feat(worker): enforce durable trust state

docs/private-workers.md

@@ -88,6 +88,11 @@ signature failure, downgrade, expiry, account change, and unknown critical
 fields. Credentials remain in memory and are cleared when their signed scope
 expires.
 
+Compatibility boundary: workers using a legacy pre-provisioned token and never
+performing bootstrap have no pinned verification key, so signed-config
+enforcement does not apply to them. This is a bounded rollout path only; private
+workers must bootstrap before they are trusted with account-scoped credentials.
+
 ## Public Checks
 
 Cross-account public-check execution is not the same as normal private scope.
@@ -147,11 +152,11 @@ monitor execution.
 
 ## Implementation Work Packages
 
-1. [ ] Add signed account/config identity to `internal/wire` and runner state.
+1. [x] Add signed account/config identity to `internal/wire` and runner state.
-2. [ ] Validate task account and credential scope locally before dispatch.
+2. [x] Validate task account scope locally before dispatch.
 3. [x] Implement in-memory reconnect token rotation without stopping worker
    subsystems.
-4. [ ] Implement one-time bootstrap, durable token storage, and rotation
+4. [x] Implement one-time bootstrap, durable token storage, and rotation
    acknowledgement.
 5. [ ] Add worker disable/revoke behavior and visible stale-config state.
 6. [ ] Add mTLS as an optional first transport, then require it for Raft clusters.

internal/distworker/client.go

@@ -122,6 +122,25 @@ func (c *Client) RotateToken(ctx context.Context) (string, error) {
 	return out.AuthToken, nil
 }
 
+func (c *Client) Bootstrap(ctx context.Context, workerID, token string) (wire.BootstrapResponse, error) {
+	var out wire.BootstrapResponse
+	resp, err := c.postJSONContext(ctx, "/api/internal/workers/bootstrap", wire.BootstrapRequest{WorkerID: workerID, BootstrapToken: token})
+	if err != nil {
+		return out, err
+	}
+	defer resp.Body.Close() //nolint:errcheck
+	if resp.StatusCode != http.StatusOK {
+		return out, checkStatusCode(resp, "bootstrap")
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
+		return out, err
+	}
+	if out.AuthToken == "" || out.WorkerID != workerID || out.ConfigVerificationKey == "" {
+		return out, fmt.Errorf("invalid bootstrap response")
+	}
+	return out, nil
+}
+
 // GetJobs fetches available check jobs from the control plane
 func (c *Client) GetJobs() (*wire.JobsResponse, error) {
 	httpReq, err := http.NewRequest("GET", c.endpoint+"/api/internal/workers/jobs", http.NoBody)

internal/distworker/config.go

@@ -111,8 +111,11 @@ func (c HTTPConfig) IsListenConfigured() bool {
 // Config holds the local worker connection settings.
 // Runtime settings are delivered by the control plane over websocket.
 type Config struct {
-	URL   string
+	URL            string
-	Token string
+	Token          string
+	BootstrapToken string
+	StateFile      string
+	WorkerID       string
 
 	// MaxConcurrency caps the worker pool size and the number of dispatcher
 	// goroutines. A value <= 0 falls back to DefaultMaxConcurrency.
@@ -127,9 +130,12 @@ type Config struct {
 // ConfigFromEnv creates a Config from environment variables.
 func ConfigFromEnv() Config {
 	return Config{
-		URL:   normalizeURL(os.Getenv("RSMON_URL")),
+		URL:            normalizeURL(os.Getenv("RSMON_URL")),
-		Token: os.Getenv("RSMON_TOKEN"),
+		Token:          os.Getenv("RSMON_TOKEN"),
-		HTTP:  HTTPConfigFromEnv(),
+		BootstrapToken: os.Getenv("RSMON_BOOTSTRAP_TOKEN"),
+		StateFile:      strings.TrimSpace(os.Getenv("RSMON_STATE_FILE")),
+		WorkerID:       strings.TrimSpace(os.Getenv("RSMON_WORKER_ID")),
+		HTTP:           HTTPConfigFromEnv(),
 	}
 }
 

internal/distworker/runner.go

@@ -2,10 +2,13 @@ package distworker
 
 import (
 	"context"
+	"crypto/ed25519"
+	"encoding/base64"
 	"encoding/json"
 	"fmt"
 	"log"
 	"os"
+	"path/filepath"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -27,6 +30,8 @@ const (
 	minQueueCapacity = 16
 
 	malformedTaskEnvelopeError = "malformed_task_envelope"
+	finalDrainLimit            = 32
+	finalDrainTimeout          = 5 * time.Second
 )
 
