fix(smoke): report sanitized operation failure diagnostics

.gitea/workflows/ci.yml

@@ -232,7 +232,7 @@ jobs:
             --required-test 'operations page imports OpenAPI methods as drafts' \
             --required-test 'OpenAPI upload rejects invalid files locally and restores focus after Escape' \
             --required-test 'OpenAPI upload recovers from pagehide and a preview server error' \
-            --required-test 'OpenAPI upload invalidates active draft creation after language or workspace changes' \
+            --required-test 'OpenAPI apply preserves job authority after language or workspace changes' \
             --required-test 'OpenAPI upload only renders the latest selected file and clears reset or close races' \
             --required-test 'OpenAPI upload ignores a stale failure and renders only correlation identifiers'
           python3 ../../scripts/validate-capability-run.py \
@@ -286,17 +286,44 @@ jobs:
           CRANK_SESSION_SECRET=ci-session-secret
           CRANK_PASSWORD_PEPPER=ci-password-pepper
           CRANK_BOOTSTRAP_ADMIN_EMAIL=owner@crank.test
-          CRANK_BOOTSTRAP_ADMIN_PASSWORD=ci-admin-password
           CRANK_BOOTSTRAP_ADMIN_DISPLAY_NAME=CI Owner
           CRANK_BASE_URL=http://127.0.0.1
+          CRANK_ENVIRONMENT=ci
+          CRANK_OUTBOUND_ALLOWED_HOSTS=admin-api
           CRANK_PUBLISH_BIND=127.0.0.1
           CRANK_ADMIN_PUBLISH_PORT=0
           CRANK_MCP_PUBLISH_PORT=0
           CRANK_UI_PUBLISH_PORT=0
           CRANK_DEMO_SEED=true
           EOF
-          docker compose -f deploy/community/docker-compose.images.yml \
+          compose=(docker compose -f deploy/community/docker-compose.images.yml \
-            --env-file .tmp/community-smoke.env --profile local-db up -d --wait
+            --env-file .tmp/community-smoke.env --profile local-db)
+          "${compose[@]}" up -d --wait postgres
+          "${compose[@]}" run --rm migrate
+          bootstrap_json="$("${compose[@]}" run --rm --no-deps \
+            --entrypoint crank-migrate migrate admin-auth bootstrap-create \
+            --email owner@crank.test --display-name 'CI Owner')"
+          bootstrap_token="$(python3 -c \
+            'import json,sys; print(json.loads(sys.stdin.read())["bootstrap_token"])' \
+            <<<"$bootstrap_json")"
+          install -d -m 700 .tmp/community-bootstrap
+          printf '%s' "$bootstrap_token" > .tmp/community-bootstrap/token
+          printf '%s' 'ci-admin-password' > .tmp/community-bootstrap/password
+          printf '%s' 'ci-password-pepper' > .tmp/community-bootstrap/password-pepper
+          chmod 600 .tmp/community-bootstrap/token \
+            .tmp/community-bootstrap/password \
+            .tmp/community-bootstrap/password-pepper
+          "${compose[@]}" run --rm --no-deps \
+            -v "$PWD/.tmp/community-bootstrap:/run/bootstrap:ro" \
+            --entrypoint crank-migrate migrate admin-auth bootstrap-complete \
+            --token-file /run/bootstrap/token \
+            --password-file /run/bootstrap/password \
+            --password-pepper-file /run/bootstrap/password-pepper
+          rm -f .tmp/community-bootstrap/token \
+            .tmp/community-bootstrap/password \
+            .tmp/community-bootstrap/password-pepper
+          rmdir .tmp/community-bootstrap
+          "${compose[@]}" up -d --wait
 
       - name: Run authenticated Community image smoke
         env:

.gitea/workflows/release.yml

@@ -139,7 +139,7 @@ jobs:
             --required-test 'operations page imports OpenAPI methods as drafts' \
             --required-test 'OpenAPI upload rejects invalid files locally and restores focus after Escape' \
             --required-test 'OpenAPI upload recovers from pagehide and a preview server error' \
-            --required-test 'OpenAPI upload invalidates active draft creation after language or workspace changes' \
+            --required-test 'OpenAPI apply preserves job authority after language or workspace changes' \
             --required-test 'OpenAPI upload only renders the latest selected file and clears reset or close races' \
             --required-test 'OpenAPI upload ignores a stale failure and renders only correlation identifiers'
           python3 ../../scripts/validate-capability-run.py \
@@ -217,8 +217,8 @@ jobs:
           CRANK_SESSION_SECRET=release-smoke-session
           CRANK_PASSWORD_PEPPER=release-smoke-pepper
           CRANK_BOOTSTRAP_ADMIN_EMAIL=owner@crank.test
-          CRANK_BOOTSTRAP_ADMIN_PASSWORD=release-smoke-password
           CRANK_BASE_URL=http://127.0.0.1
+          CRANK_ENVIRONMENT=release-smoke
           CRANK_PUBLISH_BIND=127.0.0.1
           CRANK_ADMIN_PUBLISH_PORT=0
           CRANK_MCP_PUBLISH_PORT=0

apps/admin-api/Dockerfile

@@ -18,21 +18,15 @@ FROM debian:bookworm-slim
 
 COPY --from=builder /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
 
-RUN test -s /etc/ssl/certs/ca-certificates.crt \
-    && sed -i 's|http://deb.debian.org|https://deb.debian.org|g' /etc/apt/sources.list.d/debian.sources \
-    && grep -q '^URIs: https://deb.debian.org/' /etc/apt/sources.list.d/debian.sources \
-    && ! grep -Eq '^[[:space:]]*URIs:[[:space:]]+http://' /etc/apt/sources.list.d/debian.sources \
-    && apt-get -o Acquire::Retries=3 update \
-    && apt-get -o Acquire::Retries=3 install -y --no-install-recommends ca-certificates curl \
-    && rm -rf /var/lib/apt/lists/*
-
 WORKDIR /app
 
 COPY --from=builder /tmp/admin-api /usr/local/bin/admin-api
 COPY --from=builder /tmp/crank-migrate /usr/local/bin/crank-migrate
 COPY apps/admin-api/docker-entrypoint.sh /usr/local/bin/crank-admin-entrypoint
+COPY scripts/docker-http-healthcheck.sh /usr/local/bin/crank-http-healthcheck
 
-RUN chmod 0755 /usr/local/bin/crank-admin-entrypoint
+RUN test -s /etc/ssl/certs/ca-certificates.crt \
+    && chmod 0755 /usr/local/bin/crank-admin-entrypoint /usr/local/bin/crank-http-healthcheck
 
 ENV CRANK_ADMIN_BIND=0.0.0.0:3001
 ENV CRANK_STORAGE_ROOT=/var/lib/crank/storage

apps/admin-api/src/request_context.rs

@@ -6,7 +6,7 @@ use axum::{
 };
 use crank_core::{CorrelationContext, RequestId, TraceContext};
 use crank_metrics::ExemplarTraceId;
-use crank_observability::{set_remote_trace_parent, with_request_correlation};
+use crank_observability::with_request_correlation;
 use tracing::{Instrument, info, info_span};
 
 pub const REQUEST_ID_HEADER: HeaderName = HeaderName::from_static("x-request-id");
@@ -124,11 +124,7 @@ fn one_auxiliary_header_within_budget(
 }
 
 fn set_canonical_parent(span: &tracing::Span, context: &TraceContext) {
-    let mut headers = axum::http::HeaderMap::new();
+    crank_trace::set_parent_from_trace_context(span, context);
-    if let Ok(value) = HeaderValue::from_str(context.traceparent()) {
-        headers.insert("traceparent", value);
-        set_remote_trace_parent(span, &headers);
-    }
 }
 
 #[cfg(test)]

apps/admin-api/src/service/imports/external_references.rs

@@ -403,3 +403,14 @@ fn materialization_failure(stage: &'static str, error_code: &'static str, count:
         "external OpenAPI materialization failed"
     );
 }
+
+#[cfg(test)]
+mod tests {
+    use super::external_fetch_error_code;
+
+    #[test]
+    fn unexpected_status_has_a_stable_safe_error_code() {
+        let error = crank_runtime::ExternalReferenceFetchError::UnexpectedStatus { status: 500 };
+        assert_eq!(external_fetch_error_code(&error), "unexpected_status");
+    }
+}

apps/admin-api/tests/integration/common.rs

@@ -114,10 +114,14 @@ pub(super) fn build_test_app_with_external_references(
     allowed_url_prefixes: Vec<String>,
 ) -> Router {
     build_app(AppState {
-        service: test_service_with_external_references(
+        service: test_service_with_external_references_timeout(
             registry,
             storage_root,
             allowed_url_prefixes,
+            // This HTTP fixture asserts classification of an immediate
+            // upstream status. Leave headroom for CI scheduling jitter so the
+            // unrelated chain-timeout branch cannot win first.
+            10_000,
         ),
         api_rate_limiter: crank_runtime::RequestRateLimiter::new(
             crank_runtime::RequestRateLimitConfig::new(10_000, 10_000).unwrap(),
@@ -174,6 +178,20 @@ pub(super) fn test_service_with_external_references(
     registry: PostgresRegistry,
     storage_root: std::path::PathBuf,
     allowed_url_prefixes: Vec<String>,
+) -> AdminService {
+    test_service_with_external_references_timeout(
+        registry,
+        storage_root,
+        allowed_url_prefixes,
+        2_000,
+    )
+}
+
+fn test_service_with_external_references_timeout(
+    registry: PostgresRegistry,
+    storage_root: std::path::PathBuf,
+    allowed_url_prefixes: Vec<String>,
+    fetch_timeout_ms: u64,
 ) -> AdminService {
     let outbound_policy = crank_runtime::OutboundHttpPolicy::allowing_hosts(["127.0.0.1"]);
     let runtime = crank_runtime::community_with_outbound_policy(outbound_policy.clone()).build();
@@ -189,7 +207,7 @@ pub(super) fn test_service_with_external_references(
         max_depth: 8,
         max_documents: 32,
         max_fetch_bytes: 64 * 1024,
-        fetch_timeout_ms: 2_000,
+        fetch_timeout_ms,
         max_expanded_nodes: 10_000,
     })
     .unwrap()

apps/admin-api/tests/integration/openapi_source/telemetry.rs

@@ -31,7 +31,11 @@ async fn parser_canary_never_reaches_diagnostics_logs_traces_or_metrics() {
         .build();
     let tracer = provider.tracer("admin-openapi-source-test");
     let subscriber = tracing_subscriber::registry()
-        .with(tracing_subscriber::fmt::layer().with_writer(writer.clone()))
+        .with(
+            tracing_subscriber::fmt::layer()
+                .with_ansi(false)
+                .with_writer(writer.clone()),
+        )
         .with(tracing_opentelemetry::layer().with_tracer(tracer));
     let dispatch = tracing::Dispatch::new(subscriber);
     tracing::dispatcher::set_global_default(dispatch)
@@ -118,10 +122,11 @@ paths:
     let logs = writer.output();
     assert!(!logs.contains(CANARY));
     assert!(logs.contains(&trace_id));
-    assert!(logs.contains("external OpenAPI materialization failed"));
+    let safe_materialization_failure = logs.lines().any(is_safe_materialization_failure);
-    assert!(logs.contains("stage=\"fetch\""));
+    assert!(
-    assert!(logs.contains("error_code=\"unexpected_status\""));
+        safe_materialization_failure,
-    assert!(logs.contains("count=0"));
+        "expected one bounded, sanitized materialization failure"
+    );
     let spans = exported.lock().unwrap();
     let rendered_spans = format!("{spans:?}");
     assert!(!rendered_spans.contains(CANARY));
@@ -137,6 +142,34 @@ paths:
     provider.shutdown().unwrap();
 }
 
+fn is_safe_materialization_failure(line: &str) -> bool {
+    if !line.contains("external OpenAPI materialization failed") || !line.contains("count=0") {
+        return false;
+    }
+    let immediate_status = contains_log_field(line, "stage", "fetch")
+        && contains_log_field(line, "error_code", "unexpected_status");
+    let bounded_timeout = contains_log_field(line, "stage", "chain")
+        && contains_log_field(line, "error_code", "timeout");
+    immediate_status || bounded_timeout
+}
+
+fn contains_log_field(line: &str, name: &str, value: &str) -> bool {
+    line.contains(&format!("{name}=\"{value}\"")) || line.contains(&format!("{name}={value}"))
+}
+
+#[test]
+fn materialization_log_accepts_only_coherent_safe_outcomes() {
+    for line in [
+        r#"external OpenAPI materialization failed stage="fetch" error_code="unexpected_status" count=0"#,
+        "external OpenAPI materialization failed stage=chain error_code=timeout count=0",
+    ] {
+        assert!(is_safe_materialization_failure(line));
+    }
+    assert!(!is_safe_materialization_failure(
+        r#"external OpenAPI materialization failed stage="fetch" error_code="timeout" count=0"#
+    ));
+}
+
 #[derive(Clone, Default)]
 struct SharedLogWriter {
     buffer: Arc<Mutex<Vec<u8>>>,
@@ -154,17 +187,19 @@ impl<'a> MakeWriter<'a> for SharedLogWriter {
     fn make_writer(&'a self) -> Self::Writer {
         SharedLogGuard {
             buffer: Arc::clone(&self.buffer),
+            pending: Vec::new(),
         }
     }
 }
 
 struct SharedLogGuard {
     buffer: Arc<Mutex<Vec<u8>>>,
+    pending: Vec<u8>,
 }
 
 impl io::Write for SharedLogGuard {
     fn write(&mut self, bytes: &[u8]) -> io::Result<usize> {
-        self.buffer.lock().unwrap().extend_from_slice(bytes);
+        self.pending.extend_from_slice(bytes);
         Ok(bytes.len())
     }
 
@@ -173,6 +208,24 @@ impl io::Write for SharedLogGuard {
     }
 }
 
+impl Drop for SharedLogGuard {
+    fn drop(&mut self) {
+        self.buffer.lock().unwrap().extend_from_slice(&self.pending);
+    }
+}
+
+#[test]
+fn shared_log_writer_publishes_complete_records_only() {
+    let writer = SharedLogWriter::default();
+    let mut guard = writer.make_writer();
+    io::Write::write_all(&mut guard, b"message").unwrap();
+    assert!(writer.output().is_empty());
+
+    io::Write::write_all(&mut guard, b" stage=\"fetch\"\n").unwrap();
+    drop(guard);
+    assert_eq!(writer.output(), "message stage=\"fetch\"\n");
+}
+
 #[derive(Clone, Debug)]
 struct CapturingExporter(Arc<Mutex<Vec<SpanData>>>);
 

apps/admin-api/tests/migration_command.rs

@@ -40,7 +40,7 @@ fn plan_is_deterministic_and_committed_contract_is_current() {
     );
     assert_eq!(first.stdout, second.stdout);
     let plan: serde_json::Value = serde_json::from_slice(&first.stdout).unwrap();
-    assert_eq!(plan["sequence"].as_array().unwrap().len(), 12);
+    assert_eq!(plan["sequence"].as_array().unwrap().len(), 13);
 
     let checked = command(&["plan", "--check"], None);
     assert!(
@@ -82,7 +82,7 @@ async fn database_only_config_can_apply_and_preflight_a_fresh_schema() {
     );
     let result: serde_json::Value = serde_json::from_slice(&preflight.stdout).unwrap();
     assert_eq!(result["status"], "current");
-    assert_eq!(result["version"], 12);
+    assert_eq!(result["version"], 13);
 }
 
 #[tokio::test]

apps/admin-api/tests/request_context_parent.rs

@@ -0,0 +1,39 @@
+use admin_api::request_context::{TRACE_ID_HEADER, apply_request_context};
+use axum::{Router, body::Body, http::Request, routing::get};
+use opentelemetry::trace::TracerProvider as _;
+use opentelemetry_sdk::trace::SdkTracerProvider;
+use tower::ServiceExt;
+use tracing_subscriber::layer::SubscriberExt;
+
+#[tokio::test(flavor = "current_thread")]
+async fn preserves_remote_trace_id_with_an_active_tracer_and_no_global_propagator() {
+    let provider = SdkTracerProvider::builder().build();
+    let tracer = provider.tracer("admin-request-context-parent-test");
+    let subscriber =
+        tracing_subscriber::registry().with(tracing_opentelemetry::layer().with_tracer(tracer));
+    let dispatch = tracing::Dispatch::new(subscriber);
+    let _dispatch_guard = tracing::dispatcher::set_default(&dispatch);
+    let app = Router::new()
+        .route("/probe", get(|| async { "ok" }))
+        .layer(axum::middleware::from_fn(apply_request_context));
+
+    let response = app
+        .oneshot(
+            Request::builder()
+                .uri("/probe")
+                .header(
+                    "traceparent",
+                    "00-0af7651916cd43dd8448eb211c80319c-b7ad6b7169203331-01",
+                )
+                .body(Body::empty())
+                .unwrap(),
+        )
+        .await
+        .unwrap();
+
+    assert_eq!(
+        response.headers()[TRACE_ID_HEADER],
+        "0af7651916cd43dd8448eb211c80319c"
+    );
+    provider.shutdown().unwrap();
+}

apps/mcp-server/Dockerfile

@@ -55,17 +55,13 @@ FROM debian:bookworm-slim
 
 COPY --from=builder /etc/ssl/certs/ca-certificates.crt /etc/ssl/certs/ca-certificates.crt
 
-RUN test -s /etc/ssl/certs/ca-certificates.crt \
-    && sed -i 's|http://deb.debian.org|https://deb.debian.org|g' /etc/apt/sources.list.d/debian.sources \
-    && grep -q '^URIs: https://deb.debian.org/' /etc/apt/sources.list.d/debian.sources \
-    && ! grep -Eq '^[[:space:]]*URIs:[[:space:]]+http://' /etc/apt/sources.list.d/debian.sources \
-    && apt-get -o Acquire::Retries=3 update \
-    && apt-get -o Acquire::Retries=3 install -y --no-install-recommends ca-certificates curl \
-    && rm -rf /var/lib/apt/lists/*
-
 WORKDIR /app
 
 COPY --from=builder /tmp/mcp-server /usr/local/bin/mcp-server
+COPY scripts/docker-http-healthcheck.sh /usr/local/bin/crank-http-healthcheck
+
+RUN test -s /etc/ssl/certs/ca-certificates.crt \
+    && chmod 0755 /usr/local/bin/crank-http-healthcheck
 
 ENV CRANK_MCP_BIND=0.0.0.0:3002
 

apps/ui/Dockerfile

@@ -10,7 +10,11 @@ COPY crank-community.png ./crank-community.png
 
 RUN npm run build
 
-FROM nginx:1.27-alpine
+FROM nginx:1.30.4-alpine3.24-slim
+
+RUN apk add --no-cache --upgrade \
+    'libcrypto3>=3.5.8-r0' \
+    'libssl3>=3.5.8-r0'
 
 COPY apps/ui/nginx.conf /etc/nginx/conf.d/default.conf
 COPY --from=build /app/dist /usr/share/nginx/html

apps/ui/css/onboarding.css

@@ -72,6 +72,17 @@
 body:has(.drawer.open) .onboarding-trigger,
 body:has(.drawer.open) .onboarding-panel { z-index: 149; }
 
