v0.32.0: self-serve decommission endpoint + intent-aware re-assert (B2a)

POST /disks/decommission mirrors eject (withGuest, user-data role gate) — no
operator signature, non-destructive (never formats): sets IntentDecommissioned,
prunes the GuestBindStore entry, unmounts. ReassertGuestBinds is now intent-aware
(skip non-enrolled) so a decommissioned-but-present drive never auto-rebinds on
agent restart — the load-bearing F9-reconnect fix. GuestBindStore.Remove added.
Operator-signed DecommissionExecutor + classify untouched. Non-hollow tests incl.
the intent-aware reassert companion (mutation-proven to fail on intent-blind code).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CHANGELOG.md

@@ -3,6 +3,33 @@
 All notable changes to **felhom-agent** are recorded here. Update on every code
 change that gets pushed.
 
+## v0.32.0 — self-serve decommission + intent-aware re-assert (B2a) (2026-06-14)
+
+Customer-self-serve storage decommission (no operator signature; non-destructive — never formats),
+plus the load-bearing fix that keeps a decommissioned drive from auto-rebinding into the guest.
+
+- **`POST /disks/decommission`** (`internal/localapi/disks.go` `handleDiskDecommission`, route in
+  `server.go`) — mirrors `handleDiskEject` exactly: `withGuest` self-scoping, `scopedFromBody`, and the
+  same **user-data role gate** (`roleForMountPath` must be `RoleUserData`, else 403; fail-safe-to-
+  protected on ambiguity) so a compromised controller can't decommission system/backup storage. It
+  records a PERMANENT `IntentDecommissioned`, prunes the `GuestBindStore` entry (hygiene), and unmounts
+  (so the drive is physically removable). It **NEVER** calls any format/mkfs path — the data stays on
+  the drive. The operator-signed `DecommissionExecutor` + `reconcile.Classify` classification are
+  untouched (the absent-drive/DR route).
+- **`ReassertGuestBinds` is now intent-aware** (THE correctness fix): the startup re-assert skips any
+  durable-id whose intent is not `enrolled`, so a decommissioned- (or ejected-) but-still-present drive
+  is never auto-rebound into the guest on agent restart. A nil intent store falls back to legacy
+  bind-all (matching the watchdog's nil-intent rule). Covers both the self-serve and the operator-
+  signed decommission paths (both land on `IntentDecommissioned`).
+- **`GuestBindStore.Remove(vmid, durableID)`** (`internal/localapi/guestbindstore.go`) — idempotent
+  (absent = no-op), atomic tmp+rename like `Record`; drops the vmid key when its set empties. Re-enroll
+  re-`Record`s via the existing `recordGuestBind` on guest-attach, so Remove doesn't break re-commission.
+- `IntentRecorder` extended with `SetDecommissioned` + `Get` (both already on `*storage.IntentStore`).
+- Non-hollow tests (`internal/localapi/decommission_test.go`): role-gate refuses system/backup (403,
+  no unmount); decommission sets intent + removes the bind + unmounts + never formats; intent-aware
+  re-assert does NOT rebind a decommissioned-but-present drive (companion: enrolled DOES rebind; the
+  intent-blind pre-fix code fails this); re-commission re-records; `Remove` idempotency + persistence.
+
 ## v0.31.0 — live-drive F9 + F20-BUG2 + F20-BUG3 (disk bind/wipe) (2026-06-14)
 
 The last live-drive findings, all disk/`localapi`-side, implemented + deployed on `felhom-pve` and

cmd/felhom-agent/main.go

@@ -35,38 +35,38 @@ import (
 	applog "gitea.dooplex.hu/admin/felhom-agent/internal/log"
 	"gitea.dooplex.hu/admin/felhom-agent/internal/pbs"
 	"gitea.dooplex.hu/admin/felhom-agent/internal/provision"
-	"gitea.dooplex.hu/admin/felhom-agent/internal/signedjobs"
 	"gitea.dooplex.hu/admin/felhom-agent/internal/proxmox"
 	"gitea.dooplex.hu/admin/felhom-agent/internal/reconcile"
+	"gitea.dooplex.hu/admin/felhom-agent/internal/signedjobs"
 	"gitea.dooplex.hu/admin/felhom-agent/internal/storage"
 )
 
