Merge branch 'release/v3.8.50' into feat/9490-opencode-plugin-warm-startup-parallel-refresh

@omniroute/opencode-plugin/README.md

@@ -196,7 +196,7 @@ npm install --prefix ~/.config/opencode/plugins/omniroute-opencode-plugin-prepro
 | Compression pipeline tags                   | Combo names get tagged with their compression pipeline (e.g. `Combo: claude-primary [rtk🟡 → caveman🟠]`) when `features.compressionMetadata: true`. Intensity tokens render as a traffic-light emoji: 🟢 lite/minimal · 🟡 standard · 🟠 aggressive/full · 🔴 ultra                                                                                                                                        | both hooks                   |
 | Provider-tag prefix                         | Prepend short upstream-provider label to enriched names (e.g. `Claude - Claude Opus 4.7` vs `Kiro - Claude Opus 4.7`, `GHM - GPT 5`) so same-id models routed via different upstream connections group visibly in the picker (default-on, opt-out via `features.providerTag: false`)                                                                                                                        | both hooks                   |
 | Usable-only filter                          | Filter to providers with at least one healthy connection in `/api/providers` (opt-in via `features.usableOnly`)                                                                                                                                                                                                                                                                                             | both hooks                   |
-| Disk-cache fallback                         | Last-known-good catalog persisted to disk; hydrates on a cold start when `/v1/models` is unreachable (default-on, opt-out via `features.diskCache: false`)                                                                                                                                                                                                                                                  | `config`                     |
+| Disk-cache fallback + warm startup           | Last-known-good catalog persisted to disk; hydrates on a cold start when `/v1/models` is unreachable. On warm startup, the snapshot is served *before* the live fetch so the provider registers immediately (~1-2s vs ~30s). All six fetchers run concurrently via `Promise.allSettled` (default-on, opt-out via `features.diskCache: false`)                                                                                                                                                                                                                                                  | `config`                     |
 | Bearer injection + suffix-spoof guard       | Adds `Authorization` on baseURL-matched requests only                                                                                                                                                                                                                                                                                                                                                       | `auth.loader.fetch`          |
 | Gemini schema sanitization                  | Strips `$schema`/`$ref`/`additionalProperties` for `gemini-*`/`google-vertex-gemini/*`                                                                                                                                                                                                                                                                                                                      | `auth.loader.fetch` wrap     |
 | Multi-instance                              | Each plugin entry binds to its own `providerId`; closures isolated                                                                                                                                                                                                                                                                                                                                          | factory                      |
@@ -226,7 +226,7 @@ Every field is optional. Defaults mirror v0.1.0 behaviour so existing `opencode.
 | `compressionMetadata` | `boolean` | `false` | Pull `/api/context/combos` so combo names get tagged with their compression pipeline, e.g. `Combo: claude-primary [rtk🟡 → caveman🟠]`. Intensity tokens render as traffic-light emoji (🟢 lite/minimal · 🟡 standard · 🟠 aggressive/full · 🔴 ultra) so the picker advertises "how compressed" each combo is at a glance.                                                                                                                                                                                                                                          |
 | `providerTag`         | `boolean` | `true`  | Prepend a short upstream-provider label to the enriched display name with `" - "` separator, so `cc/claude-opus-4-7 → Claude - Claude Opus 4.7` differs visibly from `kr/claude-opus-4-7 → Kiro - Claude Opus 4.7` in the OC TUI model picker. Label resolution: use `/api/pricing/models[<alias>].name` verbatim when ≤8 chars (e.g. `Claude`, `Kiro`, `Codex`, `Qwen`), otherwise fall back to `UPPER(alias)` (e.g. `GitHub Models` → `GHM`, `Gemini` → `GEMINI`). Idempotent. Combos intentionally skipped (the `Combo:` prefix already conveys multi-upstream). |
 | `usableOnly`          | `boolean` | `false` | Read `/api/providers` and filter the catalog to providers that have at least one connection with `isActive: true` AND `testStatus: 'active'`. Subtract-filter semantics: providers unknown to BOTH the pricing-models catalog AND the connection table pass through (so synthetic prefixes like `agentrouter/*` survive). On fetch failure the filter is disabled for the refresh — never hides the whole catalog.                                                                                                                                                   |
-| `diskCache`           | `boolean` | `true`  | Persist the last successful `/v1/models` + `/api/combos` + enrichment + connections + compression snapshot to `${OPENCODE_DATA_DIR ?? ~/.local/share/opencode}/plugins/omniroute-<providerId>.json`. On a subsequent cold start where `/v1/models` throws (network down / IP whitelist drop / 5xx) the static block hydrates from the snapshot so OC's model picker survives offline. Soft-fail on read/write — never blocks publishing.                                                                                                                             |
+| `diskCache`           | `boolean` | `true`  | Persist the last successful `/v1/models` + `/api/combos` + enrichment + connections + compression snapshot to `${OPENCODE_DATA_DIR ?? ~/.local/share/opencode}/plugins/omniroute-<providerId>.json`. On warm startup, the snapshot is served *before* the live fetch so the provider registers immediately (~1-2s vs ~30s). All six fetchers run concurrently via `Promise.allSettled`. A failed refresh keeps the snapshot (no overwrite). On a cold start where `/v1/models` throws (network down / IP whitelist drop / 5xx) the static block hydrates from the snapshot so OC's model picker survives offline. Soft-fail on read/write — never blocks publishing.                                                                                                                             |
 | `geminiSanitization`  | `boolean` | `true`  | Strip `$schema`/`$ref`/`additionalProperties` from tool params when the model id matches `gemini`                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
 | `mcpAutoEmit`         | `boolean` | `false` | Auto-write an `mcp.<providerId>` remote entry into the OC config pointing at `<baseURL>/api/mcp/stream` with the resolved Bearer token                                                                                                                                                                                                                                                                                                                                                                                                                               |
 | `mcpToken`            | `string`  | _unset_ | Optional separate Bearer for the auto-emitted MCP entry. Falls back to the provider's `apiKey` (from `auth.json`) when unset                                                                                                                                                                                                                                                                                                                                                                                                                                         |
@@ -294,7 +294,7 @@ If you want a narrower-scoped Bearer for MCP (different from the chat/inference
 ```
 
 - `usableOnly: true` drops models whose canonical provider has no healthy connection in your OmniRoute instance — your `/models` picker stays focused on what you can actually call.