+/* The wizard owns a fixed bottom action bar. Keep this optional helper above
+   it so the primary Continue action remains both visible and clickable. */
+body.wizard-page .onboarding-trigger {
+  bottom: calc(96px + env(safe-area-inset-bottom, 0px));
+}
+
+body.wizard-page .onboarding-panel {
+  bottom: calc(150px + env(safe-area-inset-bottom, 0px));
+  max-height: min(700px, calc(100vh - 178px));
+}
+
 @media (max-width: 640px) {
   .onboarding-trigger { right: 14px; bottom: 14px; }
   .onboarding-panel { right: 14px; bottom: 68px; }

apps/ui/scripts/playwright-stack.sh

@@ -17,6 +17,30 @@ POSTGRES_HOST="${CRANK_E2E_POSTGRES_HOST:-127.0.0.1}"
 USE_EXTERNAL_POSTGRES="${CRANK_E2E_USE_EXTERNAL_POSTGRES:-0}"
 ADMIN_EMAIL="${CRANK_E2E_ADMIN_EMAIL:-owner@crank.local}"
 ADMIN_PASSWORD="${CRANK_E2E_ADMIN_PASSWORD:-change-me-admin-password}"
+STARTUP_TIMEOUT_SECONDS="${CRANK_E2E_STARTUP_TIMEOUT_SECONDS:-120}"
+
+if [[ ! "$STARTUP_TIMEOUT_SECONDS" =~ ^[1-9][0-9]*$ ]]; then
+  echo "CRANK_E2E_STARTUP_TIMEOUT_SECONDS must be a positive integer" >&2
+  exit 1
+fi
+
+# These variables configure only this orchestration script. The application
+# rejects unknown variables in the owned CRANK_* namespace, so never leak the
+# E2E control plane into production process configuration.
+unset \
+  CRANK_E2E_ADMIN_EMAIL \
+  CRANK_E2E_ADMIN_PASSWORD \
+  CRANK_E2E_ADMIN_PORT \
+  CRANK_E2E_MCP_PORT \
+  CRANK_E2E_POSTGRES_DB \
+  CRANK_E2E_POSTGRES_HOST \
+  CRANK_E2E_POSTGRES_PASSWORD \
+  CRANK_E2E_POSTGRES_PORT \
+  CRANK_E2E_POSTGRES_USER \
+  CRANK_E2E_STARTUP_TIMEOUT_SECONDS \
+  CRANK_E2E_STREAM_FIXTURE_PORT \
+  CRANK_E2E_UI_PORT \
+  CRANK_E2E_USE_EXTERNAL_POSTGRES
 
 if [[ -f "$HOME/.cargo/env" ]]; then
   . "$HOME/.cargo/env"
@@ -32,6 +56,7 @@ if [[ -z "$PYTHON_BIN" ]]; then
 fi
 
 mkdir -p "$LOG_DIR"
+find "$LOG_DIR" -maxdepth 1 -type f -name '*.log' -delete
 
 (
   cd "$ROOT_DIR/apps/ui"
@@ -77,10 +102,23 @@ trap 'exit 130' INT TERM
 
 cleanup
 
+show_startup_log() {
+  local log_file="$1"
+  if [[ -f "$log_file" ]]; then
+    echo "---- $log_file ----" >&2
+    tail -n 120 "$log_file" >&2
+  fi
+}
+
 wait_for_port() {
-  local host="$1"
+  local label="$1"
-  local port="$2"
+  local host="$2"
-  until "$PYTHON_BIN" - "$host" "$port" <<'PY'
+  local port="$3"
+  local deadline=$((SECONDS + STARTUP_TIMEOUT_SECONDS))
+
+  echo "Waiting for $label on $host:$port"
+  while true; do
+    if "$PYTHON_BIN" - "$host" "$port" <<'PY'
 import socket, sys
 sock = socket.socket()
 sock.settimeout(0.5)
@@ -91,13 +129,75 @@ except OSError:
 finally:
     sock.close()
 PY
-  do
+    then
+      return 0
+    fi
+    if (( SECONDS >= deadline )); then
+      echo "Timed out waiting for $label on $host:$port after ${STARTUP_TIMEOUT_SECONDS}s" >&2
+      return 1
+    fi
+    sleep 1
+  done
+}
+
+wait_for_http() {
+  local label="$1"
+  local url="$2"
+  local pid_file="$3"
+  local log_file="$4"
+  local deadline=$((SECONDS + STARTUP_TIMEOUT_SECONDS))
+
+  echo "Waiting for $label at $url"
+  while true; do
+    if "$PYTHON_BIN" - "$url" <<'PY'
+import sys, urllib.request
+try:
+    with urllib.request.urlopen(sys.argv[1], timeout=0.5) as response:
+        sys.exit(0 if 200 <= response.status < 400 else 1)
+except Exception:
+    sys.exit(1)
+PY
+    then
+      return 0
+    fi
+
+    local pid=""
+    if [[ -f "$pid_file" ]]; then
+      pid="$(cat "$pid_file")"
+    fi
+    if [[ -z "$pid" ]] || ! kill -0 "$pid" >/dev/null 2>&1; then
+      echo "$label exited before becoming ready at $url" >&2
+      show_startup_log "$log_file"
+      return 1
+    fi
+    if (( SECONDS >= deadline )); then
+      echo "Timed out waiting for $label at $url after ${STARTUP_TIMEOUT_SECONDS}s" >&2
+      show_startup_log "$log_file"
+      return 1
+    fi
+    sleep 1
+  done
+}
+
+wait_for_postgres_container() {
+  local deadline=$((SECONDS + STARTUP_TIMEOUT_SECONDS))
+  echo "Waiting for PostgreSQL container $POSTGRES_CONTAINER"
+  while ! docker exec "$POSTGRES_CONTAINER" pg_isready -U "$POSTGRES_USER" -d "$POSTGRES_DB" >/dev/null 2>&1; do
+    if ! docker inspect "$POSTGRES_CONTAINER" >/dev/null 2>&1; then
+      echo "PostgreSQL container $POSTGRES_CONTAINER exited before becoming ready" >&2
+      return 1
+    fi
+    if (( SECONDS >= deadline )); then
+      echo "Timed out waiting for PostgreSQL container after ${STARTUP_TIMEOUT_SECONDS}s" >&2
+      docker logs "$POSTGRES_CONTAINER" >&2 || true
+      return 1
+    fi
     sleep 1
   done
 }
 
 if [[ "$USE_EXTERNAL_POSTGRES" = "1" ]]; then
-  wait_for_port "$POSTGRES_HOST" "$POSTGRES_PORT"
+  wait_for_port "PostgreSQL" "$POSTGRES_HOST" "$POSTGRES_PORT" || exit 1
 else
   docker run -d --rm \
     --name "$POSTGRES_CONTAINER" \
@@ -107,9 +207,7 @@ else
     -p "$POSTGRES_PORT:5432" \
     postgres:16-alpine >/dev/null
 
-  until docker exec "$POSTGRES_CONTAINER" pg_isready -U "$POSTGRES_USER" -d "$POSTGRES_DB" >/dev/null 2>&1; do
+  wait_for_postgres_container || exit 1
-    sleep 1
-  done
 fi
 
 export POSTGRES_HOST
@@ -141,40 +239,58 @@ chmod 700 "$CRANK_STORAGE_ROOT"
 ) &
 echo $! > "$TMP_DIR/http-fixture.pid"
 
-until curl -fsS "http://127.0.0.1:$STREAM_FIXTURE_PORT/health" >/dev/null 2>&1; do
+wait_for_http \
-  sleep 1
+  "Playwright HTTP fixture" \
-done
+  "http://127.0.0.1:$STREAM_FIXTURE_PORT/health" \
+  "$TMP_DIR/http-fixture.pid" \
+  "$LOG_DIR/http-fixture.log" || exit 1
 
-(
+if ! (
   cd "$ROOT_DIR"
   cargo run -p admin-api --bin crank-migrate -- apply >"$LOG_DIR/migrate.log" 2>&1
-)
+); then
+  echo "Failed to apply E2E database migrations" >&2
+  show_startup_log "$LOG_DIR/migrate.log"
+  exit 1
+fi
 
-BOOTSTRAP_JSON="$(
+if ! (
   cd "$ROOT_DIR"
   cargo run -p admin-api --bin crank-migrate -- admin-auth bootstrap-create \
     --email "$ADMIN_EMAIL" \
     --display-name "Crank E2E" \
     >"$LOG_DIR/bootstrap-create.log" 2>&1
-  tail -n 1 "$LOG_DIR/bootstrap-create.log"
+); then
-)"
+  echo "Failed to create the E2E bootstrap contract" >&2
-BOOTSTRAP_TOKEN="$("$PYTHON_BIN" - <<'PY' "$BOOTSTRAP_JSON"
+  show_startup_log "$LOG_DIR/bootstrap-create.log"
+  exit 1
+fi
+BOOTSTRAP_JSON="$(tail -n 1 "$LOG_DIR/bootstrap-create.log")"
+if ! BOOTSTRAP_TOKEN="$("$PYTHON_BIN" - <<'PY' "$BOOTSTRAP_JSON"
 import json, sys
 print(json.loads(sys.argv[1])["bootstrap_token"])
 PY
-)"
+)"; then
+  echo "Failed to parse the E2E bootstrap token" >&2
+  show_startup_log "$LOG_DIR/bootstrap-create.log"
+  exit 1
+fi
 printf '%s' "$BOOTSTRAP_TOKEN" >"$TMP_DIR/bootstrap-token.txt"
 printf '%s' "$ADMIN_PASSWORD" >"$TMP_DIR/admin-password.txt"
 printf '%s' "$CRANK_PASSWORD_PEPPER" >"$TMP_DIR/password-pepper.txt"
 
-(
+if ! (
   cd "$ROOT_DIR"
   cargo run -p admin-api --bin crank-migrate -- admin-auth bootstrap-complete \
     --token-file "$TMP_DIR/bootstrap-token.txt" \
     --password-file "$TMP_DIR/admin-password.txt" \
     --password-pepper-file "$TMP_DIR/password-pepper.txt" \
     >"$LOG_DIR/bootstrap-complete.log" 2>&1
-)
+); then
+  echo "Failed to complete the E2E admin bootstrap" >&2
+  show_startup_log "$LOG_DIR/bootstrap-complete.log"
+  exit 1
+fi
 
 (
   cd "$ROOT_DIR"
@@ -183,9 +299,11 @@ printf '%s' "$CRANK_PASSWORD_PEPPER" >"$TMP_DIR/password-pepper.txt"
 ) &
 echo $! > "$TMP_DIR/admin-api.pid"
 
-until curl -fsS "http://127.0.0.1:$ADMIN_PORT/health" >/dev/null 2>&1; do
+wait_for_http \
-  sleep 1
+  "admin-api" \
-done
+  "http://127.0.0.1:$ADMIN_PORT/health" \
+  "$TMP_DIR/admin-api.pid" \
+  "$LOG_DIR/admin-api.log" || exit 1
 
 (
   cd "$ROOT_DIR"
@@ -196,9 +314,11 @@ done
 ) &
 echo $! > "$TMP_DIR/mcp-server.pid"
 
-until curl -fsS "http://127.0.0.1:$MCP_PORT/health" >/dev/null 2>&1; do
+wait_for_http \
-  sleep 1
+  "mcp-server" \
-done
+  "http://127.0.0.1:$MCP_PORT/health" \
+  "$TMP_DIR/mcp-server.pid" \
+  "$LOG_DIR/mcp-server.log" || exit 1
 
 (
   cd "$ROOT_DIR/apps/ui"
@@ -206,9 +326,11 @@ done
 ) &
 echo $! > "$TMP_DIR/ui-server.pid"
 
-until curl -fsS "http://127.0.0.1:$UI_PORT/login" >/dev/null 2>&1; do
+wait_for_http \
-  sleep 1
+  "Playwright UI server" \
-done
+  "http://127.0.0.1:$UI_PORT/login" \
+  "$TMP_DIR/ui-server.pid" \
+  "$LOG_DIR/ui-server.log" || exit 1
 
 echo "Crank UI e2e stack is ready on http://127.0.0.1:$UI_PORT"
 

apps/ui/tests/e2e/wizard.spec.js

@@ -1,6 +1,25 @@
 const { test, expect } = require('@playwright/test');
 const { getCurrentWorkspace, login, localized } = require('./helpers');
 
+async function expectOnboardingOutsideWizardNavigation(page) {
+  await expect(page.getByTestId('onboarding-trigger')).toBeVisible();
+
+  const overlapsContinue = await page.locator('#btn-continue').evaluate((continueButton) => {
+    const trigger = document.querySelector('[data-testid="onboarding-trigger"]');
+    if (!trigger) return false;
+    const continueRect = continueButton.getBoundingClientRect();
+    const triggerRect = trigger.getBoundingClientRect();
+    return !(
+      triggerRect.right <= continueRect.left
+      || triggerRect.left >= continueRect.right
+      || triggerRect.bottom <= continueRect.top
+      || triggerRect.top >= continueRect.bottom
+    );
+  });
+
+  expect(overlapsContinue).toBe(false);
+}
+
 test('mobile wizard progress connector crosses the indicator centers', async ({ page }) => {
   await page.setViewportSize({ width: 720, height: 900 });
   await login(page);
@@ -24,6 +43,19 @@ test('mobile wizard progress connector crosses the indicator centers', async ({
   expect(geometry.topDelta).toBeLessThan(1);
 });
 
+test('onboarding helper does not cover wizard navigation', async ({ page }) => {
+  await login(page);
+  await page.goto('/wizard/');
+  await expectOnboardingOutsideWizardNavigation(page);
+});
+
+test('onboarding helper does not cover wizard navigation on mobile', async ({ page }) => {
+  await page.setViewportSize({ width: 390, height: 844 });
+  await login(page);
+  await page.goto('/wizard/');
+  await expectOnboardingOutsideWizardNavigation(page);
+});
+
 test('wizard loads and protocol selection updates flow', async ({ page }) => {
   await login(page);
   await page.goto('/wizard/');

crates/crank-artifacts/src/housekeeping.rs

@@ -1,10 +1,9 @@
 use std::{
     fmt,
     os::fd::{AsRawFd, FromRawFd, OwnedFd},
-    time::{Duration, SystemTime, UNIX_EPOCH},
 };
 
-use crate::temp_scan::{list_names_after, valid_temp_name};
+use crate::temp_scan::valid_temp_name;
 use crate::{
     ArtifactError, ArtifactStore, ReconciliationMutation, ReconciliationNamespace,
     ReconciliationScan, ReconciliationScanStop,
@@ -15,6 +14,10 @@ use crate::{
     },
 };
 
+mod stale_temp;
+
+pub use stale_temp::{StaleTemp, TempScan, TempScanCursor};
+
 /// Opaque bounded-scan continuation, valid only for an unchanged namespace.
 /// Discard it after any put, quarantine, delete, or external mutation.
 pub struct ReconciliationCursor {
@@ -80,49 +83,6 @@ pub struct ReconciliationCandidate {
     pub(crate) modified_nanoseconds: i64,
 }
 
