fix: resolve schema against runtime metadata in plugin builds; gate cache overlay by version (#1764)

* fix(schema): fall back to runtime catalog when no embedded metadata

Binaries built from the bare Go module (plugin builds) embed only the
empty meta_data_default.json stub because meta_data.json is gitignored
and fetched at build time. The schema command, its completion, and the
affordance command-form resolver read the embedded-only catalog, so
every schema lookup failed with "Unknown service" even though the
runtime registry had already sync-fetched full metadata.

Add registry.SchemaCatalog(): embedded when compiled in (official
builds unchanged, still deterministic), otherwise the merged runtime
catalog seeded from cache or remote fetch. When neither source has
data (offline plugin build with a cold cache), schema now returns a
failed_precondition error with an actionable hint instead of
"Unknown service" with an empty candidate list.

* fix(registry): gate cached meta overlay on version newer than embedded

The cached remote meta was overlaid onto the embedded meta_data.json
unconditionally, so after a CLI upgrade an equal- or older-version
cache kept shadowing the freshly shipped embedded definitions until a
later refresh happened to rewrite it.

Only overlay when the cache version is strictly newer than the
embedded baseline. The bare-module stub baseline is "0.0.0", so plugin
builds without compiled metadata still take any real cached version
(TestOverlayGate_StubEmbedded_OverlaysRealCache) and the schema
runtime fallback keeps working offline from a warm cache.

Ports #1376 onto the typed meta model.

---------

Co-authored-by: liangshuo-1 <266696938+liangshuo-1@users.noreply.github.com>

cmd/schema/schema.go

@@ -65,13 +65,13 @@ func NewCmdSchema(f *cmdutil.Factory, runF func(*SchemaOptions) error) *cobra.Co
 	return cmd
 }
 
-// completeSchemaPath is a thin adapter over the embedded catalog's Complete.
-// It uses the embedded source so completion candidates match what `schema`
-// execution can resolve (both overlay-free).
+// completeSchemaPath is a thin adapter over the schema catalog's Complete.
+// It uses the same source as schema execution so completion candidates match
+// what `schema` can resolve.
 func completeSchemaPath(f *cmdutil.Factory) func(*cobra.Command, []string, string) ([]string, cobra.ShellCompDirective) {
 	return func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
 		mode := f.ResolveStrictMode(cmd.Context())
-		completions, noSpace := registry.EmbeddedCatalog().Complete(args, toComplete, registry.FilterForStrictMode(mode))
+		completions, noSpace := registry.SchemaCatalog().Complete(args, toComplete, registry.FilterForStrictMode(mode))
 		directive := cobra.ShellCompDirectiveNoFileComp
 		if noSpace {
 			directive |= cobra.ShellCompDirectiveNoSpace
@@ -86,13 +86,19 @@ func schemaRun(opts *SchemaOptions) error {
 	return runSchema(out, apicatalog.ParsePath(opts.Args), mode)
 }
 
-// runSchema resolves the path through the embedded catalog and renders the
+// runSchema resolves the path through the schema catalog and renders the
 // matching envelope(s). The catalog owns navigation (Resolve + MethodRefs) and
 // schema owns rendering (Envelope/Envelopes); this adapter only chooses the
 // output shape — a single resolved method renders as one envelope object,
 // anything broader as an array — and maps resolve failures to hints.
 func runSchema(out io.Writer, parts []string, mode core.StrictMode) error {
-	catalog := registry.EmbeddedCatalog()
+	catalog := registry.SchemaCatalog()
+	if len(catalog.Services()) == 0 {
+		// No embedded metadata and the runtime fallback is empty too: offline
+		// with a cold cache, remote meta off, or an unwritable cache dir.
+		return errs.NewValidationError(errs.SubtypeFailedPrecondition, "No API metadata available").
+			WithHint("this binary has no embedded API metadata; run any command with network access to the open platform once so metadata can be fetched and cached")
+	}
 	target, err := catalog.Resolve(parts)
 	if err != nil {
 		return resolveError(err)

