Remove slop

Author: MasterPtato

.claude/skills/sanity-check/SKILL.md

@@ -15,7 +15,10 @@ Run a quick end-to-end sanity check of a published rivetkit version: copy the he
 
 ## Inputs
 1. **Version or tag** (required) — explicit pkg-pr-new preview, npm dist-tag, or semver. If not provided in the user's message, ask for it.
-2. **Additional test behavior** (optional) — e.g., "also verify workflows persist" or "check that KV works." If provided, extend `src/index.ts` + `test.mjs` using the menu in "Extending with custom tests" below.
+2. **Engine mode** (optional, defaults to bundled) — the hello-world example's `registry.start()` only spawns the bundled engine binary when `RIVET_RUN_ENGINE=1` is set. Without it, the runtime boots in serverful mode and immediately fails trying to connect to a non-existent engine at `:6420` (`get local datacenters` → `Connection refused`).
+   - **bundled engine** (default): export `RIVET_RUN_ENGINE=1` before `node test.mjs`. Pairs the published `@rivetkit/engine-cli` binary with the published `rivetkit` runtime — what you want for end-to-end version verification.
+   - **external engine**: only use if the user explicitly says they have their own engine already running on `:6420` (e.g. `self-host/compose/dev`). Do not set the env var.
+3. **Additional test behavior** (optional) — e.g., "also verify workflows persist" or "check that KV works." If provided, extend `src/index.ts` + `test.mjs` using the menu in "Extending with custom tests" below.
 
 ## Usage
 - `/sanity-check <version>` — run in a temp directory on the host
@@ -151,22 +154,25 @@ try {
 
 ### 3. Install + run
 
-**Default (host):**
+**Default (host, bundled engine):**
 ```bash
 cd "$SANITY_DIR"
 npm install
-node test.mjs
+RIVET_RUN_ENGINE=1 node test.mjs
 ```
 
+Drop `RIVET_RUN_ENGINE=1` only if the user explicitly said they're running an external engine on `:6420`.
+
 If you need to inspect a failure after the fact, tee the output:
 ```bash
-node test.mjs 2>&1 | tee /tmp/sanity-check.log
+RIVET_RUN_ENGINE=1 node test.mjs 2>&1 | tee /tmp/sanity-check.log
 echo "exit=$?"
 ```
 
 **Docker mode:**
 ```bash
 docker run --rm \
+  -e RIVET_RUN_ENGINE=1 \
   -v "$SANITY_DIR":/app \
   -w /app \
   node:22 \

engine/sdks/rust/envoy-client/src/actor.rs

@@ -408,39 +408,12 @@ async fn begin_stop(
 	_http_request_tasks: &mut JoinSet<()>,
 	reason: protocol::StopActorReason,
 ) -> StopProgress {
-	// A Lost stop must surface as Stopped(Error). The runner side detected its
-	// own WS to pegboard-envoy was unhealthy and gave up on the actor; that is
-	// not a graceful exit. If we emitted Stopped(Ok) here, pegboard's
-	// `handle_stopped` would see Stopped(Ok) from `Transition::Running` (no
-	// prior `ActorIntent` was sent) and take `Decision::Destroy`, wiping the
-	// actor and its KV after every transient WS flap that exceeds
-	// `envoy_lost_threshold`.
-	let (mut stop_code, mut stop_message) = if let Some(err) = ctx.error.clone() {
-		(protocol::StopCode::Error, Some(err))
-	} else if matches!(reason, protocol::StopActorReason::Lost) {
-		(
-			protocol::StopCode::Error,
-			Some("envoy connection lost".to_string()),
-		)
+	let mut stop_code = if ctx.error.is_some() {
+		protocol::StopCode::Error
 	} else {
-		(protocol::StopCode::Ok, None)
+		protocol::StopCode::Ok
 	};
-
-	let reason_label = stop_actor_reason_label(&reason);
-	tracing::info!(
-		?reason,
-		?stop_code,
-		error = ?ctx.error,
-		"actor stopping"
-	);
-	crate::metrics::METRICS
-		.actor_stop_total
-		.with_label_values(&[reason_label])
-		.inc();
-	crate::metrics::METRICS
-		.actor_lifetime_seconds
-		.with_label_values(&[reason_label])
-		.observe(ctx.started_at.elapsed().as_secs_f64());
+	let mut stop_message = ctx.error.clone();
 	let (stop_tx, mut stop_rx) = oneshot::channel();
 
 	let stop_result = ctx