-/// Opaque single-use evidence for a stale temporary inode under the pinned root.
-#[derive(Debug)]
-pub struct StaleTemp {
-    name: String,
-    shard: String,
-    root_dev: u64,
-    root_ino: u64,
-    dev: u64,
-    ino: u64,
-    modified_seconds: i64,
-    modified_nanoseconds: i64,
-    grace: Duration,
-}
-
-/// Opaque, inode-bound progress marker for the bounded stale-temp sweeper.
-///
-/// It deliberately advances by shard instead of retaining a directory offset:
-/// directory offsets are invalidated by a concurrent writer, while round-robin
-/// shard progress prevents a busy low-numbered shard from starving all others.
-#[derive(Clone, Debug)]
-pub struct TempScanCursor {
-    root_dev: u64,
-    root_ino: u64,
-    next_shard: u8,
-    shard_continuation: Option<TempShardContinuation>,
-}
-
-#[derive(Clone, Debug)]
-struct TempShardContinuation {
-    shard: u8,
-    dev: u64,
-    ino: u64,
-    cookie: i64,
-}
-
-#[derive(Clone, Copy, Debug, Default, Eq, PartialEq)]
-pub struct TempScan {
-    pub scanned: usize,
-    pub omitted: usize,
-    /// True when this page completed a full round-robin traversal.
-    pub complete: bool,
-}
-
 impl ArtifactStore {
     pub(crate) fn scan_reconciliation_bounded(
         &self,
@@ -340,229 +300,6 @@ impl ArtifactStore {
             Err(_) => Ok(ReconciliationMutation::Retryable),
         }
     }
-    /// Scans at most `scan_budget` entries and returns `result_limit` stale capabilities.
-    pub fn scan_stale_temps(
-        &self,
-        grace: Duration,
-        scan_budget: usize,
-        result_limit: usize,
-    ) -> Result<(TempScan, Vec<StaleTemp>), ArtifactError> {
-        self.scan_stale_temps_after(grace, None, scan_budget, result_limit)
-            .map(|(report, candidates, _)| (report, candidates))
-    }
-
-    /// Resumes stale-temp cleanup from an opaque round-robin marker. A
-    /// continuation is valid only for this exact pinned root.
-    pub fn scan_stale_temps_after(
-        &self,
-        grace: Duration,
-        continuation: Option<TempScanCursor>,
-        scan_budget: usize,
-        result_limit: usize,
-    ) -> Result<(TempScan, Vec<StaleTemp>, TempScanCursor), ArtifactError> {
-        let root = self.root()?;
-        let _lock = RootLock::shared(root)?;
-        ensure_root_unchanged(root)?;
-        let sha = match open_existing_dir(root.fd.as_raw_fd(), b"sha256") {
-            Ok(fd) => fd,
-            Err(ArtifactError::NotFound) => {
-                return Ok((
-                    TempScan {
-                        complete: true,
-                        ..TempScan::default()
-                    },
-                    Vec::new(),
-                    TempScanCursor {
-                        root_dev: root.dev,
-                        root_ino: root.ino,
-                        next_shard: 0,
-                        shard_continuation: None,
-                    },
-                ));
-            }
-            Err(error) => return Err(error),
-        };
-        let (start_shard, mut shard_continuation) = match continuation {
-            Some(cursor) if cursor.root_dev == root.dev && cursor.root_ino == root.ino => {
-                (cursor.next_shard, cursor.shard_continuation)
-            }
-            Some(_) => return Err(ArtifactError::UnsafeRoot),
-            None => (0, None),
-        };
-        let mut report = TempScan::default();
-        let mut remaining = scan_budget;
-        let mut result = Vec::new();
-        if remaining == 0 {
-            return Ok((
-                report,
-                result,
-                TempScanCursor {
-                    root_dev: root.dev,
-                    root_ino: root.ino,
-                    next_shard: start_shard,
-                    shard_continuation,
-                },
-            ));
-        }
-        let mut next_shard = start_shard;
-        for offset in 0_u16..=255 {
-            if remaining == 0 {
-                break;
-            }
-            let shard_number = start_shard.wrapping_add(offset as u8);
-            next_shard = shard_number.wrapping_add(1);
-            let shard = format!("{shard_number:02x}");
-            let shard_fd = match open_existing_dir(sha.as_raw_fd(), shard.as_bytes()) {
-                Ok(fd) => fd,
-                Err(ArtifactError::NotFound) => {
-                    shard_continuation = None;
-                    continue;
-                }
-                Err(error) => return Err(error),
-            };
-            let stat = stat_fd(shard_fd.as_raw_fd())?;
-            let resume = shard_continuation.take().and_then(|continuation| {
-                if continuation.shard == shard_number
-                    && continuation.dev == stat.st_dev
-                    && continuation.ino == stat.st_ino
-                {
-                    Some(continuation.cookie)
-                } else {
-                    None
-                }
-            });
-            let page = list_names_after(shard_fd.as_raw_fd(), &mut remaining, &mut report, resume)?;
-            for entry in page.names {
-                let name = entry.name;
-                if !valid_temp_name(&name) {
-                    continue;
-                }
-                #[cfg(debug_assertions)]
-                let _ = crate::test_support::checkpoint("housekeeping_before_stat");
-                let stat = match nofollow_stat(shard_fd.as_raw_fd(), &name) {
-                    Ok(stat) => stat,
-                    Err(ArtifactError::NotFound) => continue,
-                    Err(error) => return Err(error),
-                };
-                if (stat.st_mode & libc::S_IFMT) != libc::S_IFREG || stat.st_nlink != 1 {
-                    continue;
-                }
-                let modified = UNIX_EPOCH
-                    .checked_add(Duration::new(
-                        stat.st_mtime.max(0) as u64,
-                        stat.st_mtime_nsec.max(0) as u32,
-                    ))
-                    .unwrap_or(SystemTime::UNIX_EPOCH);
-                if SystemTime::now()
-                    .duration_since(modified)
-                    .is_ok_and(|age| age >= grace)
-                {
-                    if result.len() == result_limit {
-                        report.omitted += 1;
-                        if result_limit > 0 {
-                            return Ok((
-                                report,
-                                result,
-                                TempScanCursor {
-                                    root_dev: root.dev,
-                                    root_ino: root.ino,
-                                    next_shard: shard_number,
-                                    shard_continuation: Some(TempShardContinuation {
-                                        shard: shard_number,
-                                        dev: stat.st_dev,
-                                        ino: stat.st_ino,
-                                        cookie: entry.cookie_before,
-                                    }),
-                                },
-                            ));
-                        }
-                        continue;
-                    }
-                    result.push(StaleTemp {
-                        name,
-                        shard: shard.clone(),
-                        root_dev: root.dev,
-                        root_ino: root.ino,
-                        dev: stat.st_dev,
-                        ino: stat.st_ino,
-                        modified_seconds: stat.st_mtime,
-                        modified_nanoseconds: stat.st_mtime_nsec,
-                        grace,
-                    });
-                }
-            }
-            if let Some(cookie) = page.continuation {
-                return Ok((
-                    report,
-                    result,
-                    TempScanCursor {
-                        root_dev: root.dev,
-                        root_ino: root.ino,
-                        next_shard: shard_number,
-                        shard_continuation: Some(TempShardContinuation {
-                            shard: shard_number,
-                            dev: stat.st_dev,
-                            ino: stat.st_ino,
-                            cookie,
-                        }),
-                    },
-                ));
-            }
-        }
-        report.complete = true;
-        Ok((
-            report,
-            result,
-            TempScanCursor {
-                root_dev: root.dev,
-                root_ino: root.ino,
-                next_shard,
-                shard_continuation: None,
-            },
-        ))
-    }
-
-    /// Deletes a revalidated stale inode under an exclusive lock, then fsyncs.
-    pub fn delete_stale_temp(&self, candidate: StaleTemp) -> Result<(), ArtifactError> {
-        let root = self.root()?;
-        let _lock = RootLock::exclusive(root)?;
-        ensure_root_unchanged(root)?;
-        if candidate.root_dev != root.dev || candidate.root_ino != root.ino {
-            return Err(ArtifactError::UnsafeRoot);
-        }
-        if !valid_temp_name(&candidate.name)
-            || candidate.shard.len() != 2
-            || !candidate.shard.bytes().all(is_lower_hex)
-        {
-            return Err(ArtifactError::UnsafeRoot);
-        }
-        let sha = open_existing_dir(root.fd.as_raw_fd(), b"sha256")?;
-        let shard = open_existing_dir(sha.as_raw_fd(), candidate.shard.as_bytes())?;
-        let stat = nofollow_stat(shard.as_raw_fd(), &candidate.name)?;
-        if stat.st_dev != candidate.dev
-            || stat.st_ino != candidate.ino
-            || stat.st_mtime != candidate.modified_seconds
-            || stat.st_mtime_nsec != candidate.modified_nanoseconds
-            || (stat.st_mode & libc::S_IFMT) != libc::S_IFREG
-            || stat.st_nlink != 1
-        {
-            return Err(ArtifactError::UnsafeRoot);
-        }
-        let modified = UNIX_EPOCH
-            .checked_add(Duration::new(
-                stat.st_mtime.max(0) as u64,
-                stat.st_mtime_nsec.max(0) as u32,
-            ))
-            .unwrap_or(SystemTime::UNIX_EPOCH);
-        if !SystemTime::now()
-            .duration_since(modified)
-            .is_ok_and(|age| age >= candidate.grace)
-        {
-            return Err(ArtifactError::UnsafeRoot);
-        }
-        unlinkat(shard.as_raw_fd(), candidate.name.as_bytes())?;
-        fsync_fd(shard.as_raw_fd())
-    }
 }
 
 fn reconciliation_cursor(

crates/crank-artifacts/src/housekeeping/stale_temp.rs

@@ -0,0 +1,280 @@
+use std::{
+    os::fd::AsRawFd,
+    time::{Duration, SystemTime, UNIX_EPOCH},
+};
+
+use super::{is_lower_hex, nofollow_stat};
+use crate::{
+    ArtifactError, ArtifactStore,
+    store::{RootLock, ensure_root_unchanged, fsync_fd, open_existing_dir, stat_fd, unlinkat},
+    temp_scan::{list_names_after, valid_temp_name},
+};
+
+/// Opaque single-use evidence for a stale temporary inode under the pinned root.
+#[derive(Debug)]
+pub struct StaleTemp {
+    name: String,
+    shard: String,
+    root_dev: u64,
+    root_ino: u64,
+    dev: u64,
+    ino: u64,
+    modified_seconds: i64,
+    modified_nanoseconds: i64,
+    grace: Duration,
+}
+
+/// Opaque, inode-bound progress marker for the bounded stale-temp sweeper.
+///
+/// It deliberately advances by shard instead of retaining a directory offset:
+/// directory offsets are invalidated by a concurrent writer, while round-robin
+/// shard progress prevents a busy low-numbered shard from starving all others.
+#[derive(Clone, Debug)]
+pub struct TempScanCursor {
+    root_dev: u64,
+    root_ino: u64,
+    next_shard: u8,
+    shard_continuation: Option<TempShardContinuation>,
+}
+
+#[derive(Clone, Debug)]
+struct TempShardContinuation {
+    shard: u8,
+    dev: u64,
+    ino: u64,
+    cookie: i64,
+}
+
+#[derive(Clone, Copy, Debug, Default, Eq, PartialEq)]
+pub struct TempScan {
+    pub scanned: usize,
+    pub omitted: usize,
+    /// True when this page completed a full round-robin traversal.
+    pub complete: bool,
+}
+
+impl ArtifactStore {
+    /// Scans at most `scan_budget` entries and returns `result_limit` stale capabilities.
+    pub fn scan_stale_temps(
+        &self,
+        grace: Duration,
+        scan_budget: usize,
+        result_limit: usize,
+    ) -> Result<(TempScan, Vec<StaleTemp>), ArtifactError> {
+        self.scan_stale_temps_after(grace, None, scan_budget, result_limit)
+            .map(|(report, candidates, _)| (report, candidates))
+    }
+
+    /// Resumes stale-temp cleanup from an opaque round-robin marker. A
+    /// continuation is valid only for this exact pinned root.
+    pub fn scan_stale_temps_after(
+        &self,
+        grace: Duration,
+        continuation: Option<TempScanCursor>,
+        scan_budget: usize,
+        result_limit: usize,
+    ) -> Result<(TempScan, Vec<StaleTemp>, TempScanCursor), ArtifactError> {
+        let root = self.root()?;
+        let _lock = RootLock::shared(root)?;
+        ensure_root_unchanged(root)?;
+        let sha = match open_existing_dir(root.fd.as_raw_fd(), b"sha256") {
+            Ok(fd) => fd,
+            Err(ArtifactError::NotFound) => {
+                return Ok((
+                    TempScan {
+                        complete: true,
+                        ..TempScan::default()
+                    },
+                    Vec::new(),
+                    TempScanCursor {
+                        root_dev: root.dev,
+                        root_ino: root.ino,
+                        next_shard: 0,
+                        shard_continuation: None,
+                    },
+                ));
+            }
+            Err(error) => return Err(error),
+        };
+        let (start_shard, mut shard_continuation) = match continuation {
+            Some(cursor) if cursor.root_dev == root.dev && cursor.root_ino == root.ino => {
+                (cursor.next_shard, cursor.shard_continuation)
+            }
+            Some(_) => return Err(ArtifactError::UnsafeRoot),
+            None => (0, None),
+        };
+        let mut report = TempScan::default();
+        let mut remaining = scan_budget;
+        let mut result = Vec::new();
+        if remaining == 0 {
+            return Ok((
+                report,
+                result,
+                TempScanCursor {
+                    root_dev: root.dev,
+                    root_ino: root.ino,
+                    next_shard: start_shard,
+                    shard_continuation,
+                },
+            ));
+        }
+        let mut next_shard = start_shard;
+        for offset in 0_u16..=255 {
+            if remaining == 0 {
+                break;
+            }
+            let shard_number = start_shard.wrapping_add(offset as u8);
+            next_shard = shard_number.wrapping_add(1);
+            let shard = format!("{shard_number:02x}");
+            let shard_fd = match open_existing_dir(sha.as_raw_fd(), shard.as_bytes()) {
+                Ok(fd) => fd,
+                Err(ArtifactError::NotFound) => {
+                    shard_continuation = None;
+                    continue;
+                }
+                Err(error) => return Err(error),
+            };
+            let stat = stat_fd(shard_fd.as_raw_fd())?;
+            let resume = shard_continuation.take().and_then(|continuation| {
+                if continuation.shard == shard_number
+                    && continuation.dev == stat.st_dev
+                    && continuation.ino == stat.st_ino
+                {
+                    Some(continuation.cookie)
+                } else {
+                    None
+                }
+            });
+            let page = list_names_after(shard_fd.as_raw_fd(), &mut remaining, &mut report, resume)?;
+            for entry in page.names {
+                let name = entry.name;
+                if !valid_temp_name(&name) {
+                    continue;
+                }
+                #[cfg(debug_assertions)]
+                let _ = crate::test_support::checkpoint("housekeeping_before_stat");
+                let stat = match nofollow_stat(shard_fd.as_raw_fd(), &name) {
+                    Ok(stat) => stat,
+                    Err(ArtifactError::NotFound) => continue,
+                    Err(error) => return Err(error),
+                };
+                if (stat.st_mode & libc::S_IFMT) != libc::S_IFREG || stat.st_nlink != 1 {
+                    continue;
+                }
+                let modified = UNIX_EPOCH
+                    .checked_add(Duration::new(
+                        stat.st_mtime.max(0) as u64,
+                        stat.st_mtime_nsec.max(0) as u32,
+                    ))
+                    .unwrap_or(SystemTime::UNIX_EPOCH);
+                if SystemTime::now()
+                    .duration_since(modified)
+                    .is_ok_and(|age| age >= grace)
+                {
+                    if result.len() == result_limit {
+                        report.omitted += 1;
+                        if result_limit > 0 {
+                            return Ok((
+                                report,
+                                result,
+                                TempScanCursor {
+                                    root_dev: root.dev,
+                                    root_ino: root.ino,
+                                    next_shard: shard_number,
+                                    shard_continuation: Some(TempShardContinuation {
+                                        shard: shard_number,
+                                        dev: stat.st_dev,
+                                        ino: stat.st_ino,
+                                        cookie: entry.cookie_before,
+                                    }),
+                                },
+                            ));
+                        }
+                        continue;
+                    }
+                    result.push(StaleTemp {
+                        name,
+                        shard: shard.clone(),
+                        root_dev: root.dev,
+                        root_ino: root.ino,
+                        dev: stat.st_dev,
+                        ino: stat.st_ino,
+                        modified_seconds: stat.st_mtime,
+                        modified_nanoseconds: stat.st_mtime_nsec,
+                        grace,
+                    });
+                }
+            }
+            if let Some(cookie) = page.continuation {
+                return Ok((
+                    report,
+                    result,
+                    TempScanCursor {
+                        root_dev: root.dev,
+                        root_ino: root.ino,
+                        next_shard: shard_number,
+                        shard_continuation: Some(TempShardContinuation {
+                            shard: shard_number,
+                            dev: stat.st_dev,
+                            ino: stat.st_ino,
+                            cookie,
+                        }),
+                    },
+                ));
+            }
+        }
+        report.complete = true;
+        Ok((
+            report,
+            result,
+            TempScanCursor {
+                root_dev: root.dev,
+                root_ino: root.ino,
+                next_shard,
+                shard_continuation: None,
+            },
+        ))
+    }
+
+    /// Deletes a revalidated stale inode under an exclusive lock, then fsyncs.
+    pub fn delete_stale_temp(&self, candidate: StaleTemp) -> Result<(), ArtifactError> {
+        let root = self.root()?;
+        let _lock = RootLock::exclusive(root)?;
+        ensure_root_unchanged(root)?;
+        if candidate.root_dev != root.dev || candidate.root_ino != root.ino {
+            return Err(ArtifactError::UnsafeRoot);
+        }
+        if !valid_temp_name(&candidate.name)
+            || candidate.shard.len() != 2
+            || !candidate.shard.bytes().all(is_lower_hex)
+        {
+            return Err(ArtifactError::UnsafeRoot);
+        }
+        let sha = open_existing_dir(root.fd.as_raw_fd(), b"sha256")?;
+        let shard = open_existing_dir(sha.as_raw_fd(), candidate.shard.as_bytes())?;
+        let stat = nofollow_stat(shard.as_raw_fd(), &candidate.name)?;
+        if stat.st_dev != candidate.dev
+            || stat.st_ino != candidate.ino
+            || stat.st_mtime != candidate.modified_seconds
+            || stat.st_mtime_nsec != candidate.modified_nanoseconds
+            || (stat.st_mode & libc::S_IFMT) != libc::S_IFREG
+            || stat.st_nlink != 1
+        {
+            return Err(ArtifactError::UnsafeRoot);
+        }
+        let modified = UNIX_EPOCH
+            .checked_add(Duration::new(
+                stat.st_mtime.max(0) as u64,
+                stat.st_mtime_nsec.max(0) as u32,
+            ))
+            .unwrap_or(SystemTime::UNIX_EPOCH);
+        if !SystemTime::now()
+            .duration_since(modified)
+            .is_ok_and(|age| age >= candidate.grace)
+        {
+            return Err(ArtifactError::UnsafeRoot);
+        }
+        unlinkat(shard.as_raw_fd(), candidate.name.as_bytes())?;
+        fsync_fd(shard.as_raw_fd())
+    }
+}