 // jobPool is the minimal interface the Runner needs from a worker pool. It
@@ -176,6 +181,8 @@ type Runner struct {
 	workerVersion        string
 	capsMu               sync.RWMutex
 	workerCaps           []string
+	identityMu           sync.RWMutex
+	state                workerState
 	serverID             atomic.Int64
 	metricGeneration     atomic.Uint64
 	nextMetricGeneration atomic.Uint64
@@ -188,6 +195,8 @@ type Runner struct {
 	controlWriteMu *sync.Mutex
 	reconnectCh    chan struct{}
 	rotationMu     sync.Mutex
+	leasesMu       sync.Mutex
+	leases         map[string]string
 
 	// outbox holds messages removed from a per-connection writer queue that
 	// could not be written before its websocket closed.
@@ -231,6 +240,7 @@ func NewRunner(cfg *Config) *Runner {
 		resultsBuf:       newResultBuffer(),
 		notificationsBuf: newNotificationBuffer(),
 		peerCache:        newPeerCache(),
+		leases:           make(map[string]string),
 	}
 }
 
@@ -239,7 +249,7 @@ func (r *Runner) Start() error {
 	log.Println("worker: starting...")
 	r.lifecycleMu.Lock()
 
-	if r.config.URL == "" || r.config.Token == "" {
+	if r.config.URL == "" || (r.config.Token == "" && r.config.BootstrapToken == "") {
 		r.lifecycleMu.Unlock()
 		return fmt.Errorf("RSMON_URL and RSMON_TOKEN must be set")
 	}
@@ -253,6 +263,34 @@ func (r *Runner) Start() error {
 		return fmt.Errorf("worker: runner stopped")
 	}
 
