Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -30,3 +30,6 @@ _testmain.go
*.prof

.DS_Store

AGENTS.md
CLAUDE.md
163 changes: 163 additions & 0 deletions AGENTS.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,163 @@
# AGENTS.md

Project instructions for **AI coding assistants and agents** (OpenAI Codex, Cursor, GitHub Copilot, Claude Code, and similar tools). This file is the canonical source of shared project instructions; `CLAUDE.md` imports this file for Claude Code.

## Project overview

Bee is the reference Go implementation of an Ethereum Swarm node. It implements decentralized storage and communication: content-addressed chunk storage, Kademlia-based routing, postage stamp accounting, push/pull syncing, PSS messaging, feeds, and storage incentives (redistribution game).

**Module**: `github.com/ethersphere/bee/v2`

**Go version**: 1.26 (see `go.mod`)

**License**: BSD 3-clause (see `LICENSE`)

Human-oriented contributing docs: `CONTRIBUTING.md`, `CODING.md`, `CODINGSTYLE.md`, `README.md`.

## Guidelines

Keep changes **minimal and focused**. Only touch code that belongs to the task. Do not refactor unrelated code, rename symbols for style only, or mix unrelated fixes in one commit or PR.

Read **`CONTRIBUTING.md`**, **`CODING.md`**, and **`CODINGSTYLE.md`** for process, patterns, and style. Prefer matching existing naming, types, imports, and log style in the files you edit.

Do **not** add, remove, or update `go.mod` dependencies unless the task **explicitly** requires it or the person asking for the work **explicitly** requests a dependency change.

Handle errors and logging the way this repo does: propagate errors with context (`fmt.Errorf("…: %w", err)`), avoid logging and returning the same error, and use structured logging with clear operator vs developer levels (see `CODING.md`).

Prefer **`package foo_test`** tests, **`export_test.go`** when you must export internals, and **`t.Parallel()`** only where it is safe. Add or update tests when behavior changes. Integration tests use **`-tags=integration`**.

## Pre-commit checklist

Before you finish a change set (especially before a commit or PR), run these and fix failures:
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this is just my preference, but i honestly am very averse to claude touching my git history. in most of my projects there's a big fat line saying - never commit anything to git. NEVER!!!. it makes things very difficult to follow since the agents add a lot of intermediate commits - i like to have my own messages and history that i as an operator can relate to.

that being said, i don't have much resistance to having this committed as it is, but i would strongly suggest to add both of these files to the .gitignore file, so that devs can be free to make their own changes locally, and we can just keep the "spine" of the document in the repo, with little changes as needed

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes good idea I agree


1. **Formatting** — `make format` (gofumpt + gci; see `CODING.md`).
2. **Compile** — `make build` (all packages) and, when you need the binary artifact, `make binary` (`dist/bee`, `CGO_ENABLED=0`).
3. **Tests** — `make test` (unit tests, `-failfast`). For a single package use `go test ./pkg/<name>/...`. Use `make test-race` when concurrency is central to the change. Use `make test-integration` only when you touch integration-tagged code.
4. **Static checks** — `make lint` and `make vet` (see `.golangci.yml`).

CI pipelines may use `make test-ci` / `make test-ci-race` (see `Makefile` for flags).

## Dev commands (quick reference)

```bash
make binary # dist/bee
make build # compile all packages
make test # unit tests
make test-race # unit tests + race detector
make lint # golangci-lint (see .golangci.yml)
make vet # go vet
make protobuf # regenerate *.pb.go after changing .proto files
```

## Architecture

### Entry point and CLI

Binary built from `cmd/bee/main.go`. CLI uses Cobra + Viper:

- `bee start` — full or light node (`cmd/bee/cmd/start.go`)
- `bee init` — initialize data directory
- `bee deploy` — deploy smart contracts
- `bee db` — database management
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we go even deeper with db repair and db nuke?

Copy link
Copy Markdown
Contributor Author

@akrem-chabchoub akrem-chabchoub Apr 16, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, it can be added, but the frequency of use is unclear. If it is not needed too often, it remains practical. When required, the agent can execute bee db to retrieve the information.

AFAIK, these files are meant to serve as a map for agents rather than a fully detailed manual.

image

- `bee version` — print version info

Configuration: option constants in `cmd/bee/cmd/cmd.go`. Viper reads CLI flags, environment variables (`BEE_` prefix), and YAML config.

### Node bootstrap

`pkg/node/node.go` is the main orchestrator. `NewBee()` wires subsystems via dependency injection; avoid global mutable state. The `Bee` struct holds service references and provides `Shutdown()` for teardown.