crates/crank-artifacts/tests/artifact_store.rs

@@ -1,6 +1,5 @@
 use std::{
     fs,
-    io::{self, Read},
     os::unix::fs::PermissionsExt,
     path::PathBuf,
     process::{Child, Command, ExitStatus},
@@ -13,7 +12,7 @@ use std::{
 use crank_artifacts::test_support::{
     FaultAction, clear_checkpoint, set_checkpoint, wait_until_held,
 };
-use crank_artifacts::{ArtifactError, ArtifactRef, ArtifactStore, MAX_ARTIFACT_BYTES, StaleTemp};
+use crank_artifacts::{ArtifactError, ArtifactRef, ArtifactStore, MAX_ARTIFACT_BYTES};
 use sha2::{Digest, Sha256};
 #[cfg(debug_assertions)]
 use std::sync::{Mutex, OnceLock};
@@ -786,318 +785,3 @@ fn legacy_oversize_is_rejected() {
         ArtifactError::Integrity
     );
 }
-
-#[test]
-fn fifo_entries_fail_without_blocking_reads() {
-    let root = TestRoot::new("fifo");
-    let expected =
-        ArtifactRef::from_digest_hex(&format!("{:x}", Sha256::digest(b"fifo bytes"))).unwrap();
-    let sha = root.0.join("sha256");
-    let shard = sha.join(&expected.digest_hex()[..2]);
-    fs::create_dir(&sha).unwrap();
-    fs::create_dir(&shard).unwrap();
-    fs::set_permissions(&sha, fs::Permissions::from_mode(0o700)).unwrap();
-    fs::set_permissions(&shard, fs::Permissions::from_mode(0o700)).unwrap();
-    let blob_fifo = shard.join(expected.digest_hex());
-    let legacy = root.0.join("legacy");
-    fs::create_dir(&legacy).unwrap();
-    fs::set_permissions(&legacy, fs::Permissions::from_mode(0o700)).unwrap();
-    let legacy_fifo = legacy.join("source");
-    for path in [&blob_fifo, &legacy_fifo] {
-        let path = std::ffi::CString::new(path.as_os_str().as_encoded_bytes()).unwrap();
-        assert_eq!(unsafe { libc::mkfifo(path.as_ptr(), 0o400) }, 0);
-    }
-
-    let executable = std::env::current_exe().unwrap();
-    for kind in ["blob", "legacy"] {
-        let child = Command::new(&executable)
-            .args(["--exact", "fifo_read_child", "--nocapture"])
-            .env("CRANK_ARTIFACTS_FIFO_ROOT", &root.0)
-            .env("CRANK_ARTIFACTS_FIFO_KIND", kind)
-            .env("CRANK_ARTIFACTS_FIFO_REF", expected.as_str())
-            .spawn()
-            .unwrap();
-        assert!(wait_for_child(child, Duration::from_secs(2)).success());
-    }
-}
-
-#[test]
-fn fifo_read_child() {
-    let Some(root) = std::env::var_os("CRANK_ARTIFACTS_FIFO_ROOT") else {
-        return;
-    };
-    let root = PathBuf::from(root);
-    let kind = std::env::var("CRANK_ARTIFACTS_FIFO_KIND").unwrap();
-    let expected = ArtifactRef::parse(&std::env::var("CRANK_ARTIFACTS_FIFO_REF").unwrap()).unwrap();
-    let store = ArtifactStore::open(&root).unwrap();
-    let result = if kind == "blob" {
-        store.read(&expected)
-    } else {
-        store
-            .with_legacy_root(root.join("legacy"))
-            .read_legacy_file_url(
-                &format!("file://{}/legacy/source", root.display()),
-                &expected,
-            )
-    };
-    assert_eq!(result.unwrap_err(), ArtifactError::Integrity);
-}
-
-struct ShortReader {
-    bytes: Vec<u8>,
-    offset: usize,
-    interrupted: bool,
-}
-
-struct FailingReader;
-
-struct HousekeepingReader {
-    store: ArtifactStore,
-    candidate: Option<StaleTemp>,
-    bytes: io::Cursor<Vec<u8>>,
-}
-
-impl Read for FailingReader {
-    fn read(&mut self, _buf: &mut [u8]) -> io::Result<usize> {
-        Err(io::Error::other("private reader detail"))
-    }
-}
-impl Read for HousekeepingReader {
-    fn read(&mut self, buffer: &mut [u8]) -> io::Result<usize> {
-        if let Some(candidate) = self.candidate.take() {
-            self.store.delete_stale_temp(candidate).unwrap();
-        }
-        self.bytes.read(buffer)
-    }
-}
-impl Read for ShortReader {
-    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
-        if !self.interrupted {
-            self.interrupted = true;
-            return Err(io::Error::from(io::ErrorKind::Interrupted));
-        }
-        if self.offset == self.bytes.len() {
-            return Ok(0);
-        }
-        let amount = 1.min(buf.len()).min(self.bytes.len() - self.offset);
-        buf[..amount].copy_from_slice(&self.bytes[self.offset..self.offset + amount]);
-        self.offset += amount;
-        Ok(amount)
-    }
-}
-
-#[test]
-fn streaming_input_handles_short_reads_and_interruption() {
-    let root = TestRoot::new("stream");
-    let store = ArtifactStore::open(&root.0).unwrap();
-    let mut reader = ShortReader {
-        bytes: b"short chunk source".to_vec(),
-        offset: 0,
-        interrupted: false,
-    };
-    let stored = store.put_reader(&mut reader).unwrap();
-    assert_eq!(
-        store.read(&stored.artifact_ref).unwrap(),
-        b"short chunk source"
-    );
-
-    assert_eq!(
-        store.put_reader(&mut FailingReader).unwrap_err(),
-        ArtifactError::Storage
-    );
-    for error in [
-        ArtifactError::InvalidReference,
-        ArtifactError::Integrity,
-        ArtifactError::Storage,
-        ArtifactError::UnsafeRoot,
-    ] {
-        let diagnostic = format!("{error} {error:?}");
-        assert!(!diagnostic.contains("private reader detail"));
-        assert!(!diagnostic.contains(std::env::temp_dir().to_string_lossy().as_ref()));
-    }
-}
-
-#[test]
-fn user_reader_can_run_housekeeping_without_self_deadlock() {
-    use std::sync::mpsc;
-
-    let root = TestRoot::new("reader-housekeeping");
-    let store = ArtifactStore::open(&root.0).unwrap();
-    let stored = store.put(b"reader housekeeping shard").unwrap();
-    let temp = root
-        .0
-        .join("sha256")
-        .join(&stored.artifact_ref.digest_hex()[..2])
-        .join(".crank-artifact-tmp-v1-00000000000000000000000000000000-1-10");
-    fs::write(&temp, b"partial").unwrap();
-    let (_, candidates) = store.scan_stale_temps(Duration::ZERO, 128, 1).unwrap();
-    let mut reader = HousekeepingReader {
-        store: store.clone(),
-        candidate: candidates.into_iter().next(),
-        bytes: io::Cursor::new(b"reader bytes".to_vec()),
-    };
-    let writer_store = store.clone();
-    let (sender, receiver) = mpsc::channel();
-    thread::spawn(move || {
-        sender.send(writer_store.put_reader(&mut reader)).unwrap();
-    });
-
-    let stored = receiver
-        .recv_timeout(Duration::from_secs(2))
-        .expect("put_reader deadlocked while its reader ran housekeeping")
-        .unwrap();
-    assert_eq!(store.read(&stored.artifact_ref).unwrap(), b"reader bytes");
-    assert!(!temp.exists());
-}
-
-#[test]
-#[cfg(debug_assertions)]
-fn forked_crash_checkpoints_publish_only_complete_or_absent_blobs() {
-    let _guard = fault_guard();
-    if let Some(root) = std::env::var_os("CRANK_ARTIFACTS_CHILD_ROOT") {
-        set_checkpoint(
-            std::env::var("CRANK_ARTIFACTS_TEST_STAGE").unwrap(),
-            FaultAction::Exit,
-        );
-        let store = ArtifactStore::open(PathBuf::from(root)).unwrap();
-        let _ = store.put(b"crash-consistent bytes");
-        panic!("checkpoint did not terminate the child");
-    }
-
-    for stage in ["write", "file_fsync", "publish", "directory_fsync"] {
-        let root = TestRoot::new(stage);
-        let status = Command::new(std::env::current_exe().unwrap())
-            .args([
-                "--exact",
-                "forked_crash_checkpoints_publish_only_complete_or_absent_blobs",
-                "--nocapture",
-            ])
-            .env("CRANK_ARTIFACTS_CHILD_ROOT", &root.0)
-            .env("CRANK_ARTIFACTS_TEST_STAGE", stage)
-            .status()
-            .unwrap();
-        assert_eq!(status.code(), Some(86), "stage={stage}");
-
-        let store = ArtifactStore::open(&root.0).unwrap();
-        let expected = ArtifactRef::from_digest_hex(&format!(
-            "{:x}",
-            Sha256::digest(b"crash-consistent bytes")
-        ))
-        .unwrap();
-        match store.read(&expected) {
-            Ok(bytes) => assert_eq!(bytes, b"crash-consistent bytes", "stage={stage}"),
-            Err(ArtifactError::NotFound) => {}
-            Err(error) => panic!("stage={stage} exposed an invalid final blob: {error}"),
-        }
-        let stored = store.put(b"crash-consistent bytes").unwrap();
-        assert_eq!(stored.artifact_ref, expected, "stage={stage}");
-        assert_eq!(
-            store.read(&expected).unwrap(),
-            b"crash-consistent bytes",
-            "stage={stage}"
-        );
-
-        let (_, candidates) = store.scan_stale_temps(Duration::ZERO, 128, 16).unwrap();
-        for candidate in candidates {
-            store.delete_stale_temp(candidate).unwrap();
-        }
-    }
-}
-
-#[test]
-#[cfg(debug_assertions)]
-fn disk_full_fault_seams_leave_a_retryable_store() {
-    let _guard = fault_guard();
-    if let Some(root) = std::env::var_os("CRANK_ARTIFACTS_CHILD_ROOT") {
-        set_checkpoint(
-            std::env::var("CRANK_ARTIFACTS_TEST_STAGE").unwrap(),
-            FaultAction::Fail,
-        );
-        let store = ArtifactStore::open(PathBuf::from(root)).unwrap();
-        assert_eq!(
-            store.put(b"disk-full seam bytes").unwrap_err(),
-            ArtifactError::Storage
-        );
-        return;
-    }
-
-    for stage in ["write", "file_fsync", "publish", "directory_fsync"] {
-        let root = TestRoot::new(&format!("full-{stage}"));
-        let status = Command::new(std::env::current_exe().unwrap())
-            .args([
-                "--exact",
-                "disk_full_fault_seams_leave_a_retryable_store",
-                "--nocapture",
-            ])
-            .env("CRANK_ARTIFACTS_CHILD_ROOT", &root.0)
-            .env("CRANK_ARTIFACTS_TEST_STAGE", stage)
-            .status()
-            .unwrap();
-        assert!(status.success(), "stage={stage}");
-
-        let store = ArtifactStore::open(&root.0).unwrap();
-        let expected =
-            ArtifactRef::from_digest_hex(&format!("{:x}", Sha256::digest(b"disk-full seam bytes")))
-                .unwrap();
-        match store.read(&expected) {
-            Ok(bytes) => assert_eq!(bytes, b"disk-full seam bytes", "stage={stage}"),
-            Err(ArtifactError::NotFound) => {}
-            Err(error) => panic!("stage={stage} exposed an invalid final blob: {error}"),
-        }
-        assert_eq!(
-            store.put(b"disk-full seam bytes").unwrap().artifact_ref,
-            expected,
-            "stage={stage}"
-        );
-        assert_eq!(
-            store.read(&expected).unwrap(),
-            b"disk-full seam bytes",
-            "stage={stage}"
-        );
-    }
-}
-
-#[test]
-fn real_process_write_limit_leaves_no_partial_final_blob() {
-    if let Some(root) = std::env::var_os("CRANK_ARTIFACTS_CHILD_ROOT") {
-        // Keep the test's host filesystem untouched while making the actual
-        // write syscall fail as it would for a quota/disk-full condition.
-        unsafe {
-            assert_ne!(libc::signal(libc::SIGXFSZ, libc::SIG_IGN), libc::SIG_ERR);
-            let limit = libc::rlimit {
-                rlim_cur: 1,
-                rlim_max: 1,
-            };
-            assert_eq!(libc::setrlimit(libc::RLIMIT_FSIZE, &limit), 0);
-        }
-        let store = ArtifactStore::open(PathBuf::from(root)).unwrap();
-        assert_eq!(
-            store.put(b"real write limit bytes").unwrap_err(),
-            ArtifactError::Storage
-        );
-        return;
-    }
-
-    let root = TestRoot::new("real-write-limit");
-    let status = Command::new(std::env::current_exe().unwrap())
-        .args([
-            "--exact",
-            "real_process_write_limit_leaves_no_partial_final_blob",
-            "--nocapture",
-        ])
-        .env("CRANK_ARTIFACTS_CHILD_ROOT", &root.0)
-        .status()
-        .unwrap();
-    assert!(status.success());
-
-    let store = ArtifactStore::open(&root.0).unwrap();
-    let expected =
-        ArtifactRef::from_digest_hex(&format!("{:x}", Sha256::digest(b"real write limit bytes")))
-            .unwrap();
-    assert_eq!(store.read(&expected).unwrap_err(), ArtifactError::NotFound);
-    assert_eq!(
-        store.put(b"real write limit bytes").unwrap().artifact_ref,
-        expected
-    );
-    assert_eq!(store.read(&expected).unwrap(), b"real write limit bytes");
-}

crates/crank-artifacts/tests/artifact_store_io.rs

@@ -0,0 +1,379 @@
+use std::{
+    fs,
+    io::{self, Read},
+    os::unix::fs::PermissionsExt,
+    path::PathBuf,
+    process::{Child, Command, ExitStatus},
+    sync::atomic::{AtomicU64, Ordering},
+    thread,
+    time::Duration,
+};
+
+#[cfg(debug_assertions)]
+use crank_artifacts::test_support::{FaultAction, set_checkpoint};
+use crank_artifacts::{ArtifactError, ArtifactRef, ArtifactStore, StaleTemp};
+use sha2::{Digest, Sha256};
+#[cfg(debug_assertions)]
+use std::sync::{Mutex, OnceLock};
+
+static NEXT_ROOT: AtomicU64 = AtomicU64::new(0);
+
+#[cfg(debug_assertions)]
+fn fault_guard() -> std::sync::MutexGuard<'static, ()> {
+    static GUARD: OnceLock<Mutex<()>> = OnceLock::new();
+    GUARD.get_or_init(|| Mutex::new(())).lock().unwrap()
+}
+
+struct TestRoot(PathBuf);
+
+impl TestRoot {
+    fn new(name: &str) -> Self {
+        let path = std::env::temp_dir().join(format!(
+            "crank-artifacts-{name}-{}-{}",
+            std::process::id(),
+            NEXT_ROOT.fetch_add(1, Ordering::Relaxed)
+        ));
+        fs::create_dir(&path).unwrap();
+        fs::set_permissions(&path, fs::Permissions::from_mode(0o700)).unwrap();
+        Self(path)
+    }
+}
+
+impl Drop for TestRoot {
+    fn drop(&mut self) {
+        let _ = fs::set_permissions(&self.0, fs::Permissions::from_mode(0o700));
+        let _ = fs::remove_dir_all(&self.0);
+    }
+}
+
+fn wait_for_child(mut child: Child, timeout: Duration) -> ExitStatus {
+    let deadline = std::time::Instant::now() + timeout;
+    loop {
+        if let Some(status) = child.try_wait().unwrap() {
+            return status;
+        }
+        if std::time::Instant::now() >= deadline {
+            let _ = child.kill();
+            let _ = child.wait();
+            panic!("child process did not finish within {timeout:?}");
+        }
+        thread::sleep(Duration::from_millis(10));
+    }
+}
+
+#[test]
+fn fifo_entries_fail_without_blocking_reads() {
+    let root = TestRoot::new("fifo");
+    let expected =
+        ArtifactRef::from_digest_hex(&format!("{:x}", Sha256::digest(b"fifo bytes"))).unwrap();
+    let sha = root.0.join("sha256");
+    let shard = sha.join(&expected.digest_hex()[..2]);
+    fs::create_dir(&sha).unwrap();
+    fs::create_dir(&shard).unwrap();
+    fs::set_permissions(&sha, fs::Permissions::from_mode(0o700)).unwrap();
+    fs::set_permissions(&shard, fs::Permissions::from_mode(0o700)).unwrap();
+    let blob_fifo = shard.join(expected.digest_hex());
+    let legacy = root.0.join("legacy");
+    fs::create_dir(&legacy).unwrap();
+    fs::set_permissions(&legacy, fs::Permissions::from_mode(0o700)).unwrap();
+    let legacy_fifo = legacy.join("source");
+    for path in [&blob_fifo, &legacy_fifo] {
+        let path = std::ffi::CString::new(path.as_os_str().as_encoded_bytes()).unwrap();
+        assert_eq!(unsafe { libc::mkfifo(path.as_ptr(), 0o400) }, 0);
+    }
+
+    let executable = std::env::current_exe().unwrap();
+    for kind in ["blob", "legacy"] {
+        let child = Command::new(&executable)
+            .args(["--exact", "fifo_read_child", "--nocapture"])
+            .env("CRANK_ARTIFACTS_FIFO_ROOT", &root.0)
+            .env("CRANK_ARTIFACTS_FIFO_KIND", kind)
+            .env("CRANK_ARTIFACTS_FIFO_REF", expected.as_str())
+            .spawn()
+            .unwrap();
+        assert!(wait_for_child(child, Duration::from_secs(2)).success());
+    }
+}
+
+#[test]
+fn fifo_read_child() {
+    let Some(root) = std::env::var_os("CRANK_ARTIFACTS_FIFO_ROOT") else {
+        return;
+    };
+    let root = PathBuf::from(root);
+    let kind = std::env::var("CRANK_ARTIFACTS_FIFO_KIND").unwrap();
+    let expected = ArtifactRef::parse(&std::env::var("CRANK_ARTIFACTS_FIFO_REF").unwrap()).unwrap();
+    let store = ArtifactStore::open(&root).unwrap();
+    let result = if kind == "blob" {
+        store.read(&expected)
+    } else {
+        store
+            .with_legacy_root(root.join("legacy"))
+            .read_legacy_file_url(
+                &format!("file://{}/legacy/source", root.display()),
+                &expected,
+            )
+    };
+    assert_eq!(result.unwrap_err(), ArtifactError::Integrity);
+}
+
+struct ShortReader {
+    bytes: Vec<u8>,
+    offset: usize,
+    interrupted: bool,
+}
+
+struct FailingReader;
+
+struct HousekeepingReader {
+    store: ArtifactStore,
+    candidate: Option<StaleTemp>,
+    bytes: io::Cursor<Vec<u8>>,
+}
+
+impl Read for FailingReader {
+    fn read(&mut self, _buf: &mut [u8]) -> io::Result<usize> {
+        Err(io::Error::other("private reader detail"))
+    }
+}
+
+impl Read for HousekeepingReader {
+    fn read(&mut self, buffer: &mut [u8]) -> io::Result<usize> {
+        if let Some(candidate) = self.candidate.take() {
+            self.store.delete_stale_temp(candidate).unwrap();
+        }
+        self.bytes.read(buffer)
+    }
+}
+
+impl Read for ShortReader {
+    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+        if !self.interrupted {
+            self.interrupted = true;
+            return Err(io::Error::from(io::ErrorKind::Interrupted));
+        }
+        if self.offset == self.bytes.len() {
+            return Ok(0);
+        }
+        let amount = 1.min(buf.len()).min(self.bytes.len() - self.offset);
+        buf[..amount].copy_from_slice(&self.bytes[self.offset..self.offset + amount]);
+        self.offset += amount;
+        Ok(amount)
+    }
+}
+
+#[test]
+fn streaming_input_handles_short_reads_and_interruption() {
+    let root = TestRoot::new("stream");
+    let store = ArtifactStore::open(&root.0).unwrap();
+    let mut reader = ShortReader {
+        bytes: b"short chunk source".to_vec(),
+        offset: 0,
+        interrupted: false,
+    };
+    let stored = store.put_reader(&mut reader).unwrap();
+    assert_eq!(
+        store.read(&stored.artifact_ref).unwrap(),
+        b"short chunk source"
+    );
+
+    assert_eq!(
+        store.put_reader(&mut FailingReader).unwrap_err(),
+        ArtifactError::Storage
+    );
+    for error in [
+        ArtifactError::InvalidReference,
+        ArtifactError::Integrity,
+        ArtifactError::Storage,
+        ArtifactError::UnsafeRoot,
+    ] {
+        let diagnostic = format!("{error} {error:?}");
+        assert!(!diagnostic.contains("private reader detail"));
+        assert!(!diagnostic.contains(std::env::temp_dir().to_string_lossy().as_ref()));
+    }
+}
+
+#[test]
+fn user_reader_can_run_housekeeping_without_self_deadlock() {
+    use std::sync::mpsc;
+
+    let root = TestRoot::new("reader-housekeeping");
+    let store = ArtifactStore::open(&root.0).unwrap();
+    let stored = store.put(b"reader housekeeping shard").unwrap();
+    let temp = root
+        .0
+        .join("sha256")
+        .join(&stored.artifact_ref.digest_hex()[..2])
+        .join(".crank-artifact-tmp-v1-00000000000000000000000000000000-1-10");
+    fs::write(&temp, b"partial").unwrap();
+    let (_, candidates) = store.scan_stale_temps(Duration::ZERO, 128, 1).unwrap();
+    let mut reader = HousekeepingReader {
+        store: store.clone(),
+        candidate: candidates.into_iter().next(),
+        bytes: io::Cursor::new(b"reader bytes".to_vec()),
+    };
+    let writer_store = store.clone();
+    let (sender, receiver) = mpsc::channel();
+    thread::spawn(move || {
+        sender.send(writer_store.put_reader(&mut reader)).unwrap();
+    });
+
+    let stored = receiver
+        .recv_timeout(Duration::from_secs(2))
+        .expect("put_reader deadlocked while its reader ran housekeeping")
+        .unwrap();
+    assert_eq!(store.read(&stored.artifact_ref).unwrap(), b"reader bytes");
+    assert!(!temp.exists());
+}
+
+#[test]
+#[cfg(debug_assertions)]
+fn forked_crash_checkpoints_publish_only_complete_or_absent_blobs() {
+    let _guard = fault_guard();
+    if let Some(root) = std::env::var_os("CRANK_ARTIFACTS_CHILD_ROOT") {
+        set_checkpoint(
+            std::env::var("CRANK_ARTIFACTS_TEST_STAGE").unwrap(),
+            FaultAction::Exit,
+        );
+        let store = ArtifactStore::open(PathBuf::from(root)).unwrap();
+        let _ = store.put(b"crash-consistent bytes");
+        panic!("checkpoint did not terminate the child");
+    }
+
+    for stage in ["write", "file_fsync", "publish", "directory_fsync"] {
+        let root = TestRoot::new(stage);
+        let status = Command::new(std::env::current_exe().unwrap())
+            .args([
+                "--exact",
+                "forked_crash_checkpoints_publish_only_complete_or_absent_blobs",
+                "--nocapture",
+            ])
+            .env("CRANK_ARTIFACTS_CHILD_ROOT", &root.0)
+            .env("CRANK_ARTIFACTS_TEST_STAGE", stage)
+            .status()
+            .unwrap();
+        assert_eq!(status.code(), Some(86), "stage={stage}");
+
+        let store = ArtifactStore::open(&root.0).unwrap();
+        let expected = ArtifactRef::from_digest_hex(&format!(
+            "{:x}",
+            Sha256::digest(b"crash-consistent bytes")
+        ))
+        .unwrap();
+        match store.read(&expected) {
+            Ok(bytes) => assert_eq!(bytes, b"crash-consistent bytes", "stage={stage}"),
+            Err(ArtifactError::NotFound) => {}
+            Err(error) => panic!("stage={stage} exposed an invalid final blob: {error}"),
+        }
+        let stored = store.put(b"crash-consistent bytes").unwrap();
+        assert_eq!(stored.artifact_ref, expected, "stage={stage}");
+        assert_eq!(
+            store.read(&expected).unwrap(),
+            b"crash-consistent bytes",
+            "stage={stage}"
+        );
+
+        let (_, candidates) = store.scan_stale_temps(Duration::ZERO, 128, 16).unwrap();
+        for candidate in candidates {
+            store.delete_stale_temp(candidate).unwrap();
+        }
+    }
+}
+
+#[test]
+#[cfg(debug_assertions)]
+fn disk_full_fault_seams_leave_a_retryable_store() {
+    let _guard = fault_guard();
+    if let Some(root) = std::env::var_os("CRANK_ARTIFACTS_CHILD_ROOT") {
+        set_checkpoint(
+            std::env::var("CRANK_ARTIFACTS_TEST_STAGE").unwrap(),
+            FaultAction::Fail,
+        );
+        let store = ArtifactStore::open(PathBuf::from(root)).unwrap();
+        assert_eq!(
+            store.put(b"disk-full seam bytes").unwrap_err(),
+            ArtifactError::Storage
+        );
+        return;
+    }
+
+    for stage in ["write", "file_fsync", "publish", "directory_fsync"] {
+        let root = TestRoot::new(&format!("full-{stage}"));
+        let status = Command::new(std::env::current_exe().unwrap())
+            .args([
+                "--exact",
+                "disk_full_fault_seams_leave_a_retryable_store",
+                "--nocapture",
+            ])
+            .env("CRANK_ARTIFACTS_CHILD_ROOT", &root.0)
+            .env("CRANK_ARTIFACTS_TEST_STAGE", stage)
+            .status()
+            .unwrap();
+        assert!(status.success(), "stage={stage}");
+
+        let store = ArtifactStore::open(&root.0).unwrap();
+        let expected =
+            ArtifactRef::from_digest_hex(&format!("{:x}", Sha256::digest(b"disk-full seam bytes")))
+                .unwrap();
+        match store.read(&expected) {
+            Ok(bytes) => assert_eq!(bytes, b"disk-full seam bytes", "stage={stage}"),
+            Err(ArtifactError::NotFound) => {}
+            Err(error) => panic!("stage={stage} exposed an invalid final blob: {error}"),
+        }
+        assert_eq!(
+            store.put(b"disk-full seam bytes").unwrap().artifact_ref,
+            expected,
+            "stage={stage}"
+        );
+        assert_eq!(
+            store.read(&expected).unwrap(),
+            b"disk-full seam bytes",
+            "stage={stage}"
+        );
+    }
+}
+
+#[test]
+fn real_process_write_limit_leaves_no_partial_final_blob() {
+    if let Some(root) = std::env::var_os("CRANK_ARTIFACTS_CHILD_ROOT") {
+        // Keep the test's host filesystem untouched while making the actual
+        // write syscall fail as it would for a quota/disk-full condition.
+        unsafe {
+            assert_ne!(libc::signal(libc::SIGXFSZ, libc::SIG_IGN), libc::SIG_ERR);
+            let limit = libc::rlimit {
+                rlim_cur: 1,
+                rlim_max: 1,
+            };
+            assert_eq!(libc::setrlimit(libc::RLIMIT_FSIZE, &limit), 0);
+        }
+        let store = ArtifactStore::open(PathBuf::from(root)).unwrap();
+        assert_eq!(
+            store.put(b"real write limit bytes").unwrap_err(),
+            ArtifactError::Storage
+        );
+        return;
+    }
+
+    let root = TestRoot::new("real-write-limit");
+    let status = Command::new(std::env::current_exe().unwrap())
+        .args([
+            "--exact",
+            "real_process_write_limit_leaves_no_partial_final_blob",
+            "--nocapture",
+        ])
+        .env("CRANK_ARTIFACTS_CHILD_ROOT", &root.0)
+        .status()
+        .unwrap();
+    assert!(status.success());
+
+    let store = ArtifactStore::open(&root.0).unwrap();
+    let expected =
+        ArtifactRef::from_digest_hex(&format!("{:x}", Sha256::digest(b"real write limit bytes")))
+            .unwrap();
+    assert_eq!(store.read(&expected).unwrap_err(), ArtifactError::NotFound);
+    assert_eq!(
+        store.put(b"real write limit bytes").unwrap().artifact_ref,
+        expected
+    );
+    assert_eq!(store.read(&expected).unwrap(), b"real write limit bytes");
+}