internal/affordance/affordance.go

@@ -77,14 +77,10 @@ func loadService(service string) map[string]json.RawMessage {
 // space→dot fallback covers domains where the two already coincide.
 func commandFormResolver(service string) func(string) string {
 	byForm := map[string]string{}
-	for _, svc := range registry.EmbeddedServicesTyped() {
-		if svc.Name != service {
-			continue
-		}
+	if svc, ok := registry.SchemaCatalog().Service(service); ok {
 		for _, ref := range apicatalog.ServiceMethods(svc, nil) {
 			byForm[strings.Join(ref.CommandPath()[1:], " ")] = ref.Method.ID
 		}
-		break
 	}
 	return func(h string) string {
 		h = strings.TrimSpace(h)

internal/registry/catalog.go

@@ -6,8 +6,7 @@ package registry
 import "github.com/larksuite/cli/internal/apicatalog"
 
 // EmbeddedCatalog returns a navigation catalog over the embedded (overlay-free)
-// metadata — deterministic across machines, for `lark-cli schema`, golden tests
-// and schema lint.
+// metadata — deterministic across machines, for golden tests and schema lint.
 func EmbeddedCatalog() apicatalog.Catalog {
 	return apicatalog.New(apicatalog.SourceEmbedded, EmbeddedServicesTyped())
 }
@@ -18,3 +17,14 @@ func EmbeddedCatalog() apicatalog.Catalog {
 func RuntimeCatalog() apicatalog.Catalog {
 	return apicatalog.New(apicatalog.SourceRuntime, ServicesTyped())
 }
+
+// SchemaCatalog returns the embedded catalog when metadata is compiled in,
+// otherwise the merged runtime catalog. Binaries built from the bare Go module
+// embed only the empty meta_data_default.json stub, so the embedded view has
+// nothing to resolve; the merged view is the only data such binaries have.
+func SchemaCatalog() apicatalog.Catalog {
+	if len(EmbeddedServicesTyped()) > 0 {
+		return EmbeddedCatalog()
+	}
+	return RuntimeCatalog()
+}

internal/registry/catalog_test.go

@@ -0,0 +1,67 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package registry
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/larksuite/cli/internal/apicatalog"
+)
+
+// swapEmbeddedMeta replaces the compiled-in metadata bytes for one test and
+// restores them (with a full state reset) on cleanup.
+func swapEmbeddedMeta(t *testing.T, data []byte) {
+	t.Helper()
+	resetInit()
+	orig := embeddedMetaJSON
+	embeddedMetaJSON = data
+	t.Cleanup(func() {
+		waitBackgroundRefresh()
+		embeddedMetaJSON = orig
+		resetInit()
+	})
+}
+
+func TestSchemaCatalog_EmbeddedWhenCompiledIn(t *testing.T) {
+	swapEmbeddedMeta(t, testCacheJSON("embedded_svc"))
+	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
+	t.Setenv("LARKSUITE_CLI_REMOTE_META", "off")
+
+	c := SchemaCatalog()
+
+	if c.Source() != apicatalog.SourceEmbedded {
+		t.Fatalf("Source = %q, want %q", c.Source(), apicatalog.SourceEmbedded)
+	}
+	if _, ok := c.Service("embedded_svc"); !ok {
+		t.Fatal("expected embedded_svc from embedded metadata")
+	}
+}
+
+// TestSchemaCatalog_FallsBackToRuntimeWhenNoEmbedded simulates a binary built
+// from the bare Go module (plugin builds): only the empty meta_data_default.json
+// stub is compiled in, so SchemaCatalog must serve the merged runtime view that
+// Init seeds via sync fetch.
+func TestSchemaCatalog_FallsBackToRuntimeWhenNoEmbedded(t *testing.T) {
+	swapEmbeddedMeta(t, embeddedMetaDataDefaultJSON)
+	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", t.TempDir())
+	t.Setenv("LARKSUITE_CLI_REMOTE_META", "on")
+
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(200)
+		w.Write(testEnvelopeJSON("remote_svc"))
+	}))
+	defer ts.Close()
+	testMetaURL = ts.URL
+
+	c := SchemaCatalog()
+
+	if c.Source() != apicatalog.SourceRuntime {
+		t.Fatalf("Source = %q, want %q", c.Source(), apicatalog.SourceRuntime)
+	}
+	if _, ok := c.Service("remote_svc"); !ok {
+		t.Fatal("expected remote_svc from runtime fallback")
+	}
+}

