Merge remote-tracking branch 'origin/main' into slice-1-webrtc-loopback
Some checks failed
CI / fmt (pull_request) Failing after 35s
CI / clippy (pull_request) Failing after 41s
CI / test (1.85) (pull_request) Failing after 36s
CI / test (stable) (pull_request) Failing after 39s
CI / deny (pull_request) Failing after 44s

# Conflicts:
#	README.md
This commit is contained in:
opencode controller
2026-06-28 13:19:16 -04:00
4 changed files with 545 additions and 32 deletions

215
CONTRIBUTING.md Normal file
View File

@@ -0,0 +1,215 @@
# Contributing
Thanks for considering a contribution to Rutster. This file is the short
form — read [`AGENTS.md`](AGENTS.md) for the full orientation any agent
(human, AI, hybrid) working in the repo should have.
---
## Trunk-based development
1. Branch from `main` for any change.
2. Open a PR targeting `main`.
3. CI gates (below) must pass before merge.
4. Squash-merge to keep `main` linear.
**Never push directly to `main`.** Branch protection is planned to
enforce this; until then, self-discipline.
### Branch naming
Single-purpose branches named after the change:
- `slice-1-webrtc-loopback` — long-running build-target branch
- `docs/quickstart` — documentation additions
- `media/opus-roundtrip-fix` — bug fix with crate scope
- `ci/deny-toml` — CI config
No strict scheme — short, descriptive, hyphen-separated.
---
## CI gates
All four must pass before merge to `main`:
```bash
cargo fmt --check # formatting
cargo clippy -- -D warnings # lints (warnings = failures)
cargo test --all # all unit + integration tests
cargo deny check # licenses, advisories, bans, sources
```
CI runs on push + PR to `main`, matrix: latest stable + the MSRV pinned
in `rust-toolchain.toml`. The `clippy` + `test` jobs install `libopus-dev`
(the one system dependency, per PORT_PLAN §7).
If your change adds a new dependency, run `cargo deny check` locally
before pushing — a license conflict or duplicate-version ban will fail
CI, and it's faster to catch locally.
---
## Commit messages
- **Imperative mood**: "Add X" / "Fix Y", not "Added X" / "Fixes Y".
- **Subject ≤ 72 chars**.
- **Body wraps at 72**, blank line between subject and body.
- **Reference ADRs / specs** by number when relevant: `ADR-0002`,
`slice-1 spec §3.4`.
- Match the style of recent commits:
```bash
git log --oneline -10
```
Example shape (from the repo history):
```
media: PcmFrame + AudioSource/Sink + Opus codec pair
PcmFrame is the canonical tap format (16-bit mono @ 24 kHz, 480 samples
per 20 ms frame — ARCHITECTURE.md). AudioSource/AudioSink are the seam
step 2 splices the tap client into (spec §3.3); EchoAudioPipe is the
slice-1 wiring of that seam. OpusDecoder/OpusEncoder wrap the opus
crate's libopus FFI with hot-path match-and-continue.
```
### Atomic commits
One logical change per commit. Doc ratifications, code, and tests each
land as separate commits when practical. Don't bundle unrelated work.
A new feature + its tests can land in one commit if the test is part of
the feature's correctness story (TDD). A refactor + a new feature should
be two commits.
### Never commit secrets
`.gitignore` covers `.env*`, `*.pem`, `*.key`. If a new secret pattern
appears in your work, extend `.gitignore` in the same commit.
---
## Code style
See [`AGENTS.md`](AGENTS.md) "Code style" for the full guide. Highlights:
- `cargo fmt` is the source of truth for formatting. Don't hand-format.
- `clippy -D warnings` is the lint bar. Fix the code, don't suppress
with `#[allow]` unless the rationale is documented inline.
- `snake_case` (functions, variables, modules, crates), `PascalCase`
(types), `UPPER_SNAKE_CASE` (constants).
- Newtype wrappers over primitives for type-safety — e.g. `ChannelId(Uuid)`,
not bare `Uuid`.