crates/crank-registry/src/migrations.rs

@@ -11,6 +11,7 @@ mod onboarding_product_events_v11;
 mod owned_relations;
 mod platform_key_name_reuse_v6;
 mod schema_guard;
+mod schema_guard_legacy_v1;
 mod schema_guard_v10;
 mod schema_guard_v11;
 mod schema_guard_v12;

crates/crank-registry/src/migrations/authority.rs

@@ -11,6 +11,9 @@ use super::platform_key_name_reuse_v6;
 use super::schema_guard::{
     OWNED_RELATIONS, relation_exists, validate_required_relations, validate_schema_fingerprint,
 };
+use super::schema_guard_legacy_v1::{
+    validate_ledgerless_baseline_fingerprint, validate_ledgerless_optional_fingerprints,
+};
 use super::{BASELINE_CHECKSUM, BASELINE_VERSION, apply_baseline};
 use crate::ext::ExtensionMigration;
 use sqlx::{PgConnection, PgPool, Row, Transaction, query};
@@ -389,6 +392,19 @@ impl MigrationAuthority {
         if from < 13 {
             artifact_cleanup_indexes_v13::apply(&mut transaction, &Self::sequence()[12]).await?;
         }
+        match inspect(&mut transaction).await? {
+            MigrationPreflight::Current {
+                version: CURRENT_VERSION,
+            } => {}
+            _ => {
+                return Err(MigrationError::new(
+                    "apply_failed",
+                    "apply.postflight",
+                    Some(CURRENT_VERSION),
+                    "restore_known_good_backup",
+                ));
+            }
+        }
         transaction
             .commit()
             .await
@@ -408,9 +424,15 @@ async fn inspect(connection: &mut PgConnection) -> Result<MigrationPreflight, Mi
             owned_exists |= relation_exists(connection, relation).await?;
         }
         if owned_exists {
+            if is_ledgerless_legacy_baseline(connection, canonical_exists).await? {
+                return Ok(MigrationPreflight::MigrationRequired {
+                    current: 0,
+                    target: CURRENT_VERSION,
+                });
+            }
             return Err(MigrationError::new(
                 "partial_sequence",
-                "preflight.core",
+                "preflight.core_missing",
                 None,
                 "restore_known_good_backup",
             ));
@@ -541,6 +563,47 @@ async fn inspect(connection: &mut PgConnection) -> Result<MigrationPreflight, Mi
         Ok(MigrationPreflight::Current { version: current })
     }
 }
+async fn is_ledgerless_legacy_baseline(
+    connection: &mut PgConnection,
+    canonical_exists: bool,
+) -> Result<bool, MigrationError> {
+    if canonical_exists {
+        return Ok(false);
+    }
+    for relation in [
+        "__crank_migration_legacy_audit",
+        "master_key_identities",
+        "master_key_rotations",
+        "admin_bootstrap_contracts",
+        "admin_login_backoff",
+        "admin_security_audit_events",
+        "product_events",
+        "product_event_daily_rollups",
+        "onboarding_selections",
+        "artifact_blobs",
+        "artifact_sources",
+    ] {
+        if relation_exists(connection, relation).await? {
+            return Ok(false);
+        }
+    }
+    for relation in owned_relations::BASELINE {
+        if !relation_exists(connection, relation).await? {
+            return Ok(false);
+        }
+    }
+    validate_required_relations(connection, owned_relations::BASELINE, 1).await?;
+    validate_ledgerless_baseline_fingerprint(connection).await?;
+    let mcp_ledger = relation_exists(connection, "__crank_mcp_migrations").await?;
+    if mcp_ledger {
+        return Ok(false);
+    }
+    let mcp_sessions = relation_exists(connection, "mcp_transport_sessions").await?;
+    let extension_ledger = relation_exists(connection, "__crank_ext_migrations").await?;
+    validate_ledgerless_optional_fingerprints(connection, mcp_sessions, extension_ledger).await?;
+    inspect_optional_legacy_for_ledgerless_baseline(connection).await?;
+    Ok(true)
+}
 async fn validate_core_ledger(connection: &mut PgConnection) -> Result<(), MigrationError> {
     let rows = query("select version, description, checksum from __crank_core_migrations order by version limit 2")
         .fetch_all(connection)
@@ -549,7 +612,7 @@ async fn validate_core_ledger(connection: &mut PgConnection) -> Result<(), Migra
     if rows.len() != 1 {
         return Err(MigrationError::new(
             "partial_sequence",
-            "preflight.core",
+            "preflight.core_cardinality",
             None,
             "restore_known_good_backup",
         ));
@@ -577,9 +640,21 @@ async fn validate_core_ledger(connection: &mut PgConnection) -> Result<(), Migra
     Ok(())
 }
 async fn inspect_optional_legacy(connection: &mut PgConnection) -> Result<(), MigrationError> {
+    inspect_optional_legacy_with_policy(connection, false).await
+}
+async fn inspect_optional_legacy_for_ledgerless_baseline(
+    connection: &mut PgConnection,
+) -> Result<(), MigrationError> {
+    inspect_optional_legacy_with_policy(connection, true).await
+}
+async fn inspect_optional_legacy_with_policy(
+    connection: &mut PgConnection,
+    allow_sessions_without_ledger: bool,
+) -> Result<(), MigrationError> {
     let mcp_ledger = relation_exists(connection, "__crank_mcp_migrations").await?;
     let mcp_sessions = relation_exists(connection, "mcp_transport_sessions").await?;
-    if mcp_ledger != mcp_sessions {
+    if mcp_ledger != mcp_sessions && !(allow_sessions_without_ledger && !mcp_ledger && mcp_sessions)
+    {
         return Err(MigrationError::new(
             "legacy_conflict",
             "preflight.legacy_mcp",

crates/crank-registry/src/migrations/authority_tests.rs

@@ -8,7 +8,7 @@ fn sequence_is_deterministic_and_append_only() {
     MigrationAuthority::validate_sequence().unwrap();
     assert_eq!(
         first.iter().map(|item| item.version).collect::<Vec<_>>(),
-        vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]
+        vec![1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13]
     );
     assert_eq!(first[0].checksum, "crank-community-baseline-v1");
     assert_eq!(first[0].source_digest, BASELINE_SOURCE_SHA256);

crates/crank-registry/src/migrations/schema_guard.rs

@@ -325,6 +325,15 @@ pub(super) async fn validate_schema_fingerprint(
             .iter()
             .filter_map(|row| row.try_get::<String, _>("column_name").ok())
             .collect::<Vec<_>>();
+        let required = required
+            .iter()
+            .copied()
+            .filter(|column| {
+                !(current_version == 1
+                    && *table == "__crank_ext_migrations"
+                    && *column == "checksum")
+            })
+            .collect::<Vec<_>>();
         if actual.len() != required.len()
             || required
                 .iter()
@@ -337,6 +346,9 @@ pub(super) async fn validate_schema_fingerprint(
         if !relation_exists(connection, table).await? {
             continue;
         }
+        if current_version == 1 && *table == "__crank_ext_migrations" && *column == "checksum" {
+            continue;
+        }
         let row = query(
             "select data_type, is_nullable from information_schema.columns
              where table_schema = current_schema() and table_name = $1 and column_name = $2",
@@ -424,23 +436,25 @@ pub(super) async fn validate_schema_fingerprint(
             return Err(schema_error(current_version));
         }
     }
-    let required_indexes = [
+    if relation_exists(connection, "mcp_transport_sessions").await? {
-        "mcp_transport_sessions_workspace_agent_idx",
+        let required_indexes = [
-        "mcp_transport_sessions_expires_at_idx",
+            "mcp_transport_sessions_workspace_agent_idx",
-    ];
+            "mcp_transport_sessions_expires_at_idx",
-    for index in required_indexes {
+        ];
-        let present = query(
+        for index in required_indexes {
-            "select exists (select 1 from pg_catalog.pg_indexes
+            let present = query(
-             where schemaname = current_schema() and indexname = $1) as present",
+                "select exists (select 1 from pg_catalog.pg_indexes
-        )
+                 where schemaname = current_schema() and indexname = $1) as present",
-        .bind(index)
+            )
-        .fetch_one(&mut *connection)
+            .bind(index)
-        .await
+            .fetch_one(&mut *connection)
-        .map_err(|_| MigrationError::storage("preflight.schema"))?
+            .await
-        .try_get::<bool, _>("present")
+            .map_err(|_| MigrationError::storage("preflight.schema"))?
-        .map_err(|_| MigrationError::storage("preflight.schema"))?;
+            .try_get::<bool, _>("present")
-        if !present {
+            .map_err(|_| MigrationError::storage("preflight.schema"))?;
-            return Err(schema_error(current_version));
+            if !present {
+                return Err(schema_error(current_version));
+            }
         }
     }
     if current_version >= 3 {

crates/crank-registry/src/migrations/schema_guard_legacy_v1.rs

@@ -0,0 +1,196 @@
+use sha2::{Digest, Sha256};
+use sqlx::PgConnection;
+
+use super::authority::MigrationError;
+
+// Exact PostgreSQL 16 catalog contract produced by the last published
+// pre-ledger Community schema (commit 8318e4b).
+const LEDGERLESS_BASELINE_FINGERPRINT_SHA256: &str =
+    "36624ca42a28c1388f9c50e6d6c489a8e5ca1f73441a894d9af89e5e115b542c";
+const LEDGERLESS_MCP_FINGERPRINTS_SHA256: &[&str] = &[
+    // Initial published session table.
+    "9240c3d85dbc9eeddb1cd99661ec8f7d8d6ece0e62ec486151dcc951a7f3c81c",
+    // Published session table after supports_elicitation was added.
+    "ac9f99a7e667552a07480d5d45379dd0a8b529b00ba98855ce6725535198249f",
+];
+const LEDGERLESS_EXTENSION_FINGERPRINT_SHA256: &str =
+    "0809a80c0bb6e80f68d0557006c4b62f2e63556f61f6fdedc5abc324950e2587";
+
+const BASELINE_RELATIONS: &[&str] = &[
+    "workspaces",
+    "users",
+    "memberships",
+    "user_sessions",
+    "invitation_tokens",
+    "platform_api_keys",
+    "operations",
+    "operation_versions",
+    "published_operations",
+    "operation_samples",
+    "descriptors",
+    "agents",
+    "agent_versions",
+    "published_agents",
+    "agent_operation_bindings",
+    "secrets",
+    "secret_versions",
+    "auth_profiles",
+    "workspace_upstreams",
+    "yaml_import_jobs",
+    "import_jobs",
+    "approval_requests",
+    "invocation_logs",
+    "usage_rollups",
+];
+
+pub(super) async fn validate_ledgerless_baseline_fingerprint(
+    connection: &mut PgConnection,
+) -> Result<(), MigrationError> {
+    let actual = catalog_fingerprint(connection, BASELINE_RELATIONS).await?;
+    if actual == LEDGERLESS_BASELINE_FINGERPRINT_SHA256 {
+        Ok(())
+    } else {
+        Err(fingerprint_error())
+    }
+}
+
+pub(super) async fn validate_ledgerless_optional_fingerprints(
+    connection: &mut PgConnection,
+    has_mcp_sessions: bool,
+    has_extension_ledger: bool,
+) -> Result<(), MigrationError> {
+    if has_mcp_sessions {
+        let actual = catalog_fingerprint(connection, &["mcp_transport_sessions"]).await?;
+        if !LEDGERLESS_MCP_FINGERPRINTS_SHA256.contains(&actual.as_str()) {
+            return Err(fingerprint_error());
+        }
+    }
+    if has_extension_ledger {
+        let actual = catalog_fingerprint(connection, &["__crank_ext_migrations"]).await?;
+        if actual != LEDGERLESS_EXTENSION_FINGERPRINT_SHA256 {
+            return Err(fingerprint_error());
+        }
+    }
+    Ok(())
+}
+
+async fn catalog_fingerprint(
+    connection: &mut PgConnection,
+    relations: &[&str],
+) -> Result<String, MigrationError> {
+    let relations = relations
+        .iter()
+        .map(|value| (*value).to_owned())
+        .collect::<Vec<_>>();
+    let unsafe_catalog_state = sqlx::query_scalar::<_, bool>(
+        "with selected(table_name) as (select unnest($1::text[]))
+         select
+             exists (
+                 select 1 from pg_catalog.pg_class c
+                 join pg_catalog.pg_namespace n on n.oid = c.relnamespace
+                 join selected s on s.table_name = c.relname
+                 where n.nspname = current_schema()
+                   and (c.relpersistence <> 'p' or c.relrowsecurity
+                        or c.relforcerowsecurity or c.relreplident <> 'd')
+             )
+             or exists (
+                 select 1 from pg_catalog.pg_index i
+                 join pg_catalog.pg_class t on t.oid = i.indrelid
+                 join pg_catalog.pg_namespace n on n.oid = t.relnamespace
+                 join selected s on s.table_name = t.relname
+                 where n.nspname = current_schema()
+                   and (not i.indisvalid or not i.indisready or not i.indislive)
+             )
+             or exists (
+                 select 1 from pg_catalog.pg_policy p
+                 join pg_catalog.pg_class t on t.oid = p.polrelid
+                 join pg_catalog.pg_namespace n on n.oid = t.relnamespace
+                 join selected s on s.table_name = t.relname
+                 where n.nspname = current_schema()
+             )
+             or exists (
+                 select 1 from pg_catalog.pg_trigger tg
+                 join pg_catalog.pg_class t on t.oid = tg.tgrelid
+                 join pg_catalog.pg_namespace n on n.oid = t.relnamespace
+                 join selected s on s.table_name = t.relname
+                 where n.nspname = current_schema() and tg.tgenabled <> 'O'
+             )",
+    )
+    .bind(relations.clone())
+    .fetch_one(&mut *connection)
+    .await
+    .map_err(|_| MigrationError::storage("preflight.legacy_fingerprint"))?;
+    if unsafe_catalog_state {
+        return Err(fingerprint_error());
+    }
+    let fingerprint = sqlx::query_scalar::<_, String>(
+        "with baseline(table_name) as (select unnest($1::text[])), relation_rows as (
+            select jsonb_build_array('relation', c.relname, c.relkind::text) item
+            from pg_catalog.pg_class c
+            join pg_catalog.pg_namespace n on n.oid = c.relnamespace
+            join baseline b on b.table_name = c.relname
+            where n.nspname = current_schema()
+         ), column_rows as (
+            select jsonb_build_array(
+                'column', c.table_name, c.column_name,
+                c.data_type, c.udt_name, c.is_nullable, coalesce(c.column_default, '')
+            ) item
+            from information_schema.columns c
+            join baseline b using (table_name)
+            where c.table_schema = current_schema()
+         ), constraint_rows as (
+            select jsonb_build_array(
+                'constraint', t.relname, c.conname, c.contype::text,
+                c.convalidated, pg_get_constraintdef(c.oid, true)
+            ) item
+            from pg_catalog.pg_constraint c
+            join pg_catalog.pg_class t on t.oid = c.conrelid
+            join pg_catalog.pg_namespace n on n.oid = t.relnamespace
+            join baseline b on b.table_name = t.relname
+            where n.nspname = current_schema()
+         ), index_rows as (
+            select jsonb_build_array(
+                'index', t.relname, idx.relname,
+                replace(pg_get_indexdef(i.indexrelid), format('%I.', current_schema()), '')
+            ) item
+            from pg_catalog.pg_index i
+            join pg_catalog.pg_class idx on idx.oid = i.indexrelid
+            join pg_catalog.pg_class t on t.oid = i.indrelid
+            join pg_catalog.pg_namespace n on n.oid = t.relnamespace
+            join baseline b on b.table_name = t.relname
+            where n.nspname = current_schema()
+         ), trigger_rows as (
+            select jsonb_build_array(
+                'trigger', t.relname, tg.tgname,
+                replace(pg_get_triggerdef(tg.oid, true), format('%I.', current_schema()), '')
+            ) item
+            from pg_catalog.pg_trigger tg
+            join pg_catalog.pg_class t on t.oid = tg.tgrelid
+            join pg_catalog.pg_namespace n on n.oid = t.relnamespace
+            join baseline b on b.table_name = t.relname
+            where n.nspname = current_schema() and not tg.tgisinternal
+         ), all_rows as (
+            select item from relation_rows
+            union all select item from column_rows
+            union all select item from constraint_rows
+            union all select item from index_rows
+            union all select item from trigger_rows
+         )
+         select coalesce(jsonb_agg(item order by item::text), '[]'::jsonb)::text
+         from all_rows",
+    )
+    .bind(relations)
+    .fetch_one(connection)
+    .await
+    .map_err(|_| MigrationError::storage("preflight.legacy_fingerprint"))?;
+    Ok(format!("{:x}", Sha256::digest(fingerprint.as_bytes())))
+}
+
+fn fingerprint_error() -> MigrationError {
+    MigrationError::new(
+        "partial_sequence",
+        "preflight.legacy_fingerprint",
+        Some(1),
+        "restore_known_good_backup",
+    )
+}

crates/crank-registry/src/postgres/import_job.rs

@@ -211,7 +211,7 @@ impl PostgresRegistry {
             let payload = row.try_get::<Value, _>("preview_payload")?;
             // The job itself is expired regardless of whether a legacy or
             // corrupt payload can be decoded. Do not let one bad row roll
-            // back cleanup for every tenant.
+            // back cleanup for every workspace.
             for source_id in cleanup_source_ids(&payload) {
                 if detach_source_in_transaction(&mut transaction, &workspace_id, &source_id, now)
                     .await?