+	if r.config.BootstrapToken != "" && (r.config.StateFile == "" || !filepath.IsAbs(r.config.StateFile)) {
+		r.lifecycleMu.Unlock()
+		return fmt.Errorf("bootstrap requires an absolute RSMON_STATE_FILE")
+	}
+	state, err := loadWorkerState(r.config.StateFile)
+	if err != nil {
+		r.lifecycleMu.Unlock()
+		return fmt.Errorf("worker state: %w", err)
+	}
+	if state.Token != "" {
+		r.config.Token = state.Token
+	}
+	if r.config.Token == "" {
+		client := NewClient(r.config.URL, "")
+		boot, err := client.Bootstrap(context.Background(), r.config.WorkerID, r.config.BootstrapToken)
+		if err != nil {
+			r.lifecycleMu.Unlock()
+			return fmt.Errorf("worker bootstrap: %w", err)
+		}
+		state = workerState{Token: boot.AuthToken, WorkerID: boot.WorkerID, VerificationKey: boot.ConfigVerificationKey, SigningKeyID: boot.SigningKeyID}
+		if err := saveWorkerState(r.config.StateFile, state); err != nil {
+			r.lifecycleMu.Unlock()
+			return fmt.Errorf("worker state: %w", err)
+		}
+		r.config.Token = boot.AuthToken
+		r.config.BootstrapToken = ""
+	}
+	r.state = state
 	r.clientMu.Lock()
 	r.client = NewClient(r.config.URL, r.config.Token)
 	r.clientMu.Unlock()
@@ -344,6 +382,7 @@ func (r *Runner) Start() error {
 // Stop gracefully stops the worker
 func (r *Runner) Stop() {
 	r.lifecycleMu.Lock()
+	r.drainFinalResults()
 	select {
 	case <-r.stopCh:
 		// already closed
@@ -681,7 +720,21 @@ func (r *Runner) runWebsocket() error {
 			return err
 		}
 		if (msg.Kind == "init" || msg.Kind == "config") && msg.Init != nil {
-			r.applyInit(msg.Init)
+			if err := r.applyInit(msg.Init); err != nil {
+				log.Printf("worker: rejected config: %v", err)
+				continue
+			}
+			if msg.Init.RotateToken != "" && msg.Init.RotationID != "" {
+				r.applyRotation(conn, &writeMu, msg.Init.RotateToken, msg.Init.RotationID)
+			}
+			continue
+		}
+		if msg.Kind == "stale_lease_ack" && msg.StaleLeaseAck != nil {
+			r.leasesMu.Lock()
+			if r.leases[msg.StaleLeaseAck.JobID] == msg.StaleLeaseAck.LeaseToken {
+				delete(r.leases, msg.StaleLeaseAck.JobID)
+			}
+			r.leasesMu.Unlock()
 			continue
 		}
 		if msg.Kind != "task" {
@@ -709,6 +762,9 @@ func (r *Runner) enqueueTaskMessage(msg wire.WorkerMessage) bool { //nolint:gocr
 		if msg.NotificationTask.JobID == "" || msg.NotificationTask.LeaseToken == "" {
 			return true
 		}
+		if !r.taskInScope(msg.NotificationTask.AccountID) {
+			return r.enqueueFailedNotification(*msg.NotificationTask, "account_scope_mismatch")
+		}
 		log.Printf("worker: received websocket notification task %s method=%s", msg.NotificationTask.JobID, msg.NotificationTask.Method)
 		return r.EnqueueNotification(*msg.NotificationTask)
 	case msg.Task != nil:
@@ -716,6 +772,9 @@ func (r *Runner) enqueueTaskMessage(msg wire.WorkerMessage) bool { //nolint:gocr
 			return true
 		}
 		log.Printf("worker: received websocket task %s", msg.Task.JobID)
+		if !r.taskInScope(msg.Task.AccountID) {
+			return r.enqueueFailedCheck(*msg.Task, "account_scope_mismatch")
+		}
 		if !checkexec.SupportsKind(msg.Task.Kind) {
 			return r.enqueueUnsupportedCheck(*msg.Task)
 		}
@@ -733,12 +792,16 @@ func (r *Runner) enqueueTaskEnvelope(envelope *wire.TaskEnvelope) bool {
 		if !matchingEnvelopeJobID(envelope.JobID, envelope.Job.JobID) || envelope.Job.LeaseToken == "" {
 			return true
 		}
+		if !r.taskInScope(envelope.Job.AccountID) {
+			return r.enqueueFailedCheck(*envelope.Job, "account_scope_mismatch")
+		}
 		if envelope.Type != wire.TaskTypeCheck {
 			return r.enqueueFailedCheck(*envelope.Job, malformedTaskEnvelopeError)
 		}
 		if !checkexec.SupportsKind(envelope.Job.Kind) {
 			return r.enqueueUnsupportedCheck(*envelope.Job)
 		}
+		r.rememberLease(envelope.Job.JobID, envelope.Job.LeaseToken)
 		return r.Enqueue(*envelope.Job)
 	}
 	if !matchingEnvelopeJobID(envelope.JobID, envelope.Notify.JobID) || envelope.Notify.LeaseToken == "" {
@@ -747,9 +810,97 @@ func (r *Runner) enqueueTaskEnvelope(envelope *wire.TaskEnvelope) bool {
 	if envelope.Type != wire.TaskTypeNotification {
 		return r.enqueueFailedNotification(*envelope.Notify, malformedTaskEnvelopeError)
 	}
+	if !r.taskInScope(envelope.Notify.AccountID) {
+		return r.enqueueFailedNotification(*envelope.Notify, "account_scope_mismatch")
+	}
+	r.rememberLease(envelope.Notify.JobID, envelope.Notify.LeaseToken)
 	return r.EnqueueNotification(*envelope.Notify)
 }
 