- Hot path (the 20 ms media loop): **never** `?`-propagate.
Match-and-continue. Dropped packet must not terminate the peer.
- Cold path: `thiserror`-derived error enums + `?`, converted to HTTP
status codes at the axum boundary.
- Never `unwrap()` / `expect()` outside tests or const-init contexts.
### Documentation comments (learner-facing — important)
**This project overrides the default "no comments" convention.** The
user is learning Rust from this codebase. Code in slice 1 (and the
spearhead steps) carries thorough educational comments:
- `//!` module docs on every `lib.rs` / `main.rs` / sub-module
- `///` item docs on every public struct / enum / fn / trait
- `//` inline comments on the *mechanism*, not the what — why this
ownership pattern, why `Arc<Mutex<>>` vs `Arc<RwLock<>>`, why an
`enum` over a struct with a `kind` field, etc. Aim: a Rust learner
reads the comment and learns a specific Rust concept they wouldn't
have inferred from the code alone.
`LEARNING.md` at the repo root indexes "to learn concept X, read file Y"
pointers. Add to it when you introduce a new pattern worth surfacing.
This verbosity is a deliberate trade-off: more tokens now, compound
educational value later. Once a pattern is established, later slices can
be sparser on the well-trodden patterns.
---
## Slice-1 boundaries — what NOT to add (yet)
If an agent (you, an AI pair, a contributor) proposes adding any of these
in slice 1, the right answer is "no, see slice-1 spec §1.2":
- Dedicated timing thread for the media loop (step 4)
- TLS on the HTTP signaling surface (step 5)
- Authn / authz / multi-tenancy (step 6)
- Trickle ICE (when NATs demand it)
- The tap itself (step 2 — slice 1 only *pre-paves* the seam)
- The brain / STT / LLM / TTS (step 3)
- Barge-in / VAD-driven playout kill (step 4)
- PSTN trunk / SIP client (step 5)
- Spend cap / abuse gate (step 6)
- Browser-based automated e2e tests / Selenium / Playwright (post-slice-1)
- Docker / compose (later rung)
- Event bus / Valkey / CDR emission (step 5)
- Transfer / park / pickup / barge features (escalation rung 2)
The spearhead depends on this sequencing. Adding things early breaks
the ordering each step is its own proof.
---
## Terminology policy
**Avoid authoritarian / exclusionary terms** in code, prose, identifiers,
and endpoint names. Use equally-descriptive alternatives:
| Avoid | Use instead |
|---|---|
| police / policing (the verb) | enforce / gate / guard |
| master / slave | primary / replica, leader / follower, controller / worker |
| blacklist / whitelist | denylist / allowlist, blocklist / safelist |
| officer | operator / handler / controller |
| censor | suppress / filter |
**Exception: protocol-convention names are kept verbatim** when they come
from upstream specs or libraries — replacing them would hurt the
educational mapping to upstream docs. **ICE** (Interactive Connectivity
Establishment, RFC 8445) stays: it's the protocol name in `str0m::ice`,
`RTCIceCandidate`, and the cargo ecosystem.
See [`AGENTS.md`](AGENTS.md) "Terminology policy" for the full table.
---
## Pull request workflow
1. Branch from `main`.
2. Make atomic commits per the guidance above.
3. Push the branch + open a PR targeting `main`.
4. CI runs: fmt, clippy, test, deny. All four must be green.
5. Reviewer checks the diff against the spec / ADRs cited in the commit
message. New dependencies require scrutiny — `cargo deny check`
enforces license + source bans, but reviewers should also sanity-check
the dependency choice against the architecture.
6. Squash-merge once approved + green.
7. Delete the branch post-merge (keeps the branch list tidy).
### Reviewing
When reviewing a PR:
- Does it cite the relevant ADR / spec section? (For substantial changes.)
- Does it add anything from the slice-1 "what NOT to add" list? (Reject
if so — refer to slice-1 spec §1.2.)
- Are educational comments present where a new pattern is introduced?
- Any `unwrap()` / `expect()` outside tests? (Reject unless justified.)