-- `diskCache: true` (default) writes a snapshot to `${OPENCODE_DATA_DIR}/plugins/omniroute-<providerId>.json` on every healthy refresh. On a cold start where `/v1/models` is unreachable (laptop offline, IP whitelist drop), the snapshot hydrates the static block so OC still shows the catalog instead of a stub.
+- `diskCache: true` (default) writes a snapshot to `${OPENCODE_DATA_DIR}/plugins/omniroute-<providerId>.json` on every healthy refresh. On warm startup, the snapshot is served *before* the live fetch so the provider registers immediately (~1-2s vs ~30s). All six fetchers run concurrently via `Promise.allSettled` instead of sequentially. A failed refresh keeps the snapshot (no overwrite). On a cold start where `/v1/models` is unreachable (laptop offline, IP whitelist drop), the snapshot hydrates the static block so OC still shows the catalog instead of a stub.
 - `compressionMetadata: true` annotates combo display names with their pipeline using traffic-light emoji for intensity (e.g. `Combo: claude-primary [rtk🟡 → caveman🟠]`) so the picker advertises which compression each combo applies and how heavy it is at a glance. Palette: 🟢 lite/minimal · 🟡 standard · 🟠 aggressive/full · 🔴 ultra. Unknown intensities fall through to raw text (`[rtk:custom-thing]`) so the plugin never hides a value OmniRoute knows but the plugin doesn't.
 - `providerTag: true` (default) prepends a short upstream-provider label so the picker shows `Claude - Claude Opus 4.7` for `cc/claude-opus-4-7`, `Kiro - Claude Opus 4.7` for `kr/claude-opus-4-7`, and `GHM - GPT 5` for `ghm/gpt-5` (slot.name `GitHub Models` > 8 chars → abbreviated). Critical when the same model id is sold through multiple upstream connections with different cost/auth/rate-limit profiles. Set to `false` to keep the pre-v3.8.3 unsuffixed format.
 

@omniroute/opencode-plugin/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@omniroute/opencode-plugin",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@omniroute/opencode-plugin",
-      "version": "0.2.0",
+      "version": "0.2.1",
       "license": "MIT",
       "dependencies": {
         "zod": "^4.4.3"

@omniroute/opencode-plugin/package.json

@@ -23,7 +23,7 @@
   "scripts": {
     "build": "tsup",
     "clean": "rm -rf dist",
-    "test": "node --import tsx/esm --test tests/scaffold.test.ts tests/auth.test.ts tests/options-schema.test.ts tests/multi-instance.test.ts tests/fetch-interceptor.test.ts tests/provider.test.ts tests/gemini-sanitize.test.ts tests/combos.test.ts tests/config-shim.test.ts tests/features.test.ts tests/feature-defaults.test.ts tests/usable-combo.test.ts tests/disk-snapshot-perms.test.ts tests/fork-features.test.ts tests/auto-combo-context.test.ts tests/provider-id-routing.test.ts tests/management-read-token.test.ts tests/auto-sync.test.ts",
+    "test": "node --import tsx/esm --test tests/scaffold.test.ts tests/auth.test.ts tests/options-schema.test.ts tests/multi-instance.test.ts tests/fetch-interceptor.test.ts tests/provider.test.ts tests/gemini-sanitize.test.ts tests/combos.test.ts tests/config-shim.test.ts tests/features.test.ts tests/feature-defaults.test.ts tests/usable-combo.test.ts tests/disk-snapshot-perms.test.ts tests/fork-features.test.ts tests/auto-combo-context.test.ts tests/provider-id-routing.test.ts tests/management-read-token.test.ts tests/auto-sync.test.ts tests/warm-startup.test.ts",
     "prepublishOnly": "npm run clean && npm run build && npm test"
   },
   "keywords": [

@omniroute/opencode-plugin/src/index.ts

@@ -1026,7 +1026,11 @@ export const OmniRoutePlugin: Plugin = async (_input, options) => {
   // Config hook: keep existing catalog shim, and register slash command
   // templates that ask the agent to call the force-sync tool (OpenCode has no
   // Pi-style registerCommand API; tools + command templates are the native path).