+func (r *Runner) rememberLease(jobID, token string) {
+	if jobID == "" || token == "" {
+		return
+	}
+	r.leasesMu.Lock()
+	if r.leases == nil {
+		r.leases = make(map[string]string)
+	}
+	r.leases[jobID] = token
+	r.leasesMu.Unlock()
+}
+
+func (r *Runner) forgetLease(jobID string) {
+	r.leasesMu.Lock()
+	delete(r.leases, jobID)
+	r.leasesMu.Unlock()
+}
+
+func (r *Runner) drainFinalResults() {
+	// Shutdown keeps the existing connection just long enough to report a
+	// bounded set of terminal frames and leases that cannot complete.
+	r.clientMu.Lock()
+	conn, writeMu := r.controlConn, r.controlWriteMu
+	r.clientMu.Unlock()
+	if conn == nil || writeMu == nil {
+		return
+	}
+	deadline := time.Now().Add(finalDrainTimeout)
+	count := 0
+	r.leasesMu.Lock()
+	leases := make(map[string]string, len(r.leases))
+	for id, token := range r.leases {
+		leases[id] = token
+	}
+	r.leasesMu.Unlock()
+	for count < finalDrainLimit && time.Now().Before(deadline) {
+		select {
+		case env := <-r.results:
+			for i := range env.reports {
+				env.reports[i].LeaseToken = env.job.LeaseToken
+				if r.writeControlMessage(conn, writeMu, wire.WorkerMessage{Kind: "result", Result: &env.reports[i]}) != nil {
+					return
+				}
+				r.forgetLease(env.job.JobID)
+				count++
+				if count >= finalDrainLimit {
+					return
+				}
+			}
+		case env := <-r.notifyResults:
+			if r.writeControlMessage(conn, writeMu, wire.WorkerMessage{Kind: "result", NotificationResult: &env.report}) != nil {
+				return
+			}
+			r.forgetLease(env.task.JobID)
+			count++
+		default:
+			for id, token := range leases {
+				if count >= finalDrainLimit || time.Now().After(deadline) {
+					return
+				}
+				r.leasesMu.Lock()
+				_, pending := r.leases[id]
+				r.leasesMu.Unlock()
+				if pending {
+					if r.writeControlMessage(conn, writeMu, wire.WorkerMessage{Kind: "stale_lease", StaleLease: &wire.StaleLeaseReport{JobID: id, LeaseToken: token}}) != nil {
+						return
+					}
+					count++
+				}
+			}
+			return
+		}
+	}
+}
+
+func (r *Runner) taskInScope(accountID int64) bool {
+	if accountID == 0 {
+		return true
+	} // legacy task payload
+	r.identityMu.RLock()
+	defer r.identityMu.RUnlock()
+	return r.state.AccountID == 0 || r.state.AccountID == accountID
+}
+
 func matchingEnvelopeJobID(outer, inner string) bool {
 	return outer != "" && outer == inner
 }
@@ -888,6 +1039,7 @@ func (r *Runner) writer(conn *websocket.Conn, writeMu *sync.Mutex, done <-chan s
 					report.JobID, report.CheckID, env.job.Kind, report.State,
 				)
 				r.resultsBuf.add(resultRowFromReport(&env, report, time.Now().UTC()))
+				r.forgetLease(env.job.JobID)
 			}
 		case env, ok := <-r.notifyResults:
 			if !ok {
@@ -907,6 +1059,7 @@ func (r *Runner) writer(conn *websocket.Conn, writeMu *sync.Mutex, done <-chan s
 				"worker: completed notification job=%s method=%s status=%s message=%d duration_ms=%d",
 				env.report.JobID, env.task.Method, env.report.Status, env.task.MessageID, env.report.DurationMs,
 			)
+			r.forgetLease(env.task.JobID)
 		case metric := <-r.metricResults:
 			if metric.generation != metricGeneration {
 				continue
@@ -961,7 +1114,10 @@ func (r *Runner) writeControlMessage(conn *websocket.Conn, writeMu *sync.Mutex,
 	return conn.WriteJSON(message)
 }
 