- Does the hot-path code use `?`? (Reject — it must match-and-continue.)
- Does the diff bundle unrelated work? (Ask for split commits.)
---
## License
By contributing, you agree your contributions are licensed under
**GPL-3.0-or-later** (ADR-0004). Strong copyleft in the Asterisk lineage.
If you contribute code with a different license header / SPDX expression
in a Cargo manifest, CI will reject it (`cargo deny check licenses`).
Don't introduce dependencies whose licenses conflict — check `deny.toml`
for the allow-list.

View File

@@ -8,47 +8,35 @@ or its architecture.
> technical builder uses to stand up a contact center* — and re-aims it at a category AI is > technical builder uses to stand up a contact center* — and re-aims it at a category AI is
> actively disrupting, instead of a PBX category UCaaS already ate. > actively disrupting, instead of a PBX category UCaaS already ate.
## Slice 1 dev loop (WebRTC media loopback) ## Quickstart
> Build prerequisite: install libopus (the `opus` crate links it via FFI):
> ```bash
> sudo apt-get install -y libopus-dev # Debian/Ubuntu
> # Fedora: sudo dnf install -y opus-devel
> # macOS: brew install opus
> ```
> This is the one system dependency in slice 1. Opus is FFI per PORT_PLAN
> §7's "🦀 Core (FFI)" disposition — the codec surface Rust doesn't need
> to re-implement.
Run the server:
```bash ```bash
# Prereqs: Rust (rustup), libopus dev headers (libopus-dev / opus-devel / brew install opus)
cargo run cargo run
# listening on http://0.0.0.0:8080 # listening on http://0.0.0.0:8080
``` ```
Open a browser to `http://localhost:8080/`, click "Start call", grant Open <http://localhost:8080/> → click "Start call" grant mic → hear yourself echo.
microphone permission. Speak — you should hear yourself back within Full walkthrough + troubleshooting: **[`docs/QUICKSTART.md`](docs/QUICKSTART.md)**.
~200 ms (no perceptible delay). Click "Hang up" to tear down; server
logs `Closing → Closed`.
Verbose tracing: > **Status:** Slice 1 (WebRTC media loopback) is the active build target. The workspace is
> landing task-by-task on the `slice-1-webrtc-loopback` branch. Design:
> [`docs/superpowers/specs/2026-06-28-slice-1-webrtc-loopback-design.md`](docs/superpowers/specs/2026-06-28-slice-1-webrtc-loopback-design.md).
> Implementation plan:
> [`docs/superpowers/plans/2026-06-28-slice-1-webrtc-loopback.md`](docs/superpowers/plans/2026-06-28-slice-1-webrtc-loopback.md).
```bash ## Documentation
RUST_LOG=rutster=debug cargo run
```
### Slice 1 "done" checklist (spec §6.5) | Doc | For when you want to… |
|---|---|
On a clean checkout: | [`docs/QUICKSTART.md`](docs/QUICKSTART.md) | Run it in 5 minutes |
1. `cargo test --all` passes. | [`docs/DEVELOPMENT.md`](docs/DEVELOPMENT.md) | Iterate on the codebase (workspace layout, per-crate testing, dev loop) |
2. `cargo fmt --check` passes. | [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) | Understand the fused per-call vertical + composable platform + agent tap |
3. `cargo clippy -- -D warnings` passes. | [`docs/PORT_PLAN.md`](docs/PORT_PLAN.md) | See every Asterisk subsystem mapped to a disposition (capability checklist, not template) |
4. `cargo deny check` passes. | [`docs/adr/`](docs/adr/) | Load-bearing architecture decisions |
5. `cargo run` + browser manual e2e: speak → hear echo within ~200 ms. | [`AGENTS.md`](AGENTS.md) | Project orientation for any agent (human/AI/hybrid) working in the repo |
6. Hang-up button triggers `Closing → Closed` in server logs. | [`CONTRIBUTING.md`](CONTRIBUTING.md) | Trunk-based dev workflow, CI gates, commit style, review checklist |
7. Every stub crate compiles; its doc-comment names its scheduled step. | [`LEARNING.md`](LEARNING.md) | Index of "to learn concept X, read file Y" (learner-facing codebase) |
8. `LEARNING.md` indexes at least 5 "to learn X, read Y" pointers.