### HTTP API

- Router: `gorilla/mux` in `pkg/api/router.go`
- Route groups in `Mount()`:
- `mountTechnicalDebug()` — `/node`, `/addresses`, `/health`, `/readiness`, `/metrics`, `/loggers`, pprof
- `mountBusinessDebug()` — topology, accounting, settlements, stamps management
- `mountAPI()` — `/bytes`, `/chunks`, `/bzz`, `/feeds`, `/soc`, `/stamps`, `/tags`, `/pins`, `/pss`, `/grantee`
- `checkRouteAvailability` can block endpoints during sync
- OpenAPI: `openapi/Swarm.yaml` (API versioning follows SemVer there; the main Bee release version does not)
- Endpoints exist at root (e.g. `/bytes`) and under `/v1/` (e.g. `/v1/bytes`)

### P2P networking

- Transport: libp2p (`pkg/p2p/libp2p/`)
- Wire formats: protobuf (gogo) — each protocol area has a `pb/` directory with `.proto` and `doc.go` (`go:generate` calling `protoc` + `--gogofaster_out`)
- Important protocol packages: `pushsync`, `pullsync`, `retrieval`, `pingpong`, `hive`, `pricing`

### Storage

- Chunk types: CAC (`pkg/cac/`), SOC (`pkg/soc/`)
- Interfaces: `pkg/storage/` (`Putter`, `Getter`, `Hasser`, `Deleter`)
- Local store: `pkg/storer/` (reserve, cache, upload, pinning)
- Blob engine: `pkg/sharky/`
- BMT: `pkg/bmt/`
- State: `pkg/statestore/` (LevelDB); `pkg/shed/` (typed LevelDB layer)

### Postage and incentives

- `pkg/postage/` — batches, stamps, services
- `pkg/postage/listener/` — on-chain events
- `pkg/postage/postagecontract/` — contract interaction
- Stamps: batch ID, depth (capacity), amount (per-chunk value)
- `pkg/storageincentives/` — redistribution / storage incentive game

## Key domain concepts

- **Address** — 32-byte hash (`pkg/swarm/`). Chunk and overlay addresses; proximity is XOR-based (more shared prefix bits = closer), not lexicographic ordering.
- **Chunk** — 4096 bytes of data (`ChunkSize = SectionSize * Branches = 32 * 128`), plus 8-byte span (`SpanSize`); `ChunkWithSpanSize = 4104`.
- **CAC** — content-addressed chunk; address from BMT root of data.
- **SOC** — single owner chunk; address from owner + id, with signature.
- **PO** — proximity order (shared prefix bits). `MaxPO = 31`, `ExtendedPO = 36`.
- **Neighborhood** — prefix / responsibility region for storage and sync.
- **Kademlia** — routing table over XOR distance (`pkg/topology/`).
- **Postage stamp** — payment signal attached to chunks.
- **Push sync / pull sync** — push new data toward neighborhood; pull historical sync between peers.
- **Redistribution** — incentive game proving reserve storage.

## Coding conventions (summary)

### Copyright (goheader)

Every `.go` file starts with:

```go
// Copyright <year> The Swarm Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
```

### Errors, logging, concurrency

- Propagate errors; do not log and return the same error. Use `fmt.Errorf("context: %w", err)`. Avoid stacking "failed to" prefixes.
- Sentinel errors: `var ErrFoo = errors.New("package: description")` — identity only, compared with `errors.Is`.
- Typed errors: a struct implementing `error` with exported fields, inspected with `errors.As` when callers need data about the failure.
- Logging: separate operator-facing (`Error`/`Warning`) from developer detail (`Debug`, V-levels). Keys: `lower_snake_case`, specific names. Runtime log tuning: `/loggers` API.
- Every goroutine needs a clear shutdown path. Channels: prefer unbuffered or size 1 unless strongly justified; an owning goroutine sends or closes.

### Testing

- Prefer external test packages: `package foo_test` not `package foo`.
- `export_test.go` in the real package to export symbols only for tests.
- Use `t.Parallel()` where safe. Avoid the word `fail` in test names. Integration: `-tags=integration`. Prefer `t.Fatal` / `t.FailNow` over `panic` in tests.

### Style and tooling

- American English (e.g. marshaling, canceled).
- Avoid `init()` where possible (`gochecknoinits`).
- Enums often start at `iota + 1` when zero should mean "unset".
- Use `time.Time` / `time.Duration`, not raw ints for time.
- `var _ Interface = (*Impl)(nil)` where useful.
- Dependency injection over mutable globals. Exit only from `main()`.

### Commits

Never commit or push to git.

## Common pitfalls

- Do not confuse `ChunkSize` (4096 data bytes) with `ChunkWithSpanSize` (4104 including span).
- XOR distance: XOR between two addresses produces smaller integers as more prefix bits are shared, do not confuse this with lexicographic ordering of addresses.
- Goroutines must be stoppable (context cancel, quit channel, etc.).
- Full node vs light node: reserve and storage incentives are full-node concerns.
- Postage batches can be unusable (expired, depleted, unsynced); check before relying on stamps.
g
1 change: 1 addition & 0 deletions CLAUDE.md
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
@AGENTS.md
Loading