-  const baseConfigHook = createOmniRouteConfigHook(resolved, { cache: sharedCache });
+  const baseConfigHook = createOmniRouteConfigHook(resolved, {
+    cache: sharedCache,
+    diskSnapshotReader: defaultDiskSnapshotReader,
+    diskSnapshotWriter: defaultDiskSnapshotWriter,
+  });
   const configWithSyncCommand = async (input: Config) => {
     await baseConfigHook(input);
     const cfg = input as Config & {
@@ -4601,7 +4605,7 @@ export type OmniRouteDiskSnapshotWriter = (
 export type OmniRouteDiskSnapshotReader = (
   providerId: string,
   identityFingerprint: string
-) => Promise<Omit<OmniRouteFetchCacheEntry, "expiresAt"> | undefined>;
+) => Promise<(Omit<OmniRouteFetchCacheEntry, "expiresAt"> & { writtenAt?: number }) | undefined>;
 
 /**
  * Bind a snapshot to the endpoint and effective credential tuple without
@@ -4684,15 +4688,36 @@ export const defaultDiskSnapshotReader: OmniRouteDiskSnapshotReader = async (
         ? parsed.rawCompressionCombos
         : [],
       rawConnections: Array.isArray(parsed.rawConnections) ? parsed.rawConnections : [],
+      writtenAt: typeof parsed.writtenAt === "number" ? parsed.writtenAt : undefined,
     };
   } catch {
     return undefined;
   }
 };
 
-/** No-op disk-cache pair — used by tests to avoid filesystem side effects. */
+/** No-op disk-cache pair — used by tests to avoid filesystem side effects.
+ * Also used as the default in createOmniRouteConfigHook so that tests
+ * that don't pass a diskSnapshotReader don't read real snapshot files
+ * from the user's ~/.local/share/opencode/plugins/ directory.
+ * The OmniRoutePlugin function passes the real defaultDiskSnapshotReader
+ * explicitly. */
+export const noopDiskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
 export const noopDiskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
 
+/**
+ * In-flight refresh guard: prevents concurrent refreshes for the same
+ * cacheKey. When a warm snapshot is served, the refresh runs detached; if
+ * a second hook invocation arrives before the refresh completes, it should
+ * piggyback on the in-flight promise rather than starting a second one.
+ * Cleared on settle so it doesn't leak.
+ */
+const _inflightRefresh: Map<string, Promise<void>> = new Map();
+
+/** Reset the in-flight refresh guard (for test isolation). */
+export function _resetInflightRefresh(): void {
+  _inflightRefresh.clear();
+}
+
 // ────────────────────────────────────────────────────────────────────────────
 // Debug logging (features.debugLog)
 // ────────────────────────────────────────────────────────────────────────────
@@ -4927,7 +4952,6 @@ export function createDebugLoggingFetch(
     }
   };
 }
-export const noopDiskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
 
 export type OmniRouteReadAuthJson = () => Promise<AuthJsonShape | undefined | null>;
 
@@ -5030,8 +5054,8 @@ export function createOmniRouteConfigHook(
   const compressionMetaFetcher =
     deps.compressionMetaFetcher ?? defaultOmniRouteCompressionMetaFetcher;
   const providersFetcher = deps.providersFetcher ?? defaultOmniRouteProvidersFetcher;
-  const diskSnapshotReader = deps.diskSnapshotReader ?? defaultDiskSnapshotReader;
-  const diskSnapshotWriter = deps.diskSnapshotWriter ?? defaultDiskSnapshotWriter;
+  const diskSnapshotReader = deps.diskSnapshotReader ?? noopDiskSnapshotReader;
+  const diskSnapshotWriter = deps.diskSnapshotWriter ?? noopDiskSnapshotWriter;
   const now = deps.now ?? Date.now;
   const cache: OmniRouteFetchCache = deps.cache ?? new Map();
   const logger = deps.logger ?? console;
@@ -5126,12 +5150,12 @@ export function createOmniRouteConfigHook(
     const t = now();
     const cached = cache.get(cacheKey);
 
-    let rawModels: OmniRouteRawModelEntry[];
-    let rawCombos: OmniRouteRawCombo[];
-    let rawAutoCombos: OmniRouteRawAutoCombo[];
-    let rawEnrichment: OmniRouteEnrichmentMap;
-    let rawCompressionCombos: OmniRouteCompressionCombo[];
-    let rawConnections: OmniRouteProviderConnection[];
+    let rawModels: OmniRouteRawModelEntry[] = [];
+    let rawCombos: OmniRouteRawCombo[] = [];
+    let rawAutoCombos: OmniRouteRawAutoCombo[] = [];
+    let rawEnrichment: OmniRouteEnrichmentMap = new Map();
+    let rawCompressionCombos: OmniRouteCompressionCombo[] = [];
+    let rawConnections: OmniRouteProviderConnection[] = [];
 
     if (cached && cached.expiresAt > t) {
       rawModels = cached.rawModels;
@@ -5141,160 +5165,275 @@ export function createOmniRouteConfigHook(
       rawCompressionCombos = cached.rawCompressionCombos;
       rawConnections = cached.rawConnections;
     } else {
-      // Fail-open fetcher errors: on /v1/models throw, fall back to empty
-      // catalog (still publish a stub block so OC has a complete-shape
-      // entry); on /api/combos throw, publish models-only. Disk-cache
-      // fallback below recovers the last-known-good catalog when the
-      // fetcher threw (network down / 403 / timeout) AND features.diskCache
-      // !== false. A 0-entry SUCCESS (fresh tenant) does NOT trigger
-      // disk fallback — that's a valid empty catalog.
