diff --git a/compose/orchestrator.go b/compose/orchestrator.go
index d6c62ce..9b667f2 100644
--- a/compose/orchestrator.go
+++ b/compose/orchestrator.go
@@ -488,6 +488,23 @@ func (o *Orchestrator) ensureService(
 
 	if len(existing) > 0 {
 		c := existing[0]
+		// Resume: reattach on-disk state exactly. Reuse the existing
+		// container unconditionally — start if stopped, attach if
+		// running — without consulting the config-hash. A restored
+		// workspace's primary image is feature-layered fresh every boot
+		// (new digest), so a hash check would force a recreate that
+		// abandons the container's upperdir and binds a new empty
+		// anonymous volume; adoption keeps both. Recreate-mode Up has
+		// already torn the project down before reaching here, so this
+		// only fires on non-recreating (resume) Ups.
+		if plan.AdoptExisting {
+			if c.State != runtime.StateRunning {
+				if err := o.rt.StartContainer(ctx, c.ID); err != nil {
+					return "", fmt.Errorf("StartContainer(resumed %s): %w", svc.Name, err)
+				}
+			}
+			return c.ID, nil
+		}
 		// Inspect to read the stored hash + image digest. Recreate if
 		// either has drifted; the image-digest check is the second
 		// line of defense for tag-update scenarios where ConfigHash
diff --git a/compose/orchestrator_test.go b/compose/orchestrator_test.go
index 6b4070e..6591f40 100644
--- a/compose/orchestrator_test.go
+++ b/compose/orchestrator_test.go
@@ -1075,3 +1075,65 @@ func TestUp_ServiceNetworkModeResolvesToContainer(t *testing.T) {
 		t.Errorf("app NetworkMode = %q, want %q", got, want)
 	}
 }
+
+// AdoptExisting (resume) reuses an existing container even when its stored
+// config-hash no longer matches — a restored workspace's primary image is
+// feature-layered fresh every boot, so the hash always drifts; adoption must
+// keep the container (and its upperdir + anon volumes) anyway.
+func TestUp_AdoptExistingReusesDespiteHashDrift(t *testing.T) {
+	rt := newMockRuntime()
+	orch := NewOrchestrator(rt, "docker")
+	proj := newProject(t, map[string][]string{"app": nil})
+
+	// A prior orchestrator-made container whose stored hash is stale.
+	rt.containers["resumed-1"] = &mockContainer{
+		id: "resumed-1", name: "dc-x-app-1", image: "alpine",
+		labels: map[string]string{
+			LabelComposeProject: "dc-x",
+			LabelComposeService: "app",
+			LabelConfigHash:     "STALE-hash-from-a-previous-boot",
+			LabelImageDigest:    "sha256:stale",
+		},
+		state: runtime.StateExited,
+	}
+
+	res, err := orch.Up(context.Background(), &Plan{
+		Project: proj, ProjectName: "dc-x", AdoptExisting: true,
+	})
+	if err != nil {
+		t.Fatalf("Up: %v", err)
+	}
+	if res.ContainerIDs["app"] != "resumed-1" {
+		t.Errorf("ContainerIDs[app] = %q, want the adopted resumed-1", res.ContainerIDs["app"])
+	}
+	if rt.removeCalls != 0 {
+		t.Errorf("removeCalls = %d; adoption must not recreate despite hash drift", rt.removeCalls)
+	}
+	if rt.runCalls != 0 {
+		t.Errorf("runCalls = %d; adoption must not create a replacement", rt.runCalls)
+	}
+	if rt.containers["resumed-1"].state != runtime.StateRunning {
+		t.Errorf("adopted container not started: state=%v", rt.containers["resumed-1"].state)
+	}
+}
+
+// Without AdoptExisting, hash drift still recreates (guards the default path).
+func TestUp_NoAdoptRecreatesOnHashDrift(t *testing.T) {
+	rt := newMockRuntime()
+	orch := NewOrchestrator(rt, "docker")
+	proj := newProject(t, map[string][]string{"app": nil})
+	rt.containers["old-1"] = &mockContainer{
+		id: "old-1", name: "dc-x-app-1", image: "alpine",
+		labels: map[string]string{
+			LabelComposeProject: "dc-x", LabelComposeService: "app",
+			LabelConfigHash: "STALE", LabelImageDigest: "sha256:stale",
+		},
+		state: runtime.StateExited,
+	}
+	if _, err := orch.Up(context.Background(), &Plan{Project: proj, ProjectName: "dc-x"}); err != nil {
+		t.Fatalf("Up: %v", err)
+	}
+	if rt.removeCalls == 0 {
+		t.Error("expected recreate (remove) on hash drift without AdoptExisting")
+	}
+}
diff --git a/compose/plan.go b/compose/plan.go
index 23f5bcb..5e346a5 100644
--- a/compose/plan.go
+++ b/compose/plan.go
@@ -35,6 +35,12 @@ type Plan struct {
 	// fills these in with the devcontainer ID label set so
 	// Engine.Attach can find the primary container.
 	Labels map[string]string
+
+	// AdoptExisting reuses any existing (project, service) container
+	// unconditionally — start if stopped, attach if running, never
+	// recreate on a config-hash / image-digest mismatch. The resume
+	// contract: reattach on-disk state exactly, do not reconcile drift.
+	AdoptExisting bool
 }
 
 // DownPlan describes a teardown request. Used by Orchestrator.Down
diff --git a/up.go b/up.go
index fe58192..d56bb87 100644
--- a/up.go
+++ b/up.go
@@ -37,6 +37,18 @@ type UpOptions struct {
 	// running container is attached to.
 	Recreate bool
 
+	// AdoptExisting, for compose sources, reuses any existing container
+	// found for a (project, service) regardless of whether its stored
+	// config-hash / image-digest still match — start it if stopped,
+	// attach if running, never recreate. This is the RESUME contract:
+	// reattach the workspace exactly as it was left, preserving every
+	// container's writable upperdir and its (possibly anonymous) volumes,
+	// rather than reconciling against config drift (a rebuilt
+	// feature-layered primary image has a fresh digest every boot, which
+	// otherwise forces a recreate that abandons the upperdir and binds a
+	// new empty anonymous volume). Ignored when Recreate is true.
+	AdoptExisting bool
+
 	// PullPolicy controls image pulling. Default IfNotPresent.
 	PullPolicy PullPolicy
 
@@ -715,9 +727,10 @@ func (e *Engine) upComposeNative(
 
 	orch := compose.NewOrchestrator(e.runtime, "")
 	res, err := orch.Up(ctx, &compose.Plan{
-		Project:     project,
-		ProjectName: projectName,
-		Services:    src.RunServices,
+		Project:       project,
+		ProjectName:   projectName,
+		Services:      src.RunServices,
+		AdoptExisting: opts.AdoptExisting && !opts.Recreate,
 		Labels: map[string]string{
 			LabelDevcontainerID: cfg.DevcontainerID,
 		},