 // version is the agent version. Overridable at build time with
 // -ldflags "-X main.version=<v>"; defaults to the in-repo CHANGELOG version.
-var version = "0.31.0"
+var version = "0.32.0"
 
 func main() {
 	var (
-		cfgPath     string
-		selftest    selftestFlag
-		vmid        int
-		watch       time.Duration
-		archive     string
-		mode        string
-		hostname    string
-		keep        bool
-		rootfsGrow  int
-		dataVolGrow int
-		dataVolMount string
-		pbsStorage  string
-		paperkey    bool
-		offline     bool
-		upload      bool
-		custID       string
-		custDomain   string
-		custName     string
-		custEmail    string
-		hubPassword  string
+		cfgPath       string
+		selftest      selftestFlag
+		vmid          int
+		watch         time.Duration
+		archive       string
+		mode          string
+		hostname      string
+		keep          bool
+		rootfsGrow    int
+		dataVolGrow   int
+		dataVolMount  string
+		pbsStorage    string
+		paperkey      bool
+		offline       bool
+		upload        bool
+		custID        string
+		custDomain    string
+		custName      string
+		custEmail     string
+		hubPassword   string
 		blobPath      string
 		expectedFP    string
 		keyDest       string
@@ -490,7 +490,7 @@ func runDaemon(cfg config.Config, logger *slog.Logger) int {
 	}
 
 	err = <-errc
-	stop()                                            // tear down the siblings on the first exit
+	stop()                                           // tear down the siblings on the first exit
 	for i := 0; i < 4+localServers+lanServers; i++ { // wait for the other goroutines
 		<-errc
 	}
@@ -1087,8 +1087,8 @@ type provisionArgs struct {
 	archive     string
 	vmid        int
 	hostname    string
-	customerID  string // baked into bootstrap (the hub config-pull target)
-	hubPassword string // the customer's hub retrieval passphrase (SECRET) — baked into bootstrap
+	customerID  string        // baked into bootstrap (the hub config-pull target)
+	hubPassword string        // the customer's hub retrieval passphrase (SECRET) — baked into bootstrap
 	sizing      bringUpSizing // OS-rootfs / Docker-data sizing for the bring-up half
 }
 
@@ -1352,7 +1352,7 @@ func runSelftestEscrowConsume(ctx context.Context, logger *slog.Logger, blobPath
 	logger.Info("escrow: consuming R-wrapped escrow (Unwrap → fingerprint-gate → install)",
 		"blob_bytes", len(blob), "key_dest", keyDest) // R is NOT logged
 	if err := escrow.Consume(ctx, blob, R, expectedFP, keyDest); err != nil {
-		R = "" // drop the reference
+		R = ""                                            // drop the reference
 		fmt.Fprintln(os.Stderr, "  [FAIL] consume:", err) // the error never contains R or key bytes
 		return 1
 	}
@@ -1410,7 +1410,7 @@ type escrowUploadRequest struct {
 	// Slice 10D.1 — optional DR bundle (identity escrow + non-secret directive). Omitted in slice-7.
 	IdentityBlobB64 string          `json:"identity_blob_b64,omitempty"`
 	DirectiveJSON   json.RawMessage `json:"directive,omitempty"`
-	CreatedAt      string `json:"created_at"`      // RFC3339
+	CreatedAt       string          `json:"created_at"` // RFC3339
 }
 
 // uploadEscrowBlob PUTs the opaque blob (and, for 10D, the identity blob + non-secret directive) to