## Why it exists ## Why it exists

195
docs/DEVELOPMENT.md Normal file
View File

@@ -0,0 +1,195 @@
# Development
The dev loop: how to iterate on the Rust workspace, run tests, and find
your way around the codebase.
> **Status:** Slice 1 is the active build target. The workspace is
> landing task-by-task on the `slice-1-webrtc-loopback` branch. Once it
> merges to `main`, everything below applies directly. Until then,
> `git checkout slice-1-webrtc-loopback` to follow along.
> Implementation plan: [`docs/superpowers/plans/2026-06-28-slice-1-webrtc-loopback.md`](superpowers/plans/2026-06-28-slice-1-webrtc-loopback.md).
---
## Workspace layout
Cargo workspace at the repo root, shaped to ADR-0002's fused per-call
vertical. One binary crate + five library crates.
```
rutster/
├── Cargo.toml # [workspace] + [workspace.dependencies]
├── deny.toml # cargo-deny config (licenses/bans/sources)
├── rust-toolchain.toml # pinned stable
├── LEARNING.md # index of "to learn concept X, read file Y"
├── crates/
│ ├── rutster/ # binary: axum signaling server + media driver
│ │ ├── src/main.rs
│ │ ├── src/session_map.rs # DashMap<ChannelId, RtcSession>
│ │ ├── src/routes.rs # HTTP routes
│ │ └── static/index.html # browser test client
│ ├── rutster-media/ # str0m WebRTC + Opus⇄PCM codec boundary
│ │ ├── src/pcm.rs # PcmFrame, AudioSource/Sink traits, EchoAudioPipe
│ │ ├── src/opus_codec.rs # OpusDecoder / OpusEncoder wrappers
│ │ ├── src/rtc_session.rs # RtcSession (per-peer owner)
│ │ └── src/loop_driver.rs # str0m poll loop
│ ├── rutster-call-model/ # the Channel/Leg object embryo
│ ├── rutster-signaling-sip/ # stub until spearhead step 5
│ ├── rutster-tap/ # stub until spearhead step 2
│ └── rutster-spend/ # stub until spearhead step 6
└── fuzz/ # placeholder cargo-fuzz dir (real harnesses: step 5)
```
The three stub crates (`rutster-signaling-sip`, `rutster-tap`,
`rutster-spend`) exist to lock the ADR-0002 boundary shape without
anticipating code. Each is a `lib.rs` with a `//!` module doc comment
describing what will land there and when. Don't fill them in early —
see [`AGENTS.md`](../AGENTS.md) "Slice-1 boundaries — what NOT to add."
### Dependency direction
- `rutster` (binary) → `rutster-media`, `rutster-call-model`
- `rutster-media``rutster-call-model`
- `rutster-call-model` is a leaf (depends on nothing in the workspace)
- The three stubs depend on nothing in slice 1
---
## Build / test / lint commands
The four CI gates — all must pass before merge. Run them locally before
pushing:
```bash
cargo fmt --check # formatting check
cargo clippy -- -D warnings # lints (warnings = failures)
cargo test --all # all unit + integration tests
cargo deny check # licenses, advisories, bans, sources
```
To auto-fix formatting + lint warnings:
```bash
cargo fmt # apply formatting
cargo clippy --fix # apply lint suggestions
cargo clippy --fix --all-features # apply across all features
```
### Per-crate iteration
When working on one crate, skip the rest for faster cycles:
```bash
cargo test -p rutster-media # one crate's tests
cargo test -p rutster-media -- --nocapture # see println! output
cargo run -p rutster # run the binary
cargo build -p rutster-call-model # type-check one crate
cargo doc -p rutster-media --open # render one crate's docs
```
### Render full API docs
```bash
cargo doc --no-deps --open
```
Every module has `//!` module docs explaining what it does + why it
exists in the architecture. `cargo doc` is genuinely useful here, not
just ceremony — the user is learning Rust from this codebase and slice 1
carries thorough educational comments per spec §7.