crates/crank-registry/tests/integration/migrations.rs

@@ -2,6 +2,7 @@ use crank_registry::{MigrationAuthority, MigrationPreflight, PostgresRegistry};
 use sqlx::Row;
 
 mod artifact_metadata;
+mod legacy_adoption;
 mod rollback;
 
 static EVENT_TRIGGER_TEST_LOCK: tokio::sync::Mutex<()> = tokio::sync::Mutex::const_new(());
@@ -153,7 +154,7 @@ async fn changed_checksum_fails_closed_without_repair() {
     );
 }
 #[tokio::test]
-async fn legacy_core_baseline_is_consolidated_without_data_loss() {
+async fn ledgerless_legacy_baseline_is_consolidated_without_data_loss() {
     let database_url = crank_test_support::postgres_schema_url("test_legacy_core").await;
     let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
     MigrationAuthority::apply(&pool).await.unwrap();
@@ -221,7 +222,8 @@ async fn legacy_core_baseline_is_consolidated_without_data_loss() {
     }
     sqlx::query(
         "drop table __crank_migrations, __crank_migration_legacy_audit,
-         __crank_mcp_migrations, mcp_transport_sessions, __crank_ext_migrations",
+         __crank_mcp_migrations, mcp_transport_sessions, __crank_ext_migrations,
+         __crank_core_migrations",
     )
     .execute(&pool)
     .await
@@ -229,7 +231,7 @@ async fn legacy_core_baseline_is_consolidated_without_data_loss() {
     assert_eq!(
         MigrationAuthority::preflight(&pool).await.unwrap(),
         MigrationPreflight::MigrationRequired {
-            current: 1,
+            current: 0,
             target: 13,
         }
     );
@@ -937,6 +939,20 @@ async fn any_owned_relation_without_core_ledger_is_partial() {
         .unwrap();
     let error = MigrationAuthority::preflight(&pool).await.unwrap_err();
     assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.core_missing");
+}
+#[tokio::test]
+async fn empty_core_ledger_is_reported_separately() {
+    let database_url = crank_test_support::postgres_schema_url("test_empty_core_ledger").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    MigrationAuthority::apply(&pool).await.unwrap();
+    sqlx::query("delete from __crank_core_migrations")
+        .execute(&pool)
+        .await
+        .unwrap();
+    let error = MigrationAuthority::preflight(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.core_cardinality");
 }
 #[tokio::test]
 async fn current_ledger_with_structural_drift_fails_closed() {

crates/crank-registry/tests/integration/migrations/legacy_adoption.rs

@@ -0,0 +1,439 @@
+use super::*;
+
+async fn ledgerless_v1(pool: &sqlx::PgPool) {
+    MigrationAuthority::apply(pool).await.unwrap();
+    remove_v4_schema(pool).await;
+    remove_v3_schema(pool).await;
+    sqlx::raw_sql(
+        "drop table __crank_migrations, __crank_migration_legacy_audit;
+         drop table __crank_mcp_migrations;
+         drop index mcp_transport_sessions_expires_at_idx;
+         drop table __crank_ext_migrations;
+         create table __crank_ext_migrations (
+             extension_name text not null,
+             version integer not null,
+             applied_at timestamptz not null default now(),
+             primary key (extension_name, version)
+         );
+         drop table __crank_core_migrations;",
+    )
+    .execute(pool)
+    .await
+    .unwrap();
+}
+
+#[tokio::test]
+async fn ledgerless_baseline_with_optional_index_drift_is_rejected_without_writes() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_index").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::raw_sql(
+        "drop index mcp_transport_sessions_workspace_agent_idx;
+         create index mcp_transport_sessions_workspace_agent_idx
+             on mcp_transport_sessions(id);",
+    )
+    .execute(&pool)
+    .await
+    .unwrap();
+
+    let error = MigrationAuthority::apply(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.legacy_fingerprint");
+    let core_exists: bool = sqlx::query_scalar(
+        "select to_regclass(format('%I.%I', current_schema(), '__crank_core_migrations')) is not null",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(!core_exists, "rejected adoption must remain read-only");
+}
+
+#[tokio::test]
+async fn ledgerless_baseline_with_disabled_integrity_triggers_is_rejected_without_writes() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_triggers").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::query("alter table memberships disable trigger all")
+        .execute(&pool)
+        .await
+        .unwrap();
+
+    let error = MigrationAuthority::apply(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.legacy_fingerprint");
+    let core_exists: bool = sqlx::query_scalar(
+        "select to_regclass(format('%I.%I', current_schema(), '__crank_core_migrations')) is not null",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(!core_exists, "rejected adoption must remain read-only");
+}
+
+#[tokio::test]
+async fn published_ledgerless_baseline_upgrades_without_data_loss() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_baseline").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::query(
+        "insert into operations
+            (id, workspace_id, name, display_name, protocol, status, created_at, updated_at)
+         values ('op_ledgerless', 'ws_default', 'ledgerless', 'Ledgerless', 'rest', 'draft', now(), now())",
+    )
+    .execute(&pool)
+    .await
+    .unwrap();
+
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::MigrationRequired {
+            current: 0,
+            target: 13,
+        }
+    );
+    MigrationAuthority::apply(&pool).await.unwrap();
+
+    let operation_count: i64 =
+        sqlx::query_scalar("select count(*) from operations where id = 'op_ledgerless'")
+            .fetch_one(&pool)
+            .await
+            .unwrap();
+    assert_eq!(operation_count, 1);
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::Current { version: 13 }
+    );
+}
+
+#[tokio::test]
+async fn published_ledgerless_historical_column_layout_upgrades_without_data_loss() {
+    let database_url =
+        crank_test_support::postgres_schema_url("test_ledgerless_historical_layout").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::raw_sql(
+        "alter table platform_api_keys
+             drop column key_kind,
+             drop column expires_at,
+             drop column allowed_origins_json;
+         alter table platform_api_keys
+             add column key_kind text not null default 'mcp_client',
+             add column expires_at timestamptz null,
+             add column allowed_origins_json jsonb not null default '[]'::jsonb;
+
+         drop index approval_requests_pending_fingerprint_idx;
+         alter table approval_requests
+             drop column execution_started_at,
+             drop column execution_attempts,
+             drop column request_fingerprint,
+             add column confirmation_title text not null default '',
+             add column confirmation_body text not null default '';
+         alter table approval_requests
+             alter column confirmation_title drop default,
+             alter column confirmation_body drop default,
+             drop column confirmation_title,
+             drop column confirmation_body,
+             add column execution_started_at timestamptz null,
+             add column execution_attempts integer not null default 0,
+             add column request_fingerprint text null;
+         create unique index approval_requests_pending_fingerprint_idx
+             on approval_requests(agent_id, operation_id, operation_version, request_fingerprint)
+             where status = 'pending' and request_fingerprint is not null;
+
+         insert into operations
+            (id, workspace_id, name, display_name, protocol, status, created_at, updated_at)
+         values ('op_historical_layout', 'ws_default', 'historical-layout', 'Historical layout', 'rest', 'draft', now(), now());
+         insert into agents
+            (id, workspace_id, slug, display_name, description, status, created_at, updated_at)
+         values ('agent_historical_layout', 'ws_default', 'historical-layout', 'Historical layout', '', 'draft', now(), now());
+         insert into platform_api_keys
+            (id, workspace_id, agent_id, name, prefix, secret_hash, key_kind, scopes_json, status,
+             created_at, expires_at, allowed_origins_json)
+         values ('key_historical_layout', 'ws_default', 'agent_historical_layout', 'Historical layout',
+                 'cp_', 'hash', 'admin', '[]'::jsonb, 'active', now(),
+                 '2030-01-02 03:04:05+00'::timestamptz, '[\"https://example.test\"]'::jsonb);
+         insert into approval_requests
+            (id, workspace_id, agent_id, operation_id, operation_version, status, risk_level,
+             request_payload_json, created_at, expires_at, execution_started_at,
+             execution_attempts, request_fingerprint)
+         values ('approval_historical_layout', 'ws_default', 'agent_historical_layout', 'op_historical_layout', 1,
+                 'pending', 'high', '{\"layout\":\"historical\"}'::jsonb, now(), now() + interval '1 hour',
+                 '2029-02-03 04:05:06+00'::timestamptz, 3, 'historical-fingerprint');",
+    )
+    .execute(&pool)
+    .await
+    .unwrap();
+
+    let historical_ordinals: Vec<(String, i32)> = sqlx::query_as(
+        "select table_name || '.' || column_name, ordinal_position
+         from information_schema.columns
+         where table_schema = current_schema()
+           and ((table_name = 'platform_api_keys'
+                 and column_name in ('key_kind', 'expires_at', 'allowed_origins_json'))
+             or (table_name = 'approval_requests'
+                 and column_name in ('execution_started_at', 'execution_attempts', 'request_fingerprint')))
+         order by table_name, ordinal_position",
+    )
+    .fetch_all(&pool)
+    .await
+    .unwrap();
+    assert_eq!(
+        historical_ordinals,
+        vec![
+            ("approval_requests.execution_started_at".to_owned(), 22),
+            ("approval_requests.execution_attempts".to_owned(), 23),
+            ("approval_requests.request_fingerprint".to_owned(), 24),
+            ("platform_api_keys.key_kind".to_owned(), 15),
+            ("platform_api_keys.expires_at".to_owned(), 16),
+            ("platform_api_keys.allowed_origins_json".to_owned(), 17),
+        ],
+        "fixture must reproduce the published in-place column layout",
+    );
+
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::MigrationRequired {
+            current: 0,
+            target: 13,
+        }
+    );
+    MigrationAuthority::apply(&pool).await.unwrap();
+
+    let key_count: i64 = sqlx::query_scalar(
+        "select count(*) from platform_api_keys
+         where id = 'key_historical_layout'
+           and key_kind = 'admin'
+           and expires_at = '2030-01-02 03:04:05+00'::timestamptz
+           and allowed_origins_json = '[\"https://example.test\"]'::jsonb",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert_eq!(
+        key_count, 1,
+        "platform key row must survive the layout upgrade"
+    );
+    let approval_count: i64 = sqlx::query_scalar(
+        "select count(*) from approval_requests
+         where id = 'approval_historical_layout'
+           and request_payload_json = '{\"layout\":\"historical\"}'::jsonb
+           and execution_started_at = '2029-02-03 04:05:06+00'::timestamptz
+           and execution_attempts = 3
+           and request_fingerprint = 'historical-fingerprint'",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert_eq!(
+        approval_count, 1,
+        "approval row must survive the layout upgrade"
+    );
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::Current { version: 13 }
+    );
+}
+
+#[tokio::test]
+async fn published_initial_mcp_layout_is_accepted() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_initial_mcp").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::raw_sql(
+        "drop table mcp_transport_sessions;
+         create table mcp_transport_sessions (
+             id text primary key,
+             protocol_version text not null,
+             initialized boolean not null default false,
+             workspace_slug text not null,
+             agent_slug text not null,
+             created_at timestamptz not null,
+             updated_at timestamptz not null,
+             expires_at timestamptz null
+         );
+         create index mcp_transport_sessions_workspace_agent_idx
+             on mcp_transport_sessions(workspace_slug, agent_slug, updated_at desc);",
+    )
+    .execute(&pool)
+    .await
+    .unwrap();
+
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::MigrationRequired {
+            current: 0,
+            target: 13,
+        }
+    );
+    MigrationAuthority::apply(&pool).await.unwrap();
+    let supports_elicitation: bool = sqlx::query_scalar(
+        "select exists (
+             select 1 from information_schema.columns
+             where table_schema = current_schema()
+               and table_name = 'mcp_transport_sessions'
+               and column_name = 'supports_elicitation'
+         )",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(supports_elicitation);
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::Current { version: 13 }
+    );
+}
+
+#[tokio::test]
+async fn published_in_place_mcp_upgrade_layout_is_accepted() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_mcp_upgrade").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::raw_sql(
+        "drop table mcp_transport_sessions;
+         create table mcp_transport_sessions (
+             id text primary key,
+             protocol_version text not null,
+             initialized boolean not null default false,
+             workspace_slug text not null,
+             agent_slug text not null,
+             created_at timestamptz not null,
+             updated_at timestamptz not null,
+             expires_at timestamptz null
+         );
+         create index mcp_transport_sessions_workspace_agent_idx
+             on mcp_transport_sessions(workspace_slug, agent_slug, updated_at desc);
+         alter table mcp_transport_sessions
+             add column supports_elicitation boolean not null default false;",
+    )
+    .execute(&pool)
+    .await
+    .unwrap();
+
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::MigrationRequired {
+            current: 0,
+            target: 13,
+        }
+    );
+    MigrationAuthority::apply(&pool).await.unwrap();
+    assert_eq!(
+        MigrationAuthority::preflight(&pool).await.unwrap(),
+        MigrationPreflight::Current { version: 13 }
+    );
+}
+
+#[tokio::test]
+async fn ledgerless_baseline_with_future_drift_is_rejected_without_writes() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_drift").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::query("alter table invocation_logs add column trace_id text")
+        .execute(&pool)
+        .await
+        .unwrap();
+
+    let error = MigrationAuthority::apply(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.legacy_fingerprint");
+    let core_exists: bool = sqlx::query_scalar(
+        "select to_regclass(format('%I.%I', current_schema(), '__crank_core_migrations')) is not null",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(!core_exists, "rejected adoption must remain read-only");
+}
+
+#[tokio::test]
+async fn ledgerless_baseline_with_default_drift_is_rejected_without_writes() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_default").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::query("alter table workspaces alter column status set default 'active'")
+        .execute(&pool)
+        .await
+        .unwrap();
+
+    let error = MigrationAuthority::apply(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.legacy_fingerprint");
+    let core_exists: bool = sqlx::query_scalar(
+        "select to_regclass(format('%I.%I', current_schema(), '__crank_core_migrations')) is not null",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(!core_exists, "rejected adoption must remain read-only");
+}
+
+#[tokio::test]
+async fn ledgerless_baseline_with_constraint_drift_is_rejected_without_writes() {
+    let database_url =
+        crank_test_support::postgres_schema_url("test_ledgerless_constraint_definition").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::query(
+        "alter table workspaces
+         add constraint workspaces_status_nonempty check (status <> '')",
+    )
+    .execute(&pool)
+    .await
+    .unwrap();
+
+    let error = MigrationAuthority::apply(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.legacy_fingerprint");
+    let core_exists: bool = sqlx::query_scalar(
+        "select to_regclass(format('%I.%I', current_schema(), '__crank_core_migrations')) is not null",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(!core_exists, "rejected adoption must remain read-only");
+}
+
+#[tokio::test]
+async fn ledgerless_baseline_with_rls_drift_is_rejected_without_writes() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_rls").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::query("alter table workspaces enable row level security")
+        .execute(&pool)
+        .await
+        .unwrap();
+
+    let error = MigrationAuthority::apply(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.legacy_fingerprint");
+    let core_exists: bool = sqlx::query_scalar(
+        "select to_regclass(format('%I.%I', current_schema(), '__crank_core_migrations')) is not null",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(!core_exists, "rejected adoption must remain read-only");
+}
+
+#[tokio::test]
+async fn ledgerless_baseline_with_nullability_drift_is_rejected_without_writes() {
+    let database_url = crank_test_support::postgres_schema_url("test_ledgerless_constraint").await;
+    let pool = sqlx::PgPool::connect(&database_url).await.unwrap();
+    ledgerless_v1(&pool).await;
+    sqlx::query("alter table workspaces alter column status drop not null")
+        .execute(&pool)
+        .await
+        .unwrap();
+
+    let error = MigrationAuthority::apply(&pool).await.unwrap_err();
+    assert_eq!(error.code(), "partial_sequence");
+    assert_eq!(error.stage(), "preflight.legacy_fingerprint");
+    let core_exists: bool = sqlx::query_scalar(
+        "select to_regclass(format('%I.%I', current_schema(), '__crank_core_migrations')) is not null",
+    )
+    .fetch_one(&pool)
+    .await
+    .unwrap();
+    assert!(!core_exists, "rejected adoption must remain read-only");
+}

deploy/community/docker-compose.images.yml

@@ -138,7 +138,7 @@ services:
     ports:
       - "${CRANK_PUBLISH_BIND:-127.0.0.1}:${CRANK_ADMIN_PUBLISH_PORT:-3001}:3001"
     healthcheck:
-      test: ["CMD", "curl", "--fail", "http://127.0.0.1:3001/ready"]
+      test: ["CMD", "/usr/local/bin/crank-http-healthcheck", "3001", "/ready"]
       interval: 15s
       timeout: 5s
       retries: 5
@@ -199,7 +199,7 @@ services:
     ports:
       - "${CRANK_PUBLISH_BIND:-127.0.0.1}:${CRANK_MCP_PUBLISH_PORT:-3002}:3002"
     healthcheck:
-      test: ["CMD", "curl", "--fail", "http://127.0.0.1:3002/ready"]
+      test: ["CMD", "/usr/local/bin/crank-http-healthcheck", "3002", "/ready"]
       interval: 15s
       timeout: 5s
       retries: 5