internal/localapi/decommission_test.go

@@ -0,0 +1,222 @@
+package localapi
+
+import (
+	"context"
+	"io"
+	"log/slog"
+	"net/http"
+	"path/filepath"
+	"sync"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-agent/internal/hub"
+	"gitea.dooplex.hu/admin/felhom-agent/internal/storage"
+)
+
+// fakeIntent implements the extended IntentRecorder (SetEnrolled/SetEjected/SetDecommissioned + Get).
+type fakeIntent struct {
+	mu sync.Mutex
+	m  map[string]storage.DriveIntent
+}
+
+func newFakeIntent() *fakeIntent { return &fakeIntent{m: map[string]storage.DriveIntent{}} }
+func (f *fakeIntent) set(id string, v storage.DriveIntent) {
+	f.mu.Lock()
+	f.m[id] = v
+	f.mu.Unlock()
+}
+func (f *fakeIntent) SetEnrolled(id string) error { f.set(id, storage.IntentEnrolled); return nil }
+func (f *fakeIntent) SetEjected(id string) error  { f.set(id, storage.IntentEjected); return nil }
+func (f *fakeIntent) SetDecommissioned(id string) error {
+	f.set(id, storage.IntentDecommissioned)
+	return nil
+}
+func (f *fakeIntent) Get(id string) storage.DriveIntent {
+	f.mu.Lock()
+	defer f.mu.Unlock()
+	return f.m[id]
+}
+
+// decommServer wires a Server with Disks + Intent + GuestBinds + GuestAttach so both the endpoint and
+// the intent-aware re-assert can be exercised. Returns the *Server (use .Handler() for HTTP tests).
+func decommServer(t *testing.T, d *fakeDiskOps, sv StorageView, gl GuestLister, intent IntentRecorder, gb *GuestBindStore, ga GuestAttacher, mounts map[int]map[string]string) *Server {
+	t.Helper()
+	srv, err := NewServer(Options{
+		ListenAddr:  "127.0.0.1:0",
+		Guests:      &fakeGuestsCfg{mounts: mounts},
+		Backups:     &fakeBackups{},
+		Store:       &fakeStore{},
+		Storage:     sv,
+		Tokens:      staticTokens{"A": 8200, "B": 9300},
+		Disks:       d,
+		DiskGate:    &fakeGate{},
+		Guests2:     gl,
+		Intent:      intent,
+		GuestBinds:  gb,
+		GuestAttach: ga,
+		HostReader:  sysOnSDA(),
+		Logger:      slog.New(slog.NewTextHandler(io.Discard, nil)),
+	})
+	if err != nil {
+		t.Fatalf("new server: %v", err)
+	}
+	srv.baseCtx = context.Background()
+	return srv
+}
+
+// userDataAndProtected is a storage view with a user-data USB, a system dir, and a backup PBS mount.
+func userDataAndProtected() fakeStorage {
+	return fakeStorage{targets: []hub.StorageTarget{
+		{Name: "bulk", Type: hub.StorageTypeUSB, BackingDevice: "/dev/sdb1", MountPath: "/mnt/bulk", DurableID: "uuid:usb-1"},
+		{Name: "local", Type: "local", MountPath: "/var/lib/vz"},
+		{Name: "felhom-pbs", Type: hub.StorageTypePBS, MountPath: "/mnt/pbs"},
+	}}
+}
+
+// TestDecommission_RoleGated: like eject, a system/backup mount is refused 403 with NO unmount; only a
+// user-data mount decommissions. (Mirrors TestEject_RoleGated — the gate is the security control.)
+func TestDecommission_RoleGated(t *testing.T) {
+	for _, where := range []string{"/var/lib/vz", "/mnt/pbs", "/mnt/unknown"} {
+		d := &fakeDiskOps{}
+		srv := decommServer(t, d, userDataAndProtected(), fakeGuestList{}, newFakeIntent(), tempBindStore(t), &fakeGuestAttacher{}, nil)
+		w := do(t, srv.Handler(), "POST", "/disks/decommission", "A", `{"where":"`+where+`"}`)
+		if w.Code != http.StatusForbidden {
+			t.Fatalf("decommission %s: got %d want 403 (%s)", where, w.Code, w.Body.String())
+		}
+		d.mu.Lock()
+		if len(d.unmountCalls) != 0 {
+			t.Fatalf("Unmount called on protected mount %s — role-gate bypassed", where)
+		}
+		d.mu.Unlock()
+	}
+}
+
+// TestDecommission_Effects: a user-data decommission sets IntentDecommissioned, removes the
+// GuestBindStore entry, and unmounts — assert ALL THREE (not just no-error). NEVER formats.
+func TestDecommission_Effects(t *testing.T) {
+	d := &fakeDiskOps{}
+	intent := newFakeIntent()
+	intent.SetEnrolled("uuid:usb-1") // currently enrolled
+	gb := tempBindStore(t)
+	_ = gb.Record(8200, "uuid:usb-1")
+
+	srv := decommServer(t, d, userDataAndProtected(), fakeGuestList{}, intent, gb, &fakeGuestAttacher{}, nil)
+	w := do(t, srv.Handler(), "POST", "/disks/decommission", "A", `{"where":"/mnt/bulk"}`)
+	if w.Code != http.StatusOK {
+		t.Fatalf("decommission user-data: got %d want 200 (%s)", w.Code, w.Body.String())
+	}
+	// 1) intent recorded as decommissioned
+	if got := intent.Get("uuid:usb-1"); got != storage.IntentDecommissioned {
+		t.Errorf("intent = %q, want decommissioned", got)
+	}