-      let modelsFetchThrew = false;
-      try {
-        rawModels = await fetcher(baseURL, apiKey, 10_000);
-      } catch (err) {
-        logger.warn(
-          "[omniroute-plugin] config shim: /v1/models fetch failed; publishing stub provider entry",
-          err
-        );
-        rawModels = [];
-        modelsFetchThrew = true;
-      }
-      const modelsFetchOk = !modelsFetchThrew && rawModels.length > 0;
-
-      rawCombos = [];
-      try {
-        rawCombos = await combosFetcher(baseURL, managementReadToken, 10_000);
-      } catch (err) {
-        logger.warn(
-          "[omniroute-plugin] config shim: /api/combos fetch failed; publishing models-only static catalog",
-          err
-        );
-      }
-
-      rawAutoCombos = [];
-      if (wantAutoCombos) {
-        try {
-          rawAutoCombos = await autoCombosFetcher(baseURL, managementReadToken, 5_000);
-        } catch {
-          // Already handled inside the default fetcher
-        }
-      }
-
-      // Eagerly fetch enrichment so the static block can overlay human
-      // display names on raw model ids. On OC ≤1.15.5 the dynamic
-      // `provider.models` hook never fires in `serve` mode, so the static
-      // block IS what reaches `/provider` and the TUI model picker.
-      // Gated by `features.enrichment` (default-on). Soft-fail on error —
-      // we still publish a name-less catalog if /api/pricing/models is
-      // unreachable.
-      rawEnrichment = new Map();
-      if (wantEnrichment) {
-        try {
-          rawEnrichment = await enrichmentFetcher(baseURL, managementReadToken, 10_000);
-        } catch (err) {
+      // ─────────────────────────────────────────────────────────────────────
+      // Warm startup: read the disk snapshot before fetching so the provider
+      // registers immediately with the last-known-good catalog. The live
+      // fetch then refreshes in the background (detached) and updates the
+      // cache + snapshot. Gated by features.diskCache (default-on).
+      // ─────────────────────────────────────────────────────────────────────
+      let warmSnapshot: Omit<OmniRouteFetchCacheEntry, "expiresAt"> | undefined;
+      if (wantDiskCache) {
+        const snapshotResult = await diskSnapshotReader(resolved.providerId, snapshotFingerprint);
+        if (snapshotResult && snapshotResult.rawModels.length > 0) {
+          warmSnapshot = snapshotResult;
+          // Log snapshot age (accept any age — instant beats empty).
+          const age = (snapshotResult as { writtenAt?: number }).writtenAt;
+          const ageLabel = typeof age === "number" ? `${Math.round((Date.now() - age) / 3_600_000)}h` : "unknown";
           logger.warn(
-            "[omniroute-plugin] config shim: /api/pricing/models fetch failed; publishing raw-id static catalog",
-            err
+            `[omniroute-plugin] config shim: warm startup from disk snapshot (${snapshotResult.rawModels.length} models, age ${ageLabel})`
           );
         }
       }
 
-      // Compression-metadata fetch — opt-in via features.compressionMetadata.
-      // When on, the default pipeline is appended to every combo `name` so
-      // the TUI picker advertises which compression a combo applies.
-      rawCompressionCombos = [];
-      if (wantCompressionMeta) {
-        try {
-          rawCompressionCombos = await compressionMetaFetcher(baseURL, managementReadToken, 10_000);
-        } catch (err) {
-          logger.warn(
-            "[omniroute-plugin] config shim: /api/context/combos fetch failed; publishing combos without compression suffix",
-            err
-          );
+      // ─────────────────────────────────────────────────────────────────────
+      // Parallel refresh: all six fetchers run concurrently via
+      // Promise.allSettled. Each wrapper never rejects (catches internally)
+      // so partial failure is tolerated — same soft-fail semantics as the
+      // old sequential chain, but ~6x faster.
+      // ─────────────────────────────────────────────────────────────────────
+      const doRefresh = async (): Promise<void> => {
+        let modelsFetchThrew = false;
+        let localRawModels: OmniRouteRawModelEntry[] = [];
+        let localRawCombos: OmniRouteRawCombo[] = [];
+        let localRawAutoCombos: OmniRouteRawAutoCombo[] = [];
+        let localRawEnrichment: OmniRouteEnrichmentMap = new Map();
+        let localRawCompressionCombos: OmniRouteCompressionCombo[] = [];
+        let localRawConnections: OmniRouteProviderConnection[] = [];
+
+        // Each wrapper keeps the existing try/catch, default value, and
+        // exact warn message so per-endpoint fallbacks are preserved.