internal/registry/loader.go

@@ -15,6 +15,7 @@ import (
 
 	"github.com/larksuite/cli/internal/core"
 	"github.com/larksuite/cli/internal/meta"
+	"github.com/larksuite/cli/internal/update"
 )
 
 //go:embed scope_priorities.json scope_overrides.json
@@ -85,7 +86,9 @@ func InitWithBrand(brand core.LarkBrand) {
 			brandChanged := metaErr == nil && cm.Brand != "" && cm.Brand != string(brand)
 
 			if !brandChanged {
-				if cached, err := loadCachedMerged(); err == nil {
+				// After a CLI upgrade the embedded data can be fresher than an old
+				// cache; an equal/older cache must not shadow it.
+				if cached, err := loadCachedMerged(); err == nil && update.IsNewer(cached.Version, embeddedVersion) {
 					overlayMergedServices(cached)
 				}
 			}

internal/registry/loader_test.go

@@ -0,0 +1,102 @@
+// Copyright (c) 2026 Lark Technologies Pte. Ltd.
+// SPDX-License-Identifier: MIT
+
+package registry
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/larksuite/cli/internal/core"
+	"github.com/larksuite/cli/internal/meta"
+)
+
+// seedCache writes a cache file + cache meta for one service whose Title is
+// marker, tagged with the given top-level data version and brand.
+func seedCache(t *testing.T, dir, name, marker, version, brand string) {
+	t.Helper()
+	cDir := filepath.Join(dir, "cache")
+	if err := os.MkdirAll(cDir, 0700); err != nil {
+		t.Fatal(err)
+	}
+	reg := MergedRegistry{
+		Version:  version,
+		Services: []meta.Service{{Name: name, Version: "cache", Title: marker}},
+	}
+	data, _ := json.Marshal(reg)
+	if err := os.WriteFile(filepath.Join(cDir, "remote_meta.json"), data, 0644); err != nil {
+		t.Fatal(err)
+	}
+	cm := CacheMeta{LastCheckAt: time.Now().Unix(), Version: version, Brand: brand}
+	mData, _ := json.Marshal(cm)
+	if err := os.WriteFile(filepath.Join(cDir, "remote_meta.meta.json"), mData, 0644); err != nil {
+		t.Fatal(err)
+	}
+}
+
+// initWithCache runs a fresh feishu-brand init with remote on, a high TTL and a
+// recent LastCheckAt (so no refresh fires), embedded meta at embeddedVer and a
+// pre-seeded cache at cacheVer — the overlay version gate is the only variable.
+func initWithCache(t *testing.T, embeddedVer, cacheVer string) {
+	t.Helper()
+	embedded, _ := json.Marshal(MergedRegistry{
+		Version:  embeddedVer,
+		Services: []meta.Service{{Name: "svc", Version: "embedded", Title: "EMBEDDED"}},
+	})
+	swapEmbeddedMeta(t, embedded)
+	tmp := t.TempDir()
+	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", tmp)
+	t.Setenv("LARKSUITE_CLI_REMOTE_META", "on")
+	t.Setenv("LARKSUITE_CLI_META_TTL", "3600")
+	seedCache(t, tmp, "svc", "CACHE", cacheVer, "feishu")
+	InitWithBrand(core.BrandFeishu)
+}
+
+func titleOf(t *testing.T, name string) string {
+	t.Helper()
+	svc, ok := ServiceTyped(name)
+	if !ok {
+		t.Fatalf("service %q not loaded", name)
+	}
+	return svc.Title
+}
+
+func TestOverlayGate_EqualVersion_UsesEmbedded(t *testing.T) {
+	initWithCache(t, "1.0.0", "1.0.0")
+	if got := titleOf(t, "svc"); got != "EMBEDDED" {
+		t.Errorf("equal version: got %q, want EMBEDDED (cache must not overlay)", got)
+	}
+}
+
+func TestOverlayGate_OlderCache_UsesEmbedded(t *testing.T) {
+	initWithCache(t, "2.0.0", "1.0.0")
+	if got := titleOf(t, "svc"); got != "EMBEDDED" {
+		t.Errorf("older cache: got %q, want EMBEDDED", got)
+	}
+}
+
+func TestOverlayGate_NewerCache_OverlaysCache(t *testing.T) {
+	initWithCache(t, "1.0.0", "2.0.0")
+	if got := titleOf(t, "svc"); got != "CACHE" {
+		t.Errorf("newer cache: got %q, want CACHE", got)
+	}
+}
+
+func TestOverlayGate_UnparseableCacheVersion_UsesEmbedded(t *testing.T) {
+	initWithCache(t, "1.0.0", "not-a-semver")
+	if got := titleOf(t, "svc"); got != "EMBEDDED" {
+		t.Errorf("unparseable cache version: got %q, want EMBEDDED", got)
+	}
+}
+
+func TestOverlayGate_StubEmbedded_OverlaysRealCache(t *testing.T) {
+	// The bare-module stub baseline is "0.0.0"; a real cache version must win so
+	// plugin builds without compiled meta_data.json still get remote data.
+	initWithCache(t, "0.0.0", "1.0.0")
+	if got := titleOf(t, "svc"); got != "CACHE" {
+		t.Errorf("stub-embedded baseline: got %q, want CACHE", got)
+	}
+}