+	// 2) guest-bind record pruned
+	if ids := gb.Guests()[8200]; len(ids) != 0 {
+		t.Errorf("guest-bind not removed: %v", ids)
+	}
+	// 3) unmounted, never formatted
+	d.mu.Lock()
+	defer d.mu.Unlock()
+	if len(d.unmountCalls) != 1 || d.unmountCalls[0] != "/mnt/bulk" {
+		t.Errorf("Unmount calls = %v, want [/mnt/bulk]", d.unmountCalls)
+	}
+	if len(d.formatCalls) != 0 {
+		t.Errorf("decommission must NEVER format; formatCalls = %v", d.formatCalls)
+	}
+}
+
+// TestReassertGuestBinds_SkipsDecommissioned is the load-bearing F9-reconnect invariant: a
+// decommissioned-but-present drive still recorded in the bind store must NOT auto-rebind on agent
+// restart. Companion: with intent=enrolled the SAME setup DOES rebind — proving the intent gate is
+// what blocks it (the pre-fix intent-blind code would rebind both → this test FAILS on it).
+func TestReassertGuestBinds_SkipsDecommissioned(t *testing.T) {
+	// decommissioned → must NOT rebind even though present + recorded.
+	gbD := tempBindStore(t)
+	_ = gbD.Record(8200, "uuid:usb-1")
+	intentD := newFakeIntent()
+	intentD.SetDecommissioned("uuid:usb-1")
+	gaD := &fakeGuestAttacher{}
+	srvD := decommServer(t, &fakeDiskOps{}, usbPresent(), fakeGuestList{}, intentD, gbD, gaD, map[int]map[string]string{
+		8200: {"mp0": "local-lvm:8,mp=/var/lib/docker"}, // bind missing → would re-add if not gated
+	})
+	srvD.ReassertGuestBinds(context.Background())
+	if gaD.count() != 0 {
+		t.Fatalf("decommissioned drive was re-bound (%d AttachBind) — intent gate missing", gaD.count())
+	}
+
+	// companion: enrolled → DOES rebind (same present drive + missing bind).
+	gbE := tempBindStore(t)
+	_ = gbE.Record(8200, "uuid:usb-1")
+	intentE := newFakeIntent()
+	intentE.SetEnrolled("uuid:usb-1")
+	gaE := &fakeGuestAttacher{}
+	srvE := decommServer(t, &fakeDiskOps{}, usbPresent(), fakeGuestList{}, intentE, gbE, gaE, map[int]map[string]string{
+		8200: {"mp0": "local-lvm:8,mp=/var/lib/docker"},
+	})
+	srvE.ReassertGuestBinds(context.Background())
+	if gaE.count() != 1 {
+		t.Fatalf("enrolled drive should rebind (got %d AttachBind) — gate too aggressive", gaE.count())
+	}
+}
+
+// TestReCommission_ReRecords: after a decommission prunes the bind, re-enrolling (SetEnrolled +
+// Record) restores it so the re-assert rebinds again. Proves Remove doesn't break re-enroll.
+func TestReCommission_ReRecords(t *testing.T) {
+	gb := tempBindStore(t)
+	intent := newFakeIntent()
+
+	// decommission removed the record + set decommissioned intent.
+	intent.SetDecommissioned("uuid:usb-1")
+	_ = gb.Remove(8200, "uuid:usb-1") // no-op (absent) — idempotent
+
+	// re-enroll (what handleDiskGuestAttach does): set enrolled + record.
+	intent.SetEnrolled("uuid:usb-1")
+	if err := gb.Record(8200, "uuid:usb-1"); err != nil {
+		t.Fatal(err)
+	}
+	ga := &fakeGuestAttacher{}
+	srv := decommServer(t, &fakeDiskOps{}, usbPresent(), fakeGuestList{}, intent, gb, ga, map[int]map[string]string{
+		8200: {"mp0": "local-lvm:8,mp=/var/lib/docker"},
+	})
+	srv.ReassertGuestBinds(context.Background())
+	if ga.count() != 1 {
+		t.Fatalf("re-commissioned drive should rebind (got %d)", ga.count())
+	}
+}
+
+// TestGuestBindStore_Remove covers idempotency (absent = no-op), present removal, empty-key drop, and
+// persistence across reopen.
+func TestGuestBindStore_Remove(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "guest-binds.json")
+	gb, err := OpenGuestBindStore(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	// absent vmid + absent id → no-op, no error
+	if err := gb.Remove(8200, "uuid:nope"); err != nil {
+		t.Fatalf("Remove absent: %v", err)
+	}
+	_ = gb.Record(8200, "uuid:a")
+	_ = gb.Record(8200, "uuid:b")
+	// remove a present id → keeps the other
+	if err := gb.Remove(8200, "uuid:a"); err != nil {
+		t.Fatal(err)
+	}
+	if ids := gb.Guests()[8200]; len(ids) != 1 || ids[0] != "uuid:b" {
+		t.Fatalf("after remove uuid:a, vmid 8200 = %v want [uuid:b]", ids)
+	}
+	// remove the last id → vmid key dropped
+	if err := gb.Remove(8200, "uuid:b"); err != nil {
+		t.Fatal(err)
+	}
+	if _, ok := gb.Guests()[8200]; ok {
+		t.Errorf("empty vmid key should be dropped")
+	}
+	// persistence: reopen and confirm empty
+	re, err := OpenGuestBindStore(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(re.Guests()) != 0 {
+		t.Errorf("store should be empty after reopen, got %v", re.Guests())
+	}
+}