deploy/community/docker-compose.yml

@@ -42,6 +42,7 @@ services:
     depends_on:
       postgres:
         condition: service_healthy
+        required: false
     environment:
       POSTGRES_HOST: ${POSTGRES_HOST:-postgres}
       POSTGRES_PORT: ${POSTGRES_PORT:-5432}
@@ -141,7 +142,7 @@ services:
     ports:
       - "${CRANK_PUBLISH_BIND:-127.0.0.1}:${CRANK_ADMIN_PUBLISH_PORT:-3001}:3001"
     healthcheck:
-      test: ["CMD", "curl", "--fail", "http://127.0.0.1:3001/ready"]
+      test: ["CMD", "/usr/local/bin/crank-http-healthcheck", "3001", "/ready"]
       interval: 15s
       timeout: 5s
       retries: 5
@@ -202,7 +203,7 @@ services:
     ports:
       - "${CRANK_PUBLISH_BIND:-127.0.0.1}:${CRANK_MCP_PUBLISH_PORT:-3002}:3002"
     healthcheck:
-      test: ["CMD", "curl", "--fail", "http://127.0.0.1:3002/ready"]
+      test: ["CMD", "/usr/local/bin/crank-http-healthcheck", "3002", "/ready"]
       interval: 15s
       timeout: 5s
       retries: 5

docker-compose.yml

@@ -127,7 +127,7 @@ services:
     ports:
       - "${CRANK_PUBLISH_BIND:-127.0.0.1}:3001:3001"
     healthcheck:
-      test: ["CMD", "curl", "--fail", "http://127.0.0.1:3001/ready"]
+      test: ["CMD", "/usr/local/bin/crank-http-healthcheck", "3001", "/ready"]
       interval: 15s
       timeout: 5s
       retries: 5
@@ -190,7 +190,7 @@ services:
     ports:
       - "${CRANK_PUBLISH_BIND:-127.0.0.1}:3002:3002"
     healthcheck:
-      test: ["CMD", "curl", "--fail", "http://127.0.0.1:3002/health"]
+      test: ["CMD", "/usr/local/bin/crank-http-healthcheck", "3002", "/health"]
       interval: 15s
       timeout: 5s
       retries: 5

docs/migrations.md

@@ -16,7 +16,7 @@ docker compose -f deploy/community/docker-compose.yml --env-file deploy/communit
 2. Создайте и проверьте согласованный backup PostgreSQL и artifact storage. Для первой пустой установки зафиксируйте, что восстанавливать нечего.
 3. Проверьте immutable plan: `cargo run -p admin-api --bin crank-migrate -- plan --check` в source checkout либо `<compose> run --rm migrate crank-migrate plan` для образа.
 4. Примените sequence: `<compose> run --rm migrate crank-migrate apply`.
-5. Повторите preflight и убедитесь в `{"status":"current","version":12}`.
+5. Повторите preflight и убедитесь в `{"status":"current","version":13}`.
 6. Только теперь запускайте long-running services: `<compose> up -d`.
 
 Обычный `up` также содержит обязательный migration job, но при upgrade он не заменяет предварительные preflight и backup. Migrator делает до десяти bounded попыток подключения с секундной паузой и затем безопасно завершается ошибкой.
@@ -90,6 +90,9 @@ cargo run -p admin-api --bin crank-migrate -- plan --check
   sensitivity и source lifecycle принадлежат scoped relation. Claim token/expiry
   зарезервированы bounded all-or-none contract для последующего reconciliation,
   но эта версия не запускает cleanup и не удаляет physical blobs.
+- V13 — expand-only indexes, поддерживающие bounded artifact/import cleanup и
+  reconciliation ранее зарезервированных claim token/expiry; эта версия сама
+  не выполняет cleanup, не переписывает данные и не удаляет physical blobs.
 - Каждая версия имеет contiguous `i64` version, стабильное имя, lowercase SHA-256, owner, phase, explicit readable schema min/max и backfill policy.
 - `migrate` требует bounded cursor/batch policy; `contract` дополнительно требует tracked compatibility evidence и закрытого окна.
 - Добавление descriptor без executable implementation блокируется `invalid_contract` до DB I/O.

docs/schemas/migration-sequence.json

@@ -192,6 +192,22 @@
       "source_digest": "052058da53cd861b2a1af81243b34cc35204d665a9276f8ece563e061f8ebbfe",
       "transactional": true,
       "version": 12
+    },
+    {
+      "backfill": {
+        "kind": "none"
+      },
+      "checksum": "abe822c967e1b2cc3966ede988e05eb2d9b2a062b539c45aaa8cad677395ab05",
+      "compatibility": "n-minus-one-readable",
+      "contract_evidence": null,
+      "name": "artifact-cleanup-indexes-v13",
+      "owner": "crank-registry",
+      "phase": "expand",
+      "readable_schema_max": 13,
+      "readable_schema_min": 12,
+      "source_digest": "abe822c967e1b2cc3966ede988e05eb2d9b2a062b539c45aaa8cad677395ab05",
+      "transactional": true,
+      "version": 13
     }
   ]
 }

scripts/authenticated-product-smoke.py

@@ -21,10 +21,17 @@ class SmokeError(RuntimeError):
     pass
 
 
-def safe_error(stage: str, code: str, status: int | None = None) -> SmokeError:
+def safe_error(
+    stage: str,
+    code: str,
+    status: int | None = None,
+    trace_id: str | None = None,
+) -> SmokeError:
     message = f"stage={stage[:64]} code={code[:96]}"
     if status is not None:
         message += f" status={status}"
+    if trace_id is not None:
+        message += f" trace_id={trace_id}"
     return SmokeError(message[:256])
 
 
@@ -134,6 +141,7 @@ class Client:
     def __init__(self, base_url: str, timeout_seconds: int) -> None:
         self.base_url = base_url.rstrip("/")
         self.timeout_seconds = timeout_seconds
+        self.csrf_token: str | None = None
         cookie_jar = http.cookiejar.CookieJar()
         self.opener = urllib.request.build_opener(
             urllib.request.HTTPCookieProcessor(cookie_jar)
@@ -156,6 +164,21 @@ class Client:
         request_headers = {"Accept": "application/json"}
         if headers:
             request_headers.update(headers)
+        if (
+            self.csrf_token
+            and method not in ("GET", "HEAD", "OPTIONS")
+            and (
+                path_or_url.startswith("/api/admin/")
+                or path_or_url.startswith("/api/auth/")
+            )
+            and path_or_url
+            not in (
+                "/api/auth/login",
+                "/api/auth/bootstrap/complete",
+                "/api/auth/session/csrf",
+            )
+        ):
+            request_headers.setdefault("x-csrf-token", self.csrf_token)
         if payload is not None:
             data = json.dumps(payload).encode("utf-8")
             request_headers["Content-Type"] = "application/json"
@@ -203,12 +226,21 @@ def admin_path(workspace_id: str, suffix: str) -> str:
 
 
 def login(client: Client, email: str, password: str) -> None:
-    client.request_json(
+    response = client.request_json(
         "POST",
         "/api/auth/login",
         {"email": email, "password": password},
         expected=(200,),
     )
+    session = require_object(response.body, "login")
+    csrf_token = session.get("csrf_token")
+    if (
+        not isinstance(csrf_token, str)
+        or not 32 <= len(csrf_token) <= 256
+        or not all(character.isalnum() or character in "-_." for character in csrf_token)
+    ):
+        raise safe_error("login", "invalid_csrf_token")
+    client.csrf_token = csrf_token
 
 
 def resolve_workspace(
@@ -261,6 +293,29 @@ def create_operation(
         raise safe_error("operation_create", "invalid_response") from error
 
 
+def is_safe_diagnostic_identifier(value: Any) -> bool:
+    return (
+        isinstance(value, str)
+        and 1 <= len(value) <= 64
+        and value[0].islower()
+        and value[0].isascii()
+        and all(
+            character.isascii()
+            and (character.islower() or character.isdigit() or character == "_")
+            for character in value
+        )
+    )
+
+
+def is_safe_trace_id(value: Any) -> bool:
+    return (
+        isinstance(value, str)
+        and len(value) == 32
+        and value != "0" * 32
+        and all(character in "0123456789abcdef" for character in value)
+    )
+
+
 def run_operation_test(
     client: Client,
     workspace_id: str,
@@ -272,9 +327,24 @@ def run_operation_test(
         admin_path(workspace_id, f"/operations/{operation_id}/test-runs"),
         {"version": operation_version, "input": {"probe": "ok"}},
     ).body
-    if not isinstance(result, dict) or result.get("ok") is not True:
+    if isinstance(result, dict) and result.get("ok") is True:
+        return
+    if not isinstance(result, dict) or result.get("ok") is not False:
         raise safe_error("operation_test", "outcome_not_ok")
 
+    errors = result.get("errors")
+    failure = errors[0] if isinstance(errors, list) and errors else None
+    code = failure.get("code") if isinstance(failure, dict) else None
+    stage = failure.get("stage") if isinstance(failure, dict) else None
+    trace_id = result.get("trace_id")
+    if (
+        is_safe_diagnostic_identifier(code)
+        and is_safe_diagnostic_identifier(stage)
+        and is_safe_trace_id(trace_id)
+    ):
+        raise safe_error(stage, code, trace_id=trace_id)
+    raise safe_error("operation_test", "outcome_not_ok")
+
 
 def operation_etag(client: Client, workspace_id: str, operation_id: str) -> str:
     response = client.request_json(
@@ -326,6 +396,17 @@ def create_agent(client: Client, workspace_id: str, agent_slug: str) -> tuple[st
         raise safe_error("agent_create", "invalid_response") from error
 
 
+def agent_etag(client: Client, workspace_id: str, agent_id: str) -> str:
+    response = client.request_json(
+        "GET",
+        admin_path(workspace_id, f"/agents/{agent_id}"),
+    )
+    etag = response.headers.get("ETag") if response.headers is not None else None
+    if not isinstance(etag, str) or len(etag) > 128 or not etag.startswith('"') or not etag.endswith('"'):
+        raise safe_error("agent_precondition", "invalid_response")
+    return etag
+
+
 def edit_and_archive_operation(
     client: Client,
     workspace_id: str,
@@ -377,11 +458,13 @@ def bind_and_publish_agent(
                 "enabled": True,
             }
         ],
+        headers={"If-Match": agent_etag(client, workspace_id, agent_id)},
     )
     published = client.request_json(
         "POST",
         admin_path(workspace_id, f"/agents/{agent_id}/publish"),
         {"version": agent_version},
+        headers={"If-Match": agent_etag(client, workspace_id, agent_id)},
     ).body
     try:
         published_version = int(published["published_version"])
@@ -436,8 +519,15 @@ def cleanup_smoke_assets(
     if agent_id:
         try:
             client.request_json(
-                "DELETE",
+                "POST",
-                admin_path(workspace_id, f"/agents/{agent_id}"),
+                admin_path(workspace_id, f"/agents/{agent_id}/unpublish"),
+                headers={"If-Match": agent_etag(client, workspace_id, agent_id)},
+                expected=(200, 404),
+            )
+            client.request_json(
+                "POST",
+                admin_path(workspace_id, f"/agents/{agent_id}/archive"),
+                headers={"If-Match": agent_etag(client, workspace_id, agent_id)},
                 expected=(200, 404),
             )
         except SmokeError as error:

scripts/collect-capability-baseline.py

@@ -90,26 +90,52 @@ def playwright_verdict(report: dict[str, Any], required_titles: list[str]) -> tu
         return "fail", counts
     tests = list(iter_tests(report))
     if not tests:
+        if required_titles:
+            counts["failed"] = len(set(required_titles))
+            return "fail", counts
         counts["not_run"] = 1
         return "not_run", counts
     required = {title: False for title in required_titles}
     for test, title in tests:
+        if required and title not in required:
+            continue
         status = test.get("status")
-        results = test.get("results") if isinstance(test.get("results"), list) else []
+        results = test.get("results")
-        result_statuses = [result.get("status") for result in results if isinstance(result, dict)]
+        if not isinstance(results, list):
-        retries = [result.get("retry", 0) for result in results if isinstance(result, dict)]
-        if len(result_statuses) != len(results):
             counts["not_run"] += 1
             continue
-        if status == "flaky" or any(isinstance(retry, int) and retry > 0 for retry in retries):
+        result_statuses: list[Any] = []
-            counts["flaky"] += 1
+        retries: list[int] = []
+        malformed = False
+        for result in results:
+            if not isinstance(result, dict):
+                malformed = True
+                break
+            result_status = result.get("status")
+            retry = result.get("retry", 0)
+            if result_status not in ("passed", "failed", "timedOut", "skipped", "interrupted") \
+                or type(retry) is not int or retry < 0:
+                malformed = True
+                break
+            result_statuses.append(result_status)
+            retries.append(retry)
+        if malformed:
+            counts["not_run"] += 1
+            continue
+        final_status = result_statuses[-1] if result_statuses else None
+        if status in ("unexpected", "failed", "timedOut", "interrupted") \
+            or final_status in ("failed", "timedOut", "interrupted"):
+            counts["failed"] += 1
+        elif status == "flaky":
+            if final_status == "passed" and any(retry > 0 for retry in retries):
+                counts["flaky"] += 1
+            else:
+                counts["not_run"] += 1
         elif status == "skipped" or (not results and status in ("skipped", "expected")):
             counts["skipped"] += 1
-        elif status in ("unexpected", "failed", "timedOut", "interrupted") or any(
+        elif any(retry > 0 for retry in retries):
-            result_status in ("failed", "timedOut", "interrupted") for result_status in result_statuses
+            counts["flaky"] += 1
-        ):
+        elif status == "expected" and results and all(result_status == "passed" for result_status in result_statuses):
-            counts["failed"] += 1
-        elif results and all(result_status == "passed" for result_status in result_statuses):
             counts["passed"] += 1
             if title in required:
                 required[title] = True

scripts/deploy-community.sh

@@ -26,13 +26,28 @@ compose_profiles=""
 if [ "$cache_backend" = "valkey" ] || [ "$cache_backend" = "redis" ]; then
     compose_profiles="--profile cache"
 fi
-
 compose() {
     # Intentional word splitting: compose_profiles is either empty or two arguments.
     # shellcheck disable=SC2086
     docker compose $compose_profiles "$@"
 }
 
+compose_up() {
+    postgres_scale=""
+    if [ "$(env_value POSTGRES_HOST postgres)" != "postgres" ] \
+        && compose config --services | grep -Fxq postgres; then
+        postgres_scale="--scale postgres=0"
+    fi
+    # Intentional word splitting: postgres_scale is either empty or two arguments.
+    # shellcheck disable=SC2086
+    compose up -d --remove-orphans $postgres_scale
+}
+
+show_failure_diagnostics() {
+    compose ps >&2 || true
+    compose logs --no-color migrate >&2 || true
+}
+
 wait_for_stack() {
     readiness_path="$1"
     attempt=1
@@ -48,6 +63,34 @@ wait_for_stack() {
     return 1
 }
 
+find_artifact_container() {
+    expected_root="$1"
+    expected_mount="$(printf 'volume\t%s' "$expected_root")"
+
+    for service in admin-api artifact-storage-init; do
+        candidates="$(compose ps -aq "$service" 2>/dev/null || true)"
+        # Intentional word splitting: Docker container IDs cannot contain whitespace.
+        # shellcheck disable=SC2086
+        set -- $candidates
+        if [ "$#" -gt 1 ]; then
+            echo "Artifact backup found multiple $service containers; refusing an ambiguous volume source" >&2
+            return 1
+        fi
+        if [ "$#" -eq 1 ]; then
+            candidate="$1"
+            oneoff="$(docker inspect --format '{{index .Config.Labels "com.docker.compose.oneoff"}}' "$candidate" 2>/dev/null || true)"
+            mounts="$(docker inspect --format '{{range .Mounts}}{{printf "%s\t%s\n" .Type .Destination}}{{end}}' "$candidate" 2>/dev/null || true)"
+            if [ "$oneoff" != "True" ] && [ "$oneoff" != "true" ] \
+                && printf '%s\n' "$mounts" | grep -Fxq "$expected_mount"; then
+                printf '%s' "$candidate"
+                return 0
+            fi
+        fi
+    done
+
+    return 1
+}
+
 create_backup() {
     timestamp="$(date -u +%Y%m%dT%H%M%SZ)"
     backup_dir="$(pwd)/backups/${timestamp}"
@@ -86,10 +129,9 @@ create_backup() {
             --username "$postgres_user" --dbname "$postgres_db" \
             --format custom --file /backup/postgres.dump
 
-    admin_container="$(compose ps -q admin-api 2>/dev/null || true)"
+    storage_root="$(env_value_from "$backup_env_file" CRANK_STORAGE_ROOT /var/lib/crank/storage)"
-    if [ -n "$admin_container" ]; then
+    if artifact_container="$(find_artifact_container "$storage_root")"; then
-        storage_root="$(env_value_from "$backup_env_file" CRANK_STORAGE_ROOT /var/lib/crank/storage)"
+        docker run --rm --volumes-from "$artifact_container:ro" \
-        docker run --rm --volumes-from "$admin_container" \
             -v "$backup_dir:/backup" alpine:3.21 \
             tar -C "$storage_root" -czf /backup/artifacts.tar.gz .
     elif [ "$previous_deployment" = true ]; then
@@ -104,6 +146,76 @@ create_backup() {
         | sort -nr | awk 'NR > 5 { print $2 }' | xargs -r rm -rf
 }
 