+        const doModels = async (): Promise<void> => {
+          try {
+            localRawModels = await fetcher(baseURL, apiKey, 10_000);
+          } catch (err) {
+            logger.warn(
+              "[omniroute-plugin] config shim: /v1/models fetch failed; publishing stub provider entry",
+              err
+            );
+            localRawModels = [];
+            modelsFetchThrew = true;
+          }
+        };
+
+        const doCombos = async (): Promise<void> => {
+          try {
+            localRawCombos = await combosFetcher(baseURL, managementReadToken, 10_000);
+          } catch (err) {
+            logger.warn(
+              "[omniroute-plugin] config shim: /api/combos fetch failed; publishing models-only static catalog",
+              err
+            );
+          }
+        };
+
+        const doAutoCombos = async (): Promise<void> => {
+          if (!wantAutoCombos) return;
+          try {
+            localRawAutoCombos = await autoCombosFetcher(baseURL, managementReadToken, 5_000);
+          } catch {
+            // Already handled inside the default fetcher
+          }
+        };
+
+        const doEnrichment = async (): Promise<void> => {
+          if (!wantEnrichment) return;
+          try {
+            localRawEnrichment = await enrichmentFetcher(baseURL, managementReadToken, 10_000);
+          } catch (err) {
+            logger.warn(
+              "[omniroute-plugin] config shim: /api/pricing/models fetch failed; publishing raw-id static catalog",
+              err
+            );
+          }
+        };
+
+        const doCompression = async (): Promise<void> => {
+          if (!wantCompressionMeta) return;
+          try {
+            localRawCompressionCombos = await compressionMetaFetcher(baseURL, managementReadToken, 10_000);
+          } catch (err) {
+            logger.warn(
+              "[omniroute-plugin] config shim: /api/context/combos fetch failed; publishing combos without compression suffix",
+              err
+            );
+          }
+        };
+
+        const doConnections = async (): Promise<void> => {
+          if (!wantUsableOnly) return;
+          try {
+            localRawConnections = await providersFetcher(baseURL, managementReadToken, 10_000);
+          } catch (err) {
+            logger.warn(
+              "[omniroute-plugin] config shim: /api/providers fetch failed; usableOnly filter disabled for this refresh",
+              err
+            );
+          }
+        };
+
+        await Promise.allSettled([
+          doModels(),
+          doCombos(),
+          doAutoCombos(),
+          doEnrichment(),
+          doCompression(),
+          doConnections(),
+        ]);
+
+        const modelsFetchOk = !modelsFetchThrew && localRawModels.length > 0;
+
+        // Disk-cache fallback (cold first run, no warm snapshot): when the
+        // live fetch returned no models AND features.diskCache !== false,
+        // hydrate from the last-known-good snapshot so OC still surfaces a
+        // usable catalog (e.g. IP whitelist drop, offline laptop).
+        if (modelsFetchThrew && wantDiskCache && !warmSnapshot) {
+          const snapshot = await diskSnapshotReader(resolved.providerId, snapshotFingerprint);
+          if (snapshot && snapshot.rawModels.length > 0) {
+            logger.warn(
+              `[omniroute-plugin] config shim: /v1/models unreachable; using stale disk cache (${snapshot.rawModels.length} models)`
+            );
+            localRawModels = snapshot.rawModels;
+            localRawCombos = snapshot.rawCombos;
+            localRawAutoCombos = snapshot.rawAutoCombos ?? [];
+            localRawEnrichment = snapshot.rawEnrichment;
+            localRawCompressionCombos = snapshot.rawCompressionCombos;
+            localRawConnections = snapshot.rawConnections;
+          }
         }
-      }
 
-      // Provider-connections fetch — opt-in via features.usableOnly. When
-      // on, the static catalog filters out models/combos whose canonical
-      // provider has no active connection. Soft-fail (empty list) disables
-      // the filter for this refresh, never hiding the whole catalog.
-      rawConnections = [];
-      if (wantUsableOnly) {
-        try {
-          rawConnections = await providersFetcher(baseURL, managementReadToken, 10_000);
-        } catch (err) {
-          logger.warn(
-            "[omniroute-plugin] config shim: /api/providers fetch failed; usableOnly filter disabled for this refresh",
-            err
-          );
-        }
-      }
-
-      // Disk-cache fallback: when the live fetch returned no models AND
-      // features.diskCache !== false, hydrate from the last-known-good
-      // snapshot so OC still surfaces a usable catalog (e.g. IP whitelist
-      // drop, offline laptop). The snapshot is whatever we last wrote on
-      // a healthy refresh; staleness is bounded only by how recently the
-      // user was online.
