# AGENTS.md
This repository keeps its agent/contributor guidance in [`CLAUDE.md`](./CLAUDE.md), which is the
single source of truth for build/test/lint commands, the architecture overview, and the
Akka-mirroring and performance conventions.
**If you are an AI coding agent (or a human contributor) working in this repo, read
[`CLAUDE.md`](./CLAUDE.md) first.** It covers everything you need: how to build, test, lint, and
benchmark; the module layout (`src/stream.rs`, `src/graph.rs`, `src/actor.rs`, `src/io/`); the
blueprint-vs-materialization rule and execution model; and the rules that govern API and
performance changes.
As of v0.4.0, the codebase includes graph cycles (`MergePreferred`/`Broadcast` feedback loops),
`StreamRefs` (`StreamRefs::source_ref`/`sink_ref`), IO adapters
(`StreamConverters::as_input_stream`/`as_output_stream`), and a fully widened typed graph executor
that covers all major junction shapes including the cyclic feedback kernel.
To avoid drift, all guidance lives in `CLAUDE.md` — update that file rather than duplicating
content here.