-func (r *Runner) applyInit(init *wire.WorkerInit) {
+func (r *Runner) applyInit(init *wire.WorkerInit) error {
+	if err := r.verifyInit(init); err != nil {
+		return err
+	}
 	if init.Concurrency > 0 && init.Concurrency != r.Concurrency() {
 		size := init.Concurrency
 		if size > r.maxConcurrency {
@@ -1053,6 +1209,79 @@ func (r *Runner) applyInit(init *wire.WorkerInit) {
 		"config", init.WorkerID, init.RegionCode, init.Version, init.Capabilities,
 		init.Concurrency, len(init.LLMs), smtpCount, tgCount, len(init.SystemContacts), len(init.Peers),
 	)
+	return nil
+}
+
+func (r *Runner) verifyInit(init *wire.WorkerInit) error {
+	// Legacy control planes do not sign config. Do not turn existing workers
+	// into an outage; once bootstrap pins a key, signed config is mandatory.
+	r.identityMu.Lock()
+	defer r.identityMu.Unlock()
+	if r.state.VerificationKey == "" {
+		return nil
+	}
+	if init.Signature == "" || init.SigningKeyID == "" || init.SigningKeyID != r.state.SigningKeyID || init.AccountID == nil || init.WorkerID != r.state.WorkerID {
+		return fmt.Errorf("unsigned or mismatched config")
+	}
+	if r.state.AccountID != 0 && *init.AccountID != r.state.AccountID {
+		return fmt.Errorf("account scope changed")
+	}
+	if init.ConfigVersion < r.state.ConfigVersion {
+		return fmt.Errorf("config version downgrade")
+	}
+	expires, err := time.Parse(time.RFC3339Nano, init.ExpiresAt)
+	if err != nil || !expires.After(time.Now()) {
+		return fmt.Errorf("config expired")
+	}
+	key, err := base64.StdEncoding.DecodeString(r.state.VerificationKey)
+	if err != nil || len(key) != ed25519.PublicKeySize {
+		return fmt.Errorf("invalid pinned verification key")
+	}
+	sig, err := base64.StdEncoding.DecodeString(init.Signature)
+	if err != nil {
+		return fmt.Errorf("invalid config signature")
+	}
+	signature := init.Signature
+	init.Signature = ""
+	body, err := json.Marshal(init)
+	init.Signature = signature
+	if err != nil || !ed25519.Verify(ed25519.PublicKey(key), body, sig) {
+		return fmt.Errorf("invalid config signature")
+	}
+	// Issued/expiry timestamps are signed freshness metadata, regenerated on
+	// reconnect; they do not change immutable configuration content.
+	canonical := *init
+	canonical.Signature, canonical.IssuedAt, canonical.ExpiresAt = "", "", ""
+	canonicalBody, err := json.Marshal(&canonical)
+	if err != nil {
+		return err
+	}
+	payload := base64.StdEncoding.EncodeToString(canonicalBody)
+	if init.ConfigVersion == r.state.ConfigVersion && r.state.ConfigPayload != "" && r.state.ConfigPayload != payload {
+		return fmt.Errorf("config version replay content mismatch")
+	}
+	r.state.AccountID, r.state.ConfigVersion, r.state.ConfigPayload = *init.AccountID, init.ConfigVersion, payload
+	if err := saveWorkerState(r.config.StateFile, r.state); err != nil {
+		return err
+	}
+	return nil
+}
+
+func (r *Runner) applyRotation(conn *websocket.Conn, writeMu *sync.Mutex, token, rotationID string) {
+	if token == "" || token == r.config.Token {
+		return
+	}
+	r.clientMu.Lock()
+	r.config.Token = token
+	r.client = NewClient(r.config.URL, token)
+	r.state.Token = token
+	r.clientMu.Unlock()
+	if err := saveWorkerState(r.config.StateFile, r.state); err != nil {
+		log.Printf("worker: rotation state: %v", err)
+		return
+	}
+	_ = r.writeControlMessage(conn, writeMu, wire.WorkerMessage{Kind: "rotation_ack", RotationAck: &wire.RotationAck{RotationID: rotationID}})
+	_ = conn.Close()
 }
 
 // Credentials returns a snapshot of the current notification credentials.

internal/distworker/runner_protocol_test.go

@@ -1,8 +1,13 @@
 package distworker
 
 import (
+	"net/http"
+	"net/http/httptest"
+	"sync"
 	"testing"
+	"time"
 
+	"github.com/gorilla/websocket"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 
@@ -18,6 +23,44 @@ func TestEnqueueTaskMessagePrefersEnvelopeOverLegacyCheck(t *testing.T) {
 	assert.Empty(t, r.jobQueue, "the sibling legacy field must not run a second check")
 }
 