---
## Running the binary
```bash
cargo run -p rutster
# listening on http://0.0.0.0:8080
```
Open <http://localhost:8080/> in a browser → click "Start call" → grant
mic → hear yourself echo. See [`QUICKSTART.md`](QUICKSTART.md) for
the full walkthrough + troubleshooting.
Verbose tracing:
```bash
RUST_LOG=rutster=debug cargo run -p rutster
```
Filter to one module:
```bash
RUST_LOG=rutster_media::loop_driver=trace,rutster=info cargo run -p rutster
```
---
## The 20 ms media loop & the "drop + observe" rule
The hot path is the str0m poll loop in
`crates/rutster-media/src/loop_driver.rs`. It runs every ~10 ms (tokio
interval, slice-1 deviation per spec §3.4 — step 4 lands a dedicated
timing thread).
Error policy on the hot path: **never** `?`-propagate. Match-and-continue.
A dropped packet must not terminate the peer. Policy: "drop + observe
(log + counter), don't crash." The eventual fuzz harness (step 5) will
test against this exact posture.
Cold path (signaling, setup, request handlers) uses `thiserror`-derived
error enums + `?` propagation, converted to HTTP status codes at the
axum boundary.
See [`AGENTS.md`](../AGENTS.md) "Error handling" for the full policy.
---
## Shipped vs. deferred
Slice 1 proves the **media core** only: WebRTC termination + the
Opus⇄PCM codec boundary. The tap, the brain, barge-in, the trunk, and
the spend cap are all explicitly deferred — each lands in a subsequent
spearhead step.
The full "what's deferred + when it returns" table is in
[`docs/superpowers/specs/2026-06-28-slice-1-webrtc-loopback-design.md` §1.2](superpowers/specs/2026-06-28-slice-1-webrtc-loopback-design.md).
The short version — **don't add these in slice 1:**
- ❌ Dedicated timing thread (step 4)
- ❌ TLS on the HTTP signaling surface (step 5)
- ❌ Authn / authz / multi-tenancy (step 6)
- ❌ Trickle ICE (when NATs demand it)
- ❌ The tap itself (step 2 — slice 1 only *pre-paves* the seam)
- ❌ The brain / STT / LLM / TTS (step 3)
- ❌ Barge-in / VAD (step 4)
- ❌ PSTN trunk / SIP (step 5)
- ❌ Spend cap (step 6)
- ❌ Event bus / Valkey / CDR emission (step 5)
- ❌ Transfer / park / pickup / barge features (escalation rung 2)
- ❌ Browser automation e2e tests (post-slice-1)
- ❌ Docker / compose (later rung)
If you find yourself reaching for any of these, the right answer is
"no, see slice-1 spec §1.2."
---
## Dev loop gotchas
- **libopus system dependency.** The `opus` crate links system libopus
via FFI. If you see `error: linking with cc failed: exit code`, install
`libopus-dev` (Debian/Ubuntu) / `opus-devel` (Fedora) / `brew install opus`
(macOS). See [`QUICKSTART.md`](QUICKSTART.md) for platform-specific
commands.
- **`cargo deny` first-time setup.** Run `cargo install cargo-deny --locked`
once. `cargo deny check` then validates licenses, advisories, and
duplicate-version bans.
- **Editor IDE.** `rust-analyzer` is recommended — it's what the plan was
written against. Open the repo root in your editor; `rust-analyzer`
picks up `rust-toolchain.toml` and the workspace manifest
automatically.
- **Slow first build.** str0m + axum + tokio compile fresh on first
build (~2 minutes). Incremental builds are fast. Use
`cargo check -p <crate>` instead of `cargo build` for type-check-only.

115
docs/QUICKSTART.md Normal file
View File

@@ -0,0 +1,115 @@
# Quickstart
Get Rutster running and hear your own voice echoed back in under 5 minutes.