+validate_backup_migration() {
+    admin_image="$(env_value CRANK_ADMIN_API_IMAGE)"
+    if [ -z "$admin_image" ]; then
+        echo "CRANK_ADMIN_API_IMAGE is required for shadow migration validation" >&2
+        return 1
+    fi
+
+    shadow_id="crank-migration-check-$$"
+    shadow_network="${shadow_id}-network"
+    shadow_postgres="${shadow_id}-postgres"
+    shadow_password="crank-shadow-migration-password"
+    shadow_admin="crank_shadow_admin"
+    shadow_user="crank_shadow_owner"
+    shadow_cleanup() {
+        docker rm -fv "$shadow_postgres" >/dev/null 2>&1 || true
+        docker network rm "$shadow_network" >/dev/null 2>&1 || true
+    }
+    trap shadow_cleanup EXIT
+    trap 'exit 130' HUP INT TERM
+
+    docker network create "$shadow_network" >/dev/null
+    docker run -d --name "$shadow_postgres" --network "$shadow_network" \
+        -e POSTGRES_USER="$shadow_admin" \
+        -e POSTGRES_PASSWORD="$shadow_password" \
+        -e POSTGRES_DB=crank \
+        postgres:16-alpine >/dev/null
+
+    shadow_ready=false
+    attempt=1
+    while [ "$attempt" -le 30 ]; do
+        if docker logs "$shadow_postgres" 2>&1 | grep -q 'PostgreSQL init process complete' \
+            && docker exec "$shadow_postgres" pg_isready --username "$shadow_admin" --dbname crank >/dev/null 2>&1; then
+            shadow_ready=true
+            break
+        fi
+        sleep 1
+        attempt=$((attempt + 1))
+    done
+    if [ "$shadow_ready" != true ]; then
+        echo "Shadow PostgreSQL did not become ready" >&2
+        return 1
+    fi
+
+    docker exec "$shadow_postgres" psql --username "$shadow_admin" --dbname crank \
+        --set ON_ERROR_STOP=1 \
+        --command "create role ${shadow_user} login password '${shadow_password}' nosuperuser nocreatedb nocreaterole noreplication" \
+        --command "alter database crank owner to ${shadow_user}"
+
+    timeout 10m docker run --rm --network "$shadow_network" \
+        -e PGPASSWORD="$shadow_password" \
+        -v "$backup_dir:/backup:ro" \
+        postgres:16-alpine \
+        pg_restore --host "$shadow_postgres" --username "$shadow_user" --dbname crank \
+            --no-owner --no-privileges --single-transaction --exit-on-error \
+            /backup/postgres.dump
+
+    shadow_database_url="postgres://${shadow_user}:${shadow_password}@${shadow_postgres}:5432/crank"
+    timeout 10m docker run --rm --network "$shadow_network" \
+        -e CRANK_DATABASE_URL="$shadow_database_url" \
+        "$admin_image" crank-migrate apply
+    shadow_preflight="$(timeout 2m docker run --rm --network "$shadow_network" \
+        -e CRANK_DATABASE_URL="$shadow_database_url" \
+        "$admin_image" crank-migrate preflight)"
+    printf '%s\n' "$shadow_preflight" | grep -Fxq '{"status":"current","version":13}'
+
+    shadow_cleanup
+    trap - EXIT HUP INT TERM
+    echo "Shadow migration validation passed"
+}
+
 rollback() {
     echo "New release failed readiness; restoring previous deployment" >&2
     if [ ! -f .env.previous ] || [ ! -f docker-compose.previous.yml ]; then
@@ -119,16 +231,17 @@ rollback() {
     if [ "$cache_backend" = "valkey" ] || [ "$cache_backend" = "redis" ]; then
         compose_profiles="--profile cache"
     fi
-    compose up -d --remove-orphans
+    compose_up
     wait_for_stack health
 }
 
 compose config -q
 create_backup
 compose pull
+validate_backup_migration
 
-if ! compose up -d --remove-orphans || ! wait_for_stack ready; then
+if ! compose_up || ! wait_for_stack ready; then
-    compose ps >&2 || true
+    show_failure_diagnostics
     rollback
     exit 1
 fi

scripts/docker-http-healthcheck.sh

@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+port="${1:?port is required}"
+path="${2:?path is required}"
+
+if [[ ! "$port" =~ ^[0-9]{1,5}$ ]] || (( port < 1 || port > 65535 )); then
+    exit 64
+fi
+
+if [[ "$path" != /* || "$path" == *$'\r'* || "$path" == *$'\n'* ]]; then
+    exit 64
+fi
+
+# Bash provides /dev/tcp without adding a network client package to the runtime
+# image. Compose still applies its own five-second timeout to the whole probe.
+exec 3<>"/dev/tcp/127.0.0.1/${port}"
+printf 'GET %s HTTP/1.1\r\nHost: 127.0.0.1\r\nConnection: close\r\n\r\n' "$path" >&3
+
+IFS=$'\r' read -r -t 2 status <&3
+case "$status" in
+    'HTTP/1.0 200 '*|'HTTP/1.1 200 '*) ;;
+    *) exit 1 ;;
+esac

scripts/restore-community.sh

@@ -32,6 +32,15 @@ compose() {
     # shellcheck disable=SC2086
     docker compose $compose_profiles "$@"
 }
+compose_up() {
+    postgres_scale=""
+    if [ "$(env_value POSTGRES_HOST postgres)" != "postgres" ] \
+        && compose config --services | grep -Fxq postgres; then
+        postgres_scale="--scale postgres=0"
+    fi
+    # shellcheck disable=SC2086
+    compose up -d --remove-orphans $postgres_scale
+}
 
 compose stop admin-api mcp-server ui
 
@@ -40,13 +49,18 @@ postgres_port="$(env_value POSTGRES_PORT 5432)"
 postgres_db="$(env_value POSTGRES_DB crank)"
 postgres_user="$(env_value POSTGRES_USER crank)"
 postgres_password="$(env_value POSTGRES_PASSWORD)"
+docker run --rm \
+    -v "$backup_dir:/backup:ro" \
+    postgres:16-alpine \
+    pg_restore --list /backup/postgres.dump >/dev/null
 docker run --rm --network host \
     -e PGPASSWORD="$postgres_password" \
     -v "$backup_dir:/backup:ro" \
     postgres:16-alpine \
     pg_restore --host "$postgres_host" --port "$postgres_port" \
         --username "$postgres_user" --dbname "$postgres_db" \
-        --clean --if-exists --no-owner --no-privileges /backup/postgres.dump
+        --clean --if-exists --no-owner --no-privileges \
+        --single-transaction --exit-on-error /backup/postgres.dump
 
 admin_container="$(compose ps -aq admin-api)"
 storage_root="$(env_value CRANK_STORAGE_ROOT /var/lib/crank/storage)"
@@ -54,7 +68,7 @@ docker run --rm --volumes-from "$admin_container" \
     -v "$backup_dir:/backup:ro" alpine:3.21 sh -eu -c \
     "find '$storage_root' -mindepth 1 -delete; tar -C '$storage_root' -xzf /backup/artifacts.tar.gz"
 
-compose up -d --remove-orphans
+compose_up
 attempt=1
 while [ "$attempt" -le 45 ]; do
     if curl --fail --silent http://127.0.0.1:3000/ >/dev/null \

tests/unit/test_authenticated_product_smoke.py

@@ -17,6 +17,48 @@ def load_smoke_module():
 
 
 class AuthenticatedProductSmokeTests(unittest.TestCase):
+    def test_client_attaches_csrf_only_to_browser_api_mutations(self) -> None:
+        smoke = load_smoke_module()
+
+        class Response:
+            status = 200
+            headers = {}
+
+            def read(self, _limit):
+                return b"{}"
+
+        class Opener:
+            def __init__(self):
+                self.requests = []
+
+            def open(self, request, timeout):
+                self.requests.append((request, timeout))
+                return Response()
+
+        client = smoke.Client("http://crank.test", 5)
+        opener = Opener()
+        client.opener = opener
+        client.csrf_token = "a" * 32
+
+        client.request_json("POST", "/api/admin/workspaces/ws/operations", {})
+        client.request_json("POST", "http://mcp.test/v1/ws/agent", {})
+
+        self.assertEqual(opener.requests[0][0].get_header("X-csrf-token"), "a" * 32)
+        self.assertIsNone(opener.requests[1][0].get_header("X-csrf-token"))
+
+    def test_login_keeps_server_issued_csrf_token(self) -> None:
+        smoke = load_smoke_module()
+
+        class FakeClient:
+            csrf_token = None
+
+            def request_json(self, *args, **kwargs):
+                return smoke.JsonResponse(200, {}, {"csrf_token": "b" * 32})
+
+        client = FakeClient()
+        smoke.login(client, "owner@crank.test", "safe-password")
+        self.assertEqual(client.csrf_token, "b" * 32)
+
     def test_operation_payload_uses_internal_upstream(self) -> None:
         smoke = load_smoke_module()
 
@@ -136,6 +178,8 @@ class AuthenticatedProductSmokeTests(unittest.TestCase):
                     return smoke.JsonResponse(200, {}, {"operation_id": "op_safe", "version": 7})
                 if path.endswith("/operations/op_safe"):
                     return smoke.JsonResponse(200, {"ETag": '"safe-etag"'}, {"id": "op_safe"})
+                if path.endswith("/agents/agent_safe"):
+                    return smoke.JsonResponse(200, {"ETag": '"safe-agent-etag"'}, {"id": "agent_safe"})
                 if path.endswith("/publish") and "/operations/" in path:
                     return smoke.JsonResponse(200, {}, {"published_version": 7})
                 if path.endswith("/agents"):
@@ -158,11 +202,14 @@ class AuthenticatedProductSmokeTests(unittest.TestCase):
         self.assertEqual((agent_id, agent_version, published_agent_version), ("agent_safe", 3, 3))
         binding = next(payload for _, path, payload, _ in client.requests if path.endswith("/bindings"))[0]
         self.assertEqual(binding["operation_version"], 7)
+        binding_request = next(request for request in client.requests if request[1].endswith("/bindings"))
+        self.assertEqual(binding_request[3]["headers"], {"If-Match": '"safe-agent-etag"'})
         publish = next(request for request in client.requests if request[1].endswith("/operations/op_safe/publish"))
         self.assertEqual(publish[3]["headers"], {"If-Match": '"safe-etag"'})
 
-    def test_admin_test_run_uses_created_version_and_rejects_failed_outcome(self) -> None:
+    def test_admin_test_run_uses_created_version_and_reports_safe_typed_failure(self) -> None:
         smoke = load_smoke_module()
+        trace_id = "0123456789abcdef0123456789abcdef"
 
         class FakeClient:
             def __init__(self, ok):
@@ -171,7 +218,22 @@ class AuthenticatedProductSmokeTests(unittest.TestCase):
 
             def request_json(self, method, path, payload=None, **kwargs):
                 self.payload = payload
-                return smoke.JsonResponse(200, {}, {"ok": self.ok, "errors": [{"message": "secret-canary"}]})
+                return smoke.JsonResponse(
+                    200,
+                    {},
+                    {
+                        "ok": self.ok,
+                        "trace_id": trace_id,
+                        "errors": [
+                            {
+                                "code": "outbound_target_rejected",
+                                "stage": "adapter",
+                                "message": "secret-canary",
+                                "context": {"url": "https://private.invalid/token-canary"},
+                            }
+                        ],
+                    },
+                )
 
         passing = FakeClient(True)
         smoke.run_operation_test(passing, "ws", "op", 9)
@@ -179,7 +241,188 @@ class AuthenticatedProductSmokeTests(unittest.TestCase):
 
         with self.assertRaises(smoke.SmokeError) as raised:
             smoke.run_operation_test(FakeClient(False), "ws", "op", 9)
-        self.assertNotIn("secret-canary", str(raised.exception))
+        self.assertEqual(
+            str(raised.exception),
+            "stage=adapter code=outbound_target_rejected trace_id=0123456789abcdef0123456789abcdef",
+        )
+        for forbidden in ("secret-canary", "private.invalid", "token-canary", "context"):
+            self.assertNotIn(forbidden, str(raised.exception))
+
+    def test_admin_test_run_fails_closed_for_malformed_diagnostics(self) -> None:
+        smoke = load_smoke_module()
+        valid_trace_id = "0123456789abcdef0123456789abcdef"
+        valid_failure = {"code": "upstream_timeout", "stage": "upstream"}
+        malformed_results = (
+            None,
+            [],
+            "secret-canary",
+            {"trace_id": valid_trace_id, "errors": [valid_failure]},
+            {"ok": None, "trace_id": valid_trace_id, "errors": [valid_failure]},
+            {"ok": 0, "trace_id": valid_trace_id, "errors": [valid_failure]},
+            {"ok": "false", "trace_id": valid_trace_id, "errors": [valid_failure]},
+            {"ok": False, "errors": []},
+            {"ok": False, "errors": {}},
+            {"ok": False, "trace_id": valid_trace_id, "errors": [None]},
+            {"ok": False, "errors": ["secret-canary"]},
+            {"ok": False, "errors": [valid_failure]},
+            {
+                "ok": False,
+                "trace_id": valid_trace_id[:-1],
+                "errors": [valid_failure],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id + "0",
+                "errors": [valid_failure],
+            },
+            {
+                "ok": False,
+                "trace_id": "0" * 32,
+                "errors": [valid_failure],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id.upper(),
+                "errors": [valid_failure],
+            },
+            {
+                "ok": False,
+                "trace_id": "a" * 31 + "١",
+                "errors": [valid_failure],
+            },
+            {
+                "ok": False,
+                "trace_id": "a" * 31 + "\n",
+                "errors": [valid_failure],
+            },
+            {
+                "ok": False,
+                "trace_id": 42,
+                "errors": [valid_failure],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "", "stage": "upstream"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "a" * 65, "stage": "upstream"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "Upstream_timeout", "stage": "upstream"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "upstream_таймаут", "stage": "upstream"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": 42, "stage": "upstream"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "upstream_timeout", "stage": ""}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "upstream_timeout", "stage": "a" * 65}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "upstream_timeout", "stage": "Upstream"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "upstream_timeout", "stage": "вверх"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "upstream_timeout", "stage": 42}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "unsafe-value://secret-canary", "stage": "upstream"}],
+            },
+            {
+                "ok": False,
+                "trace_id": valid_trace_id,
+                "errors": [{"code": "upstream_timeout", "stage": "unsafe stage secret-canary"}],
+            },
+            {
+                "ok": False,
+                "trace_id": "not-a-trace-id-secret-canary",
+                "errors": [{"code": "upstream_timeout", "stage": "upstream"}],
+            },
+        )
+
+        class FakeClient:
+            def __init__(self, result):
+                self.result = result
+
+            def request_json(self, *args, **kwargs):
+                return smoke.JsonResponse(200, {}, self.result)
+
+        for result in malformed_results:
+            with self.subTest(result=result), self.assertRaises(smoke.SmokeError) as raised:
+                smoke.run_operation_test(FakeClient(result), "ws", "op", 9)
+            self.assertEqual(str(raised.exception), "stage=operation_test code=outcome_not_ok")
+            self.assertNotIn("secret-canary", str(raised.exception))
+
+    def test_admin_test_run_accepts_maximum_length_safe_diagnostics(self) -> None:
+        smoke = load_smoke_module()
+        identifier = "a" * 64
+        trace_id = "0123456789abcdef0123456789abcdef"
+
+        class FakeClient:
+            def request_json(self, *args, **kwargs):
+                return smoke.JsonResponse(
+                    200,
+                    {},
+                    {
+                        "ok": False,
+                        "trace_id": trace_id,
+                        "request_preview": {"credential": "secret-canary"},
+                        "response_preview": {"token": "secret-canary"},
+                        "errors": [{"code": identifier, "stage": identifier}],
+                    },
+                )
+
+        with self.assertRaises(smoke.SmokeError) as raised:
+            smoke.run_operation_test(FakeClient(), "ws", "op", 9)
+        rendered = str(raised.exception)
+        self.assertEqual(rendered, f"stage={identifier} code={identifier} trace_id={trace_id}")
+        self.assertLessEqual(len(rendered.encode("ascii")), 256)
+        self.assertNotIn("secret-canary", rendered)
+
+    def test_admin_test_run_success_ignores_malformed_diagnostics(self) -> None:
+        smoke = load_smoke_module()
+
+        class FakeClient:
+            def request_json(self, *args, **kwargs):
+                return smoke.JsonResponse(
+                    200,
+                    {},
+                    {
+                        "ok": True,
+                        "trace_id": "secret-canary",
+                        "errors": "secret-canary",
+                        "request_preview": {"token": "secret-canary"},
+                    },
+                )
+
+        smoke.run_operation_test(FakeClient(), "ws", "op", 9)
 
     def test_edit_and_archive_use_fresh_operation_preconditions(self) -> None:
         smoke = load_smoke_module()

tests/unit/test_collect_capability_baseline.py

@@ -57,6 +57,7 @@ class CapabilityBaselineCollectorTests(unittest.TestCase):
     def test_required_openapi_tests_fail_closed_when_missing_skipped_or_flaky(self) -> None:
         required = ["OpenAPI required scenario"]
         reports = [
+            {"suites": []},
             {"suites": [{"specs": [{"title": "another scenario", "tests": [{"status": "expected", "results": [{"status": "passed", "retry": 0}]}]}]}]},
             {"suites": [{"specs": [{"title": required[0], "tests": [{"status": "skipped", "results": []}]}]}]},
             {"suites": [{"specs": [{"title": required[0], "tests": [{"status": "flaky", "results": [{"status": "failed", "retry": 0}, {"status": "passed", "retry": 1}]}]}]}]},
@@ -70,6 +71,73 @@ class CapabilityBaselineCollectorTests(unittest.TestCase):
                 self.assertEqual(candidate["execution_verdict"], "fail")
                 self.assertFalse(candidate["accepted"])
 
+    def test_required_scope_ignores_unrelated_flaky_tests(self) -> None:
+        required = "OpenAPI required scenario"
+        report = {
+            "suites": [
+                {
+                    "specs": [
+                        {
+                            "title": required,
+                            "tests": [{"status": "expected", "results": [{"status": "passed", "retry": 0}]}],
+                        },
+                        {
+                            "title": "Unrelated wizard scenario",
+                            "tests": [
+                                {
+                                    "status": "flaky",
+                                    "results": [
+                                        {"status": "failed", "retry": 0},
+                                        {"status": "passed", "retry": 1},
+                                    ],
+                                }
+                            ],
+                        },
+                        {"title": "Unrelated failure", "tests": [{"status": "unexpected", "results": [{"status": "failed", "retry": 0}]}]},
+                        {"title": "Unrelated skip", "tests": [{"status": "skipped", "results": []}]},
+                        {"title": "Unrelated malformed", "tests": [{"status": "expected", "results": None}]},
+                    ]
+                }
+            ]
+        }
+        result, output, temporary = self.run_playwright(report, [required])
+        self.addCleanup(temporary.cleanup)
+        self.assertEqual(result.returncode, 0, result.stderr)
+        candidate = json.loads(output.read_text(encoding="utf-8"))
+        self.assertEqual(candidate["execution_verdict"], "pass")
+        self.assertTrue(candidate["accepted"])
+        self.assertEqual(candidate["summary"], {"passed": 1, "failed": 0, "flaky": 0, "skipped": 0, "not_run": 0})
+
+    def test_final_failure_after_retry_is_failed_not_flaky(self) -> None:
+        for status in ("unexpected", "flaky"):
+            with self.subTest(status=status):
+                report = {
+                    "suites": [
+                        {
+                            "specs": [
+                                {
+                                    "tests": [
+                                        {
+                                            "status": status,
+                                            "results": [
+                                                {"status": "failed", "retry": 0},
+                                                {"status": "failed", "retry": 1},
+                                            ],
+                                        }
+                                    ]
+                                }
+                            ]
+                        }
+                    ]
+                }
+                result, output, temporary = self.run_playwright(report)
+                self.addCleanup(temporary.cleanup)
+                self.assertEqual(result.returncode, 0, result.stderr)
+                candidate = json.loads(output.read_text(encoding="utf-8"))
+                self.assertEqual(candidate["execution_verdict"], "fail")
+                self.assertEqual(candidate["summary"]["failed"], 1)
+                self.assertEqual(candidate["summary"]["flaky"], 0)
+
     def test_raw_report_content_never_reaches_candidate_or_error(self) -> None:
         canary = "Bearer secret-canary /home/private/workspace https://private.invalid?q=secret"
         report = {"suites": [], "errors": [{"message": canary}], "stdout": [canary]}
@@ -84,6 +152,8 @@ class CapabilityBaselineCollectorTests(unittest.TestCase):
         reports = [
             {"errors": [{"message": "fatal"}], "suites": [{"specs": [{"tests": [{"status": "expected", "results": [{"status": "passed"}]}]}]}]},
             {"suites": [{"specs": [{"tests": [{"status": "expected", "results": ["not-an-object"]}]}]}]},
+            {"suites": [{"specs": [{"tests": [{"status": "expected", "results": None}]}]}]},
+            {"suites": [{"specs": [{"tests": [{"status": "expected", "results": [{"status": "passed", "retry": True}]}]}]}]},
         ]
         for report in reports:
             with self.subTest(report=report):