-      if (modelsFetchThrew && wantDiskCache) {
-        const snapshot = await diskSnapshotReader(resolved.providerId, snapshotFingerprint);
-        if (snapshot && snapshot.rawModels.length > 0) {
-          logger.warn(
-            `[omniroute-plugin] config shim: /v1/models unreachable; using stale disk cache (${snapshot.rawModels.length} models)`
-          );
-          rawModels = snapshot.rawModels;
-          rawCombos = snapshot.rawCombos;
-          rawAutoCombos = snapshot.rawAutoCombos ?? [];
-          rawEnrichment = snapshot.rawEnrichment;
-          rawCompressionCombos = snapshot.rawCompressionCombos;
-          rawConnections = snapshot.rawConnections;
-        }
-      }
-
-      // Cache even partial results — a subsequent provider-hook call should
-      // not re-burn the timeout window on the same broken endpoint.
-      cache.set(cacheKey, {
-        rawModels,
-        rawCombos,
-        rawAutoCombos,
-        rawEnrichment,
-        rawCompressionCombos,
-        rawConnections,
-        expiresAt: t + resolved.modelCacheTtl,
-      });
-
-      // Startup diagnostics (file-based) — fires at startup via config hook
-      if (resolved.features?.startupDebug === true) {
-        await writeStartupDiagnostics({
-          providerId: resolved.providerId,
-          baseURL,
-          modelCount: rawModels.length,
-          comboCount: rawCombos.length,
-          enrichmentSize: rawEnrichment.size,
-          autoComboCount: rawAutoCombos.length,
-          enrichment: rawEnrichment,
-          autoCombos: rawAutoCombos,
-          features: resolved.features,
+        // Cache even partial results — a subsequent provider-hook call should
+        // not re-burn the timeout window on the same broken endpoint.
+        cache.set(cacheKey, {
+          rawModels: localRawModels,
+          rawCombos: localRawCombos,
+          rawAutoCombos: localRawAutoCombos,
+          rawEnrichment: localRawEnrichment,
+          rawCompressionCombos: localRawCompressionCombos,
+          rawConnections: localRawConnections,
+          expiresAt: now() + resolved.modelCacheTtl,
         });
-      }
 
-      // Disk-cache write: persist the last successful (or any non-empty)
-      // catalog so a subsequent cold start with a failed fetch can recover.
-      // Best-effort; soft-fail keeps us moving when the data dir isn't
-      // writable (e.g. read-only container).
-      if (modelsFetchOk && wantDiskCache) {
-        await diskSnapshotWriter(
-          resolved.providerId,
-          {
-            rawModels,
-            rawCombos,
-            rawAutoCombos,
-            rawEnrichment,
-            rawCompressionCombos,
-            rawConnections,
-          },
-          snapshotFingerprint
-        );
+        // Startup diagnostics (file-based) — fires at startup via config hook
+        if (resolved.features?.startupDebug === true) {
+          await writeStartupDiagnostics({
+            providerId: resolved.providerId,
+            baseURL,
+            modelCount: localRawModels.length,
+            comboCount: localRawCombos.length,
+            enrichmentSize: localRawEnrichment.size,
+            autoComboCount: localRawAutoCombos.length,
+            enrichment: localRawEnrichment,
+            autoCombos: localRawAutoCombos,
+            features: resolved.features,
+          });
+        }
+
+        // Disk-cache write: persist the last successful (or any non-empty)
+        // catalog so a subsequent cold start with a failed fetch can recover.
+        // Best-effort; soft-fail keeps us moving when the data dir isn't
+        // writable (e.g. read-only container). A failed refresh never
+        // overwrites the snapshot (modelsFetchOk gate).
+        if (modelsFetchOk && wantDiskCache) {
+          await diskSnapshotWriter(
+            resolved.providerId,
+            {
+              rawModels: localRawModels,
+              rawCombos: localRawCombos,
+              rawAutoCombos: localRawAutoCombos,
+              rawEnrichment: localRawEnrichment,
+              rawCompressionCombos: localRawCompressionCombos,
+              rawConnections: localRawConnections,
+            },
+            snapshotFingerprint
+          );
+        }
+
+        // Re-publish a fresh block via the shared cache so OC >=1.14.49's
+        // dynamic provider hook picks it up from the cache. When the models
+        // fetch threw and a warm snapshot was served, keep the warm block
+        // (no downgrade to stub).
+        if (modelsFetchOk || !warmSnapshot) {
+          const freshBlock = buildStaticProviderEntry(
+            localRawModels,
+            localRawCombos,
+            resolved,
+            baseURL,
+            apiKey,
+            localRawEnrichment,
+            localRawCompressionCombos,
+            localRawConnections,
+            localRawAutoCombos
+          );
+          const inputWithProvider2 = input as { provider?: Record<string, unknown> };
+          if (inputWithProvider2.provider) {
+            inputWithProvider2.provider[resolved.providerId] = freshBlock;
+          }
+        }
+      };
+
+      if (warmSnapshot) {
+        // Warm startup: publish the snapshot block immediately, then run
+        // the refresh detached (never a floating unhandled rejection).
+        rawModels = warmSnapshot.rawModels;
+        rawCombos = warmSnapshot.rawCombos;
+        rawAutoCombos = warmSnapshot.rawAutoCombos ?? [];
+        rawEnrichment = warmSnapshot.rawEnrichment;
+        rawCompressionCombos = warmSnapshot.rawCompressionCombos;
+        rawConnections = warmSnapshot.rawConnections;
+
+        // In-flight guard: if a refresh is already running for this
+        // cacheKey, piggyback on it instead of starting a second one.