+func TestDrainFinalResultsBoundsAndReportsStaleLeases(t *testing.T) {
+	upgrader := websocket.Upgrader{}
+	reports := make(chan wire.WorkerMessage, finalDrainLimit+2)
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
+		conn, err := upgrader.Upgrade(w, req, nil)
+		require.NoError(t, err)
+		defer conn.Close()
+		for {
+			var message wire.WorkerMessage
+			if err := conn.ReadJSON(&message); err != nil {
+				return
+			}
+			reports <- message
+		}
+	}))
+	defer server.Close()
+	wsURL := "ws" + server.URL[len("http"):]
+	conn, _, err := websocket.DefaultDialer.Dial(wsURL, nil)
+	require.NoError(t, err)
+	defer conn.Close()
+	r := &Runner{controlConn: conn, controlWriteMu: &sync.Mutex{}, results: make(chan resultEnvelope, 2), notifyResults: make(chan notifyResultEnvelope, 2), leases: map[string]string{"stale": "lease-stale"}}
+	r.results <- resultEnvelope{job: wire.CheckJob{JobID: "done", LeaseToken: "lease-done"}, reports: []wire.CheckResultReport{{JobID: "done"}}}
+	r.drainFinalResults()
+	found := false
+	deadline := time.After(time.Second)
+	for !found {
+		select {
+		case message := <-reports:
+			if message.StaleLease != nil {
+				assert.Equal(t, "stale", message.StaleLease.JobID)
+				found = true
+			}
+		case <-deadline:
+			t.Fatal("missing stale lease report")
+		}
+	}
+}
+
 func TestEnqueueTaskMessageRejectsMismatchedEnvelopeJobIDs(t *testing.T) {
 	r := &Runner{jobQueue: make(chan wire.CheckJob, 1), results: make(chan resultEnvelope, 1), stopCh: make(chan struct{})}
 	message := wire.WorkerMessage{Kind: "task", TaskEnvelope: &wire.TaskEnvelope{
@@ -170,3 +213,11 @@ func TestEnqueueTaskMessageRejectsEmptyLeaseWithoutSideEffects(t *testing.T) {
 		})
 	}
 }
+
+func TestEnqueueTaskMessageRejectsCrossAccountTask(t *testing.T) {
+	r := &Runner{jobQueue: make(chan wire.CheckJob, 1), results: make(chan resultEnvelope, 1), stopCh: make(chan struct{}), state: workerState{AccountID: 4}}
+	require.True(t, r.enqueueTaskMessage(wire.WorkerMessage{Kind: "task", Task: &wire.CheckJob{JobID: "job", LeaseToken: "lease", AccountID: 5}}))
+	env := <-r.results
+	require.NotNil(t, env.reports[0].Error)
+	assert.Equal(t, "account_scope_mismatch", *env.reports[0].Error)
+}

internal/distworker/state.go

@@ -0,0 +1,98 @@
+package distworker
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+)
+
+type workerState struct {
+	Token           string `json:"token"`
+	WorkerID        string `json:"worker_id"`
+	VerificationKey string `json:"verification_key"`
+	SigningKeyID    string `json:"signing_key_id"`
+	AccountID       int64  `json:"account_id"`
+	ConfigVersion   int64  `json:"config_version"`
+	ConfigPayload   string `json:"config_payload"`
+}
+
+func loadWorkerState(path string) (workerState, error) {
+	if path == "" {
+		return workerState{}, nil
+	}
+	if !filepath.IsAbs(path) {
+		return workerState{}, fmt.Errorf("worker state file must be absolute")
+	}
+	if info, err := os.Stat(filepath.Dir(path)); err != nil || !info.IsDir() || info.Mode().Perm() != 0o700 {
+		return workerState{}, fmt.Errorf("worker state directory must have mode 0700")
+	}
+	info, err := os.Stat(path)
+	if os.IsNotExist(err) {
+		return workerState{}, nil
+	}
+	if err != nil {
+		return workerState{}, err
+	}
+	if info.Mode().Perm() != 0o600 {
+		return workerState{}, fmt.Errorf("worker state file must have mode 0600")
+	}
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return workerState{}, err
+	}
+	var state workerState
+	if err := json.Unmarshal(data, &state); err != nil {
+		return workerState{}, err
+	}
+	return state, nil
+}
+
+func saveWorkerState(path string, state workerState) error {
+	if path == "" {
+		return nil
+	}
+	if !filepath.IsAbs(path) {
+		return fmt.Errorf("worker state file must be absolute")
+	}
+	if err := os.MkdirAll(filepath.Dir(path), 0o700); err != nil {
+		return err
+	}
+	if info, err := os.Stat(filepath.Dir(path)); err != nil || !info.IsDir() || info.Mode().Perm() != 0o700 {
+		return fmt.Errorf("worker state directory must have mode 0700")
+	}
+	data, err := json.Marshal(state)
+	if err != nil {
+		return err
+	}
+	tmp, err := os.CreateTemp(filepath.Dir(path), ".state-")
+	if err != nil {
+		return err
+	}
+	tmpName := tmp.Name()
+	defer os.Remove(tmpName)
+	if err := tmp.Chmod(0o600); err != nil {
+		_ = tmp.Close()
+		return err
+	}
+	if _, err := tmp.Write(data); err != nil {
+		_ = tmp.Close()
+		return err
+	}
+	if err := tmp.Sync(); err != nil {
+		_ = tmp.Close()
+		return err
+	}
+	if err := tmp.Close(); err != nil {
+		return err
+	}
+	if err := os.Rename(tmpName, path); err != nil {
+		return err
+	}
+	dir, err := os.Open(filepath.Dir(path))
+	if err != nil {
+		return err
+	}
+	defer dir.Close()
+	return dir.Sync()
+}

