message tolerant. Ships three example pipelines (map-reduce, classify-and-extract, fan-out/fan-in) and a /splitty-design skill that authors a pipeline YAML from a natural-language request.