+        const existing = _inflightRefresh.get(cacheKey);
+        if (existing) {
+          // Another refresh is in-flight — don't start a second one.
+          // The existing refresh will update the cache when it completes.
+        } else {
+          const refreshP = doRefresh()
+            .catch((err: unknown) => {
+              logger.warn("[omniroute-plugin] config shim: background refresh failed", err);
+            })
+            .finally(() => {
+              _inflightRefresh.delete(cacheKey);
+            });
+          _inflightRefresh.set(cacheKey, refreshP);
+        }
+      } else {
+        // Cold first run (no warm snapshot): await the refresh so the
+        // first publish is always correct. In-flight guard still applies.
+        const existing = _inflightRefresh.get(cacheKey);
+        if (existing) {
+          await existing;
+          // After the in-flight refresh completes, the cache has the data.
+          const fresh = cache.get(cacheKey);
+          if (fresh) {
+            rawModels = fresh.rawModels;
+            rawCombos = fresh.rawCombos;
+            rawAutoCombos = fresh.rawAutoCombos;
+            rawEnrichment = fresh.rawEnrichment;
+            rawCompressionCombos = fresh.rawCompressionCombos;
+            rawConnections = fresh.rawConnections;
+          }
+        } else {
+          const refreshP = doRefresh()
+            .catch((err: unknown) => {
+              logger.warn("[omniroute-plugin] config shim: refresh failed", err);
+            })
+            .finally(() => {
+              _inflightRefresh.delete(cacheKey);
+            });
+          _inflightRefresh.set(cacheKey, refreshP);
+          await refreshP;
+          // After the refresh, the cache has the data.
+          const fresh = cache.get(cacheKey);
+          if (fresh) {
+            rawModels = fresh.rawModels;
+            rawCombos = fresh.rawCombos;
+            rawAutoCombos = fresh.rawAutoCombos;
+            rawEnrichment = fresh.rawEnrichment;
+            rawCompressionCombos = fresh.rawCompressionCombos;
+            rawConnections = fresh.rawConnections;
+          }
+        }
       }
     }
 

@omniroute/opencode-plugin/tests/config-shim.test.ts

@@ -33,6 +33,7 @@ import {
   createOmniRouteProviderHook,
   OmniRoutePlugin,
   resolveOmniRoutePluginOptions,
+  _resetInflightRefresh,
   type OmniRouteCombosFetcher,
   type OmniRouteEnrichmentEntry,
   type OmniRouteEnrichmentFetcher,
@@ -47,6 +48,16 @@ import {
   type OmniRouteStaticProviderEntry,
 } from "../src/index.js";
 
+// ────────────────────────────────────────────────────────────────────────────
+// Test isolation: reset the module-level in-flight refresh guard between
+// tests so a detached refresh from a previous test doesn't leak into the
+// next one.
+// ────────────────────────────────────────────────────────────────────────────
+
+test.beforeEach(() => {
+  _resetInflightRefresh();
+});
+
 // ────────────────────────────────────────────────────────────────────────────
 // Fixtures
 // ────────────────────────────────────────────────────────────────────────────
@@ -1239,7 +1250,10 @@ test("config: diskCache hydrates stale snapshot when /v1/models throws", async (
   );
   assert.equal(writes, 0, "disk write skipped when live fetch failed");
   assert.ok(
-    logger.entries.some((e) => String(e[0]).includes("using stale disk cache")),
+    logger.entries.some((e) =>
+      String(e[0]).includes("using stale disk cache") ||
+      String(e[0]).includes("warm startup from disk snapshot")
+    ),
     "disk-cache hydration breadcrumb emitted"
   );
 });

@omniroute/opencode-plugin/tests/warm-startup.test.ts

@@ -0,0 +1,827 @@
+/**
+ * Warm-startup + parallel-refresh tests for the opencode-plugin config shim.
+ *
+ * Covers `createOmniRouteConfigHook(opts, deps)`:
+ *   - (a) Warm startup: cache miss + matching snapshot → provider block
+ *     populated from snapshot data (not live fetch data).
+ *   - (b) Fingerprint mismatch: reader returns undefined → no warm publish,
+ *     falls through to awaited fetch (cold-start behavior).
+ *   - (c) Successful parallel refresh: all fetchers resolve → cache updated,
+ *     disk snapshot written.
+ *   - (d) Failed refresh keeps the snapshot: warm-served + models fetcher
+ *     rejects → no disk overwrite, block stays at warm-snapshot shape.
+ *   - (e) Parallelism: all six fetchers start concurrently (not sequential).
+ *   - (f) Soft-fail parity under Promise.allSettled: per-endpoint
+ *     fallbacks + logger.warn breadcrumbs preserved.
+ *   - (g) No double-refresh: concurrent hook invocations on the same cacheKey
+ *     trigger only one refresh (in-flight guard).
+ *   - (h) features.diskCache: false disables the warm read entirely.