internal/localapi/disks.go

@@ -71,12 +71,16 @@ type GuestAttacher interface {
 	RebootGuest(ctx context.Context, vmid int) error
 }
 
-// IntentRecorder persists drive enroll/eject INTENT (slice 10 P3 self-heal), keyed by durable-id, so
-// the watchdog reconciles only enrolled drives and respects an official eject. Satisfied by
-// *storage.IntentStore. Optional — when nil, the local API records no intent (self-heal is ungated).
+// IntentRecorder persists drive enroll/eject/decommission INTENT (slice 10 P3 self-heal), keyed by
+// durable-id, so the watchdog reconciles only enrolled drives and respects an official eject /
+// permanent decommission. Get lets the startup re-assert be intent-aware (B2 — skip non-enrolled).
+// Satisfied by *storage.IntentStore. Optional — when nil, the local API records no intent (self-heal
+// is ungated and the re-assert falls back to legacy bind-all behavior).
 type IntentRecorder interface {
 	SetEnrolled(durableID string) error
 	SetEjected(durableID string) error
+	SetDecommissioned(durableID string) error
+	Get(durableID string) storage.DriveIntent
 }
 
 // ---- handlers ---------------------------------------------------------------------------
@@ -88,14 +92,14 @@ type DiskInfo struct {
 	State         string `json:"state"`          // attached | disconnected
 	BackingDevice string `json:"backing_device"` // /dev/sdb1, … ("" for network/lvm)
 	MountPath     string `json:"mount_path"`
-	Class         string `json:"class"`        // fast | slow | ""
+	Class         string `json:"class"` // fast | slow | ""
 	// Role is the agent's AUTHORITATIVE protection tier (system | backup | user-data), derived from
 	// the agent's own storage view + host topology — never from the controller. The controller drives
 	// the UI from it: system/backup get a lock badge and NO destructive controls; user-data is
 	// customer-manageable. Defense in depth — the agent re-enforces role at wipe time regardless.
-	Role          string `json:"role"`
-	DataBearing   bool   `json:"data_bearing"` // agent device-inspection verdict (UI hint)
-	DataReason    string `json:"data_reason,omitempty"`
+	Role        string `json:"role"`
+	DataBearing bool   `json:"data_bearing"` // agent device-inspection verdict (UI hint)
+	DataReason  string `json:"data_reason,omitempty"`
 	// Capacity (from the agent's storage view) — for the controller's capacity bar. 0 when unknown.
 	TotalBytes   int64   `json:"total_bytes"`
 	UsedBytes    int64   `json:"used_bytes"`
@@ -246,6 +250,58 @@ func (s *Server) handleDiskEject(w http.ResponseWriter, r *http.Request, vmid in
 	writeOK(w, map[string]any{"vmid": vmid, "ejected": req.Where, "dependent_guests": dependents})
 }
 