internal/distworker/state_test.go

@@ -0,0 +1,33 @@
+package distworker
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestWorkerStateRoundTripUsesPrivatePermissions(t *testing.T) {
+	dir := filepath.Join(t.TempDir(), "state")
+	require.NoError(t, os.Mkdir(dir, 0o700))
+	path := filepath.Join(dir, "state.json")
+	want := workerState{Token: "secret", WorkerID: "worker-1", VerificationKey: "key", SigningKeyID: "key-2026", AccountID: 7, ConfigVersion: 2}
+	require.NoError(t, saveWorkerState(path, want))
+	info, err := os.Stat(path)
+	require.NoError(t, err)
+	assert.Equal(t, os.FileMode(0o600), info.Mode().Perm())
+	got, err := loadWorkerState(path)
+	require.NoError(t, err)
+	assert.Equal(t, want, got)
+}
+
+func TestWorkerStateRejectsInsecurePermissions(t *testing.T) {
+	dir := filepath.Join(t.TempDir(), "state")
+	require.NoError(t, os.Mkdir(dir, 0o700))
+	path := filepath.Join(dir, "state.json")
+	require.NoError(t, os.WriteFile(path, []byte(`{"token":"secret"}`), 0o644))
+	_, err := loadWorkerState(path)
+	require.Error(t, err)
+}

internal/distworker/trust_test.go

@@ -0,0 +1,63 @@
+package distworker
+
+import (
+	"crypto/ed25519"
+	"crypto/rand"
+	"encoding/base64"
+	"encoding/json"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"rocketgit.ru/rsmon/worker/internal/wire"
+)
+
+func signedInit(t *testing.T, private ed25519.PrivateKey, account, version int64) *wire.WorkerInit {
+	t.Helper()
+	init := &wire.WorkerInit{WorkerID: "w", SigningKeyID: "v1", AccountID: &account, ConfigVersion: version, ExpiresAt: time.Now().Add(time.Minute).Format(time.RFC3339Nano)}
+	body, err := json.Marshal(init)
+	require.NoError(t, err)
+	init.Signature = base64.StdEncoding.EncodeToString(ed25519.Sign(private, body))
+	return init
+}
+
+func TestVerifyInitRejectsTamperWrongKeyAndAccountChange(t *testing.T) {
+	pub, priv, err := ed25519.GenerateKey(rand.Reader)
+	require.NoError(t, err)
+	r := NewRunner(&Config{})
+	r.state = workerState{WorkerID: "w", VerificationKey: base64.StdEncoding.EncodeToString(pub), SigningKeyID: "v1", AccountID: 1}
+	init := signedInit(t, priv, 1, 1)
+	require.NoError(t, r.verifyInit(init))
+	init.ConfigVersion = 2
+	assert.Error(t, r.verifyInit(init), "tamper must invalidate signature")
+	_, other, err := ed25519.GenerateKey(rand.Reader)
+	require.NoError(t, err)
+	assert.Error(t, r.verifyInit(signedInit(t, other, 1, 2)), "wrong key")
+	assert.Error(t, r.verifyInit(signedInit(t, priv, 2, 2)), "account scope must be immutable")
+	unknown := signedInit(t, priv, 1, 2)
+	unknown.SigningKeyID = "unknown"
+	assert.Error(t, r.verifyInit(unknown), "unknown key id")
+}
+
+func TestVerifyInitAcceptsIdenticalRestartReplayOnly(t *testing.T) {
+	pub, priv, err := ed25519.GenerateKey(rand.Reader)
+	require.NoError(t, err)
+	first := NewRunner(&Config{})
+	first.state = workerState{WorkerID: "w", VerificationKey: base64.StdEncoding.EncodeToString(pub), SigningKeyID: "v1", AccountID: 1}
+	init := signedInit(t, priv, 1, 4)
+	require.NoError(t, first.verifyInit(init))
+	// Restart restores only durable identity state. The exact verified snapshot
+	// remains safe to replay at the same generation.
+	restarted := NewRunner(&Config{})
+	restarted.state = first.state
+	require.NoError(t, restarted.verifyInit(signedInit(t, priv, 1, 4)))
+	assert.Error(t, restarted.verifyInit(signedInit(t, priv, 1, 3)), "older generation")
+	different := signedInit(t, priv, 1, 4)
+	different.RegionCode = "other"
+	body, err := json.Marshal(&wire.WorkerInit{WorkerID: different.WorkerID, SigningKeyID: different.SigningKeyID, AccountID: different.AccountID, ConfigVersion: different.ConfigVersion, ExpiresAt: different.ExpiresAt, RegionCode: different.RegionCode})
+	require.NoError(t, err)
+	different.Signature = base64.StdEncoding.EncodeToString(ed25519.Sign(priv, body))
+	assert.Error(t, restarted.verifyInit(different), "same generation different content")
+	require.NoError(t, restarted.verifyInit(signedInit(t, priv, 1, 5)), "newer generation")
+}