internal/registry/remote_test.go

@@ -72,9 +72,11 @@ func hasEmbeddedServices() bool {
 }
 
 // testRegistry returns a minimal MergedRegistry with one service.
+// The version is a real semver newer than the embedded stub baseline ("0.0.0")
+// so cache overlay passes the version gate in InitWithBrand.
 func testRegistry(name string) MergedRegistry {
 	return MergedRegistry{
-		Version: "test-1.0",
+		Version: "1.0.0",
 		Services: []meta.Service{
 			{
 				Name:        name,
@@ -160,7 +162,7 @@ func TestRemoteOff_SkipsRemoteLogic(t *testing.T) {
 }
 
 func TestCacheHit_WithinTTL(t *testing.T) {
-	resetInit()
+	swapEmbeddedMeta(t, nil) // overlay must depend only on the cache version, not the ambient embedded meta
 	tmp := t.TempDir()
 	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", tmp)
 	t.Setenv("LARKSUITE_CLI_REMOTE_META", "on")
@@ -197,7 +199,7 @@ func TestCacheHit_WithinTTL(t *testing.T) {
 }
 
 func TestNetworkError_SilentDegradation(t *testing.T) {
-	resetInit()
+	swapEmbeddedMeta(t, nil) // overlay must depend only on the cache version, not the ambient embedded meta
 	tmp := t.TempDir()
 	t.Setenv("LARKSUITE_CLI_CONFIG_DIR", tmp)
 	t.Setenv("LARKSUITE_CLI_REMOTE_META", "on")
@@ -371,8 +373,8 @@ func TestFetchRemoteMerged_200(t *testing.T) {
 	if data == nil {
 		t.Fatal("expected non-nil data")
 	}
-	if reg.Version != "test-1.0" {
-		t.Errorf("expected version test-1.0, got %s", reg.Version)
+	if reg.Version != "1.0.0" {
+		t.Errorf("expected version 1.0.0, got %s", reg.Version)
 	}
 }