+// handleDiskDecommission is the SELF-SERVE, NON-DESTRUCTIVE permanent removal of a user-data drive
+// (B2). It mirrors handleDiskEject EXACTLY — withGuest self-scoping, scopedFromBody, and the same
+// user-data ROLE GATE (a system/backup mount is refused 403; fail-safe-to-protected on ambiguity) so
+// a compromised controller can't decommission protected storage. Unlike the operator-signed
+// DecommissionExecutor it needs no signature: it is the customer's own drive. It records a PERMANENT
+// decommission intent (the self-heal watchdog + the intent-aware re-assert never auto-mount/re-bind it
+// again), prunes the guest-bind record (hygiene), and unmounts so the drive is physically removable.
+// It NEVER calls any format/mkfs path — the data stays on the drive.
+func (s *Server) handleDiskDecommission(w http.ResponseWriter, r *http.Request, vmid int) {
+	if s.disks == nil {
+		writeErr(w, http.StatusServiceUnavailable, "disk management not configured on this host")
+		return
+	}
+	var req ejectRequest
+	if !decodeBody(w, r, &req) {
+		return
+	}
+	if !s.scopedFromBody(w, req.VMID, vmid, r.URL.Path) {
+		return
+	}
+	if strings.TrimSpace(req.Where) == "" {
+		writeErr(w, http.StatusBadRequest, "where (mountpoint) is required")
+		return
+	}
+	// ROLE GATE (same as eject): user-data only. The agent classifies from its OWN view, never the
+	// caller's claim; an unresolvable mount fails safe to protected → refused.
+	if role := s.roleForMountPath(r.Context(), req.Where); role != storage.RoleUserData {
+		s.logger.Warn("local-api: protected — decommission refused by role",
+			"vmid", vmid, "where", req.Where, "role", role)
+		writeErr(w, http.StatusForbidden, "mount is system/backup-protected — decommission refused (role: "+string(role)+")")
+		return
+	}
+	dependents := s.dependentGuests(r.Context(), req.Where)
+	// Resolve the durable-id BEFORE unmounting (it still resolves while mounted) for the bind prune.
+	id := s.durableIDForMount(r.Context(), req.Where)
+	// Record the PERMANENT decommission intent first (self-heal never re-mounts it again).
+	s.recordIntent(r.Context(), req.Where, "decommissioned")
+	// Hygiene: drop the guest-bind record so the startup re-assert carries no stale id.
+	if id != "" && s.guestBinds != nil {
+		if err := s.guestBinds.Remove(vmid, id); err != nil {
+			s.logger.Warn("local-api: guest-bind remove failed", "vmid", vmid, "durable_id", id, "err", err)
+		}
+	}
+	// Unmount (mirror eject) — benign, data preserved. NEVER format/mkfs here.
+	if err := s.disks.Unmount(r.Context(), req.Where); err != nil {
+		s.logger.Error("local-api: disk decommission", "vmid", vmid, "where", req.Where, "err", err)
+		writeErr(w, http.StatusBadRequest, "decommission failed: "+err.Error())
+		return
+	}
+	writeOK(w, map[string]any{"vmid": vmid, "decommissioned": req.Where, "dependent_guests": dependents})
+}
+
 type guestAttachRequest struct {
 	VMID  int    `json:"vmid"`
 	Where string `json:"where"` // the host mount path of the enrolled drive (e.g. /mnt/felhom-usb)
@@ -670,6 +726,15 @@ func (s *Server) ReassertGuestBinds(ctx context.Context) {
 			}
 		}
 		for _, id := range ids {
+			// Intent-aware (B2, the load-bearing correctness fix): NEVER re-bind a drive that is not
+			// currently `enrolled` — an ejected or decommissioned drive must not auto-rebind into the
+			// guest on agent restart, even if it is still host-mounted. Covers both the self-serve and
+			// the operator-signed decommission paths (both land on IntentDecommissioned). A nil intent
+			// store falls back to legacy bind-all (ungated), matching the watchdog's nil-intent rule.
+			if s.intent != nil && s.intent.Get(id) != storage.IntentEnrolled {
+				s.logger.Warn("F9 re-assert: skipping non-enrolled drive (intent-gated)", "vmid", vmid, "durable_id", id, "intent", string(s.intent.Get(id)))
+				continue
+			}
 			where, present := mountByDurable[id]
 			if !present {
 				s.logger.Warn("F9 re-assert: enrolled drive not present (durable-id absent) — skipping", "vmid", vmid, "durable_id", id)
@@ -727,6 +792,8 @@ func (s *Server) recordIntent(ctx context.Context, where, action string) {
 		err = s.intent.SetEnrolled(id)
 	case "ejected":
 		err = s.intent.SetEjected(id)
+	case "decommissioned":
+		err = s.intent.SetDecommissioned(id)
 	}
 	if err != nil {
 		s.logger.Warn("local-api: intent record failed", "where", where, "action", action, "durable_id", id, "err", err)

internal/localapi/guestbindstore.go

@@ -71,6 +71,37 @@ func (s *GuestBindStore) Record(vmid int, durableID string) error {
 	return s.saveLocked()
 }
 