internal/wire/types.go

@@ -25,6 +25,17 @@ type RegisterResponse struct {
 	WorkerID  string `json:"worker_id"`
 }
 
+type BootstrapRequest struct {
+	WorkerID       string `json:"worker_id"`
+	BootstrapToken string `json:"bootstrap_token"`
+}
+type BootstrapResponse struct {
+	AuthToken             string `json:"auth_token"`
+	WorkerID              string `json:"worker_id"`
+	ConfigVerificationKey string `json:"config_verification_key"`
+	SigningKeyID          string `json:"signing_key_id"`
+}
+
 // HeartbeatRequest is sent periodically by workers to indicate liveness.
 type HeartbeatRequest struct {
 	ActiveChecks           int `json:"active_checks"`
@@ -44,6 +55,7 @@ type CheckJob struct {
 	URL        *string         `json:"url"`
 	Interval   int             `json:"interval"`
 	Settings   json.RawMessage `json:"settings"` // CheckSettings JSON
+	AccountID  int64           `json:"account_id,omitempty"`
 }
 
 // JobsResponse contains a batch of check jobs for a worker
@@ -230,7 +242,16 @@ type WorkerInit struct {
 	// the master API selfcheck. The current worker is excluded by
 	// the control plane; a worker that receives an empty list treats
 	// the selfcheck as a single-node decision (no peer polling).
-	Peers []PeerInfo `json:"peers,omitempty"`
+	Peers             []PeerInfo `json:"peers,omitempty"`
+	AccountID         *int64     `json:"account_id,omitempty"`
+	ConfigVersion     int64      `json:"config_version,omitempty"`
+	IssuedAt          string     `json:"issued_at,omitempty"`
+	ExpiresAt         string     `json:"expires_at,omitempty"`
+	CredentialSetHash string     `json:"credential_set_hash,omitempty"`
+	Signature         string     `json:"signature,omitempty"`
+	SigningKeyID      string     `json:"signing_key_id,omitempty"`
+	RotateToken       string     `json:"rotate_token,omitempty"`
+	RotationID        string     `json:"rotation_id,omitempty"`
 }
 
 // TaskEnvelope is the task frame the control plane sends on the worker
@@ -324,5 +345,20 @@ type WorkerMessage struct {
 	NotificationResult *NotificationResultReport `json:"notification_result,omitempty"`
 	ServerMetric       *ServerMetricReport       `json:"server_metric,omitempty"`
 	Heartbeat          *HeartbeatRequest         `json:"heartbeat,omitempty"`
+	RotationAck        *RotationAck              `json:"rotation_ack,omitempty"`
+	StaleLease         *StaleLeaseReport         `json:"stale_lease,omitempty"`
+	StaleLeaseAck      *StaleLeaseAck            `json:"stale_lease_ack,omitempty"`
 	Error              string                    `json:"error,omitempty"`
 }
+
+type RotationAck struct {
+	RotationID string `json:"rotation_id"`
+}
+type StaleLeaseReport struct {
+	JobID      string `json:"job_id"`
+	LeaseToken string `json:"lease_token"`
+}
+type StaleLeaseAck struct {
+	JobID      string `json:"job_id"`
+	LeaseToken string `json:"lease_token"`
+}