> **Status:** Slice 1 (WebRTC media loopback) is the active build target.
> If the workspace isn't on `main` yet, check the `slice-1-webrtc-loopback`
> branch — that's where the implementation is landing task-by-task.
> See [`docs/superpowers/specs/2026-06-28-slice-1-webrtc-loopback-design.md`](superpowers/specs/2026-06-28-slice-1-webrtc-loopback-design.md)
> for the full design.
---
## Prerequisites
### 1. Rust toolchain
Install via [rustup](https://rustup.rs/):
```bash
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
```
The repo pins a specific stable channel in `rust-toolchain.toml``rustup`
will pick it up automatically on first `cargo` invocation. No manual
toolchain selection needed.
### 2. libopus (FFI dependency)
The `opus` crate links system libopus via FFI (per PORT_PLAN §7's
"🦀 Core (FFI)" disposition — Opus is the codec surface Rust doesn't need
to re-implement). Install the dev headers:
| Platform | Command |
|---|---|
| Debian/Ubuntu | `sudo apt-get install -y libopus-dev` |
| Fedora | `sudo dnf install -y opus-devel` |
| Arch | `sudo pacman -S opus` |
| macOS (Homebrew) | `brew install opus` |
Verify: `pkg-config --cflags opus` should print a path with no error.
That's the only system dependency in slice 1. Everything else is pure
Rust from crates.io.
---
## Run the server
```bash
cargo run
# listening on http://0.0.0.0:8080
```
First build takes ~2 minutes (str0m + axum + tokio compile fresh).
Subsequent builds are incremental.
---
## Hear the echo
1. Open a browser to <http://localhost:8080/>.
2. Click **Start call**.
3. Grant microphone permission when the browser prompts.
4. Speak — you should hear yourself back within ~200 ms
(no perceptible delay).
5. Click **Hang up** to tear down. The server logs
`Closing → Closed` for the session.
Verbose tracing for debugging:
```bash
RUST_LOG=rutster=debug cargo run
```
---
## Troubleshooting
| Symptom | Likely cause / fix |
|---|---|
| `error: linking with cc failed` / `could not find opus` | libopus dev headers not installed. Re-run the install command above. |
| Browser shows no mic prompt | Another tab/app holding the mic, or mic permissions disabled for `localhost`. Check browser settings. |
| `ICE connection failed` in the browser | Shouldn't happen on loopback (host candidates only). If it does, check the server console for the str0m error. |
| Click Start call, nothing happens | Open the browser console (F12). The page logs ICE state + connection state to a `<pre>` element. Look for the failure there. |
| Port 8080 already in use | Another process holding the port. Either stop it or edit `crates/rutster/src/main.rs` to bind a different port. |
The browser test page at `GET /` is a single self-contained HTML file
with inline JS — no build step. View source to see exactly what the
client side is doing.
---
## What's happening
When you click "Start call":
1. Browser captures microphone audio via `getUserMedia`.
2. Browser creates an `RTCPeerConnection` and generates an SDP offer
(audio-only, Opus codec).
3. Browser POSTs the offer to `POST /v1/sessions/:id/offer`.
4. The Rutster core (built on [`str0m`](https://docs.rs/str0m), a sans-IO
WebRTC implementation) accepts the offer, generates an SDP answer with
its DTLS fingerprint + ICE credentials.
5. Browser sets the answer as remote description; ICE + DTLS handshake
completes.
6. RTP starts flowing: browser → core terminates DTLS-SRTP → decodes
Opus to 16-bit PCM @ 24 kHz mono → echoes PCM back → re-encodes to
Opus → DTLS-SRTP → browser plays it.
The "codec-to-PCM boundary" is the canonical point where, in a future
slice, the audio tap for an external AI brain splices in. Slice 1 just
echoes; step 2 of the spearhead swaps the echo for a real tap.
For the why, see [`ARCHITECTURE.md`](ARCHITECTURE.md). For the dev loop,
see [`DEVELOPMENT.md`](DEVELOPMENT.md).