+// Remove drops (vmid, durableID) from the enrolled set. Idempotent — absent (vmid or id) is a no-op
+// returning nil. Atomic write (tmp+rename) like Record/saveLocked. Called by the self-serve
+// decommission endpoint so a permanently-removed drive no longer lingers in the startup re-assert
+// record (hygiene — the intent-aware ReassertGuestBinds is the load-bearing guard).
+func (s *GuestBindStore) Remove(vmid int, durableID string) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	ids, ok := s.m[vmid]
+	if !ok {
+		return nil
+	}
+	kept := ids[:0:0]
+	found := false
+	for _, id := range ids {
+		if id == durableID {
+			found = true
+			continue
+		}
+		kept = append(kept, id)
+	}
+	if !found {
+		return nil // idempotent: nothing to remove
+	}
+	if len(kept) == 0 {
+		delete(s.m, vmid)
+	} else {
+		s.m[vmid] = kept
+	}
+	return s.saveLocked()
+}
+
 // Guests returns a copy of the vmid → enrolled-durable-ids map.
 func (s *GuestBindStore) Guests() map[int][]string {
 	s.mu.Lock()

internal/localapi/server.go

@@ -193,16 +193,16 @@ func NewServer(o Options) (*Server, error) {
 		cadence = defaultBackupCadence
 	}
 	s := &Server{
-		addr:    o.ListenAddr,
-		cert:    o.Cert,
-		guests:  o.Guests,
-		backups: o.Backups,
-		store:   o.Store,
-		storage: o.Storage,
-		tokens:  o.Tokens,
-		cadence:   cadence,
-		logger:    o.Logger,
-		now:       func() time.Time { return time.Now().UTC() },
+		addr:        o.ListenAddr,
+		cert:        o.Cert,
+		guests:      o.Guests,
+		backups:     o.Backups,
+		store:       o.Store,
+		storage:     o.Storage,
+		tokens:      o.Tokens,
+		cadence:     cadence,
+		logger:      o.Logger,
+		now:         func() time.Time { return time.Now().UTC() },
 		disks:       o.Disks,
 		diskGate:    o.DiskGate,
 		guestList:   o.Guests2,
@@ -237,6 +237,7 @@ func (s *Server) Handler() http.Handler {
 	mux.HandleFunc("GET /disks", s.withGuest(s.handleDisks))
 	mux.HandleFunc("POST /disks/assign", s.withGuest(s.handleDiskAssign))
 	mux.HandleFunc("POST /disks/eject", s.withGuest(s.handleDiskEject))
+	mux.HandleFunc("POST /disks/decommission", s.withGuest(s.handleDiskDecommission))
 	mux.HandleFunc("POST /disks/format", s.withGuest(s.handleDiskFormat))
 	mux.HandleFunc("GET /disks/format/status", s.withGuest(s.handleDiskFormatStatus))
 	// Guest data-drive passthrough (slice 10 P2): bind an enrolled drive's felhom-data namespace in.
@@ -583,11 +584,11 @@ func (s *Server) handleBackupDue(w http.ResponseWriter, r *http.Request, vmid in
 // BackupStatusResponse is GET /backup/status (slice 8B): the current/last job phase + the latest
 // recorded backup. Phase is idle when no job has run this process lifetime.
 type BackupStatusResponse struct {
-	VMID    int         `json:"vmid"`
-	Phase   string      `json:"phase"`             // idle | running | done | failed
-	JobID   string      `json:"job_id,omitempty"`
-	Error   string      `json:"error,omitempty"`
-	Backup  *hub.Backup `json:"backup,omitempty"`  // latest recorded backup for this guest
+	VMID   int         `json:"vmid"`
+	Phase  string      `json:"phase"` // idle | running | done | failed
+	JobID  string      `json:"job_id,omitempty"`
+	Error  string      `json:"error,omitempty"`
+	Backup *hub.Backup `json:"backup,omitempty"` // latest recorded backup for this guest
 }
 
 func (s *Server) handleBackupStatus(w http.ResponseWriter, r *http.Request, vmid int) {