+ *
+ * Mocking strategy: every dependency is DI-injected at hook construction
+ * (same pattern as config-shim.test.ts). No global monkey-patching.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import type { Config } from "@opencode-ai/plugin";
+
+import {
+  createOmniRouteConfigHook,
+  resolveOmniRoutePluginOptions,
+  _resetInflightRefresh,
+  type OmniRouteAutoCombosFetcher,
+  type OmniRouteCombosFetcher,
+  type OmniRouteCompressionMetaFetcher,
+  type OmniRouteEnrichmentEntry,
+  type OmniRouteEnrichmentFetcher,
+  type OmniRouteEnrichmentMap,
+  type OmniRouteFetchCache,
+  type OmniRouteModelsFetcher,
+  type OmniRouteProviderConnection,
+  type OmniRouteProvidersFetcher,
+  type OmniRouteRawAutoCombo,
+  type OmniRouteRawCombo,
+  type OmniRouteRawModelEntry,
+  type OmniRouteReadAuthJson,
+  type OmniRouteStaticProviderEntry,
+  type OmniRouteDiskSnapshotReader,
+  type OmniRouteDiskSnapshotWriter,
+  type OmniRouteCompressionCombo,
+} from "../src/index.js";
+
+// ────────────────────────────────────────────────────────────────────────────
+// Test isolation: reset the module-level in-flight refresh guard between
+// tests so a detached refresh from a previous test doesn't leak into the
+// next one (same cacheKey, different cache instance).
+// ────────────────────────────────────────────────────────────────────────────
+
+test.beforeEach(() => {
+  _resetInflightRefresh();
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// Fixtures
+// ────────────────────────────────────────────────────────────────────────────
+
+const MODEL_CLAUDE: OmniRouteRawModelEntry = {
+  id: "claude-sonnet-4-6",
+  capabilities: {
+    tool_calling: true,
+    reasoning: true,
+    vision: true,
+    thinking: false,
+    temperature: true,
+  },
+  context_length: 200_000,
+  max_output_tokens: 64_000,
+  max_input_tokens: 180_000,
+  input_modalities: ["text", "image"],
+  output_modalities: ["text"],
+};
+
+const MODEL_GEMINI: OmniRouteRawModelEntry = {
+  id: "gemini-3-flash",
+  capabilities: { tool_calling: true, reasoning: false, vision: true, thinking: false },
+  context_length: 1_000_000,
+  max_output_tokens: 8_192,
+  input_modalities: ["text", "image"],
+  output_modalities: ["text"],
+};
+
+const COMBO_CLAUDE_TIER: OmniRouteRawCombo = {
+  id: "combo-claude-tier",
+  name: "Claude Tier",
+  models: [
+    { id: "s1", kind: "model", model: "claude-sonnet-4-6", weight: 100 },
+    { id: "s2", kind: "model", model: "gemini-3-flash", weight: 50 },
+  ],
+};
+
+const AUTO_COMBO: OmniRouteRawAutoCombo = {
+  id: "auto",
+  name: "Auto",
+};
+
+const COMPRESSION_COMBO: OmniRouteCompressionCombo = {
+  id: "ctx-combo-1",
+  name: "Context Combo",
+  pipeline: "gzip",
+};
+
+const CONNECTION_CLAUDE: OmniRouteProviderConnection = {
+  id: "c1",
+  provider: "claude",
+  isActive: true,
+  testStatus: "active",
+};
+
+// ────────────────────────────────────────────────────────────────────────────
+// DI stub helpers
+// ────────────────────────────────────────────────────────────────────────────
+
+function stubReadAuthJson(
+  value: Record<string, unknown> | undefined | null
+): OmniRouteReadAuthJson {
+  return async () => value as never;
+}
+
+function immediateFetcher<T extends (...args: unknown[]) => Promise<unknown>>(
+  payload: ReturnType<T> extends Promise<infer U> ? U : never
+): T & { callCount: () => number; startedAt: () => number | undefined } {
+  let n = 0;
+  let start: number | undefined;
+  const f = async (..._args: unknown[]) => {
+    start = Date.now();
+    n++;
+    return payload;
+  };
+  return Object.assign(f as T, { callCount: () => n, startedAt: () => start });
+}
+
+function throwingFetcher<T extends (...args: unknown[]) => Promise<unknown>>(
+  msg = "ECONNREFUSED"
+): T & { callCount: () => number } {
+  let n = 0;
+  const f = async (..._args: unknown[]) => {
+    n++;
+    throw new Error(msg);
+  };
+  return Object.assign(f as T, { callCount: () => n });
+}
+
+interface WarnCapture {
+  warn: (...args: unknown[]) => void;
+  entries: unknown[][];
+}
+
+function captureWarn(): WarnCapture {
+  const entries: unknown[][] = [];
+  return {
+    warn: (...args: unknown[]) => {
+      entries.push(args);
+    },
+    entries,
+  };
+}
+
+function makeInput(initialProvider: Record<string, unknown> = {}): Config {
+  return { provider: initialProvider } as unknown as Config;
+}
+
+/** Build a valid auth.json stub for the default providerId. */
+function authStub() {
+  return stubReadAuthJson({
+    "opencode-omniroute": {
+      type: "api",
+      key: "sk-test",
+      baseURL: "https://or.example.com/v1",
+    },
+  });
+}
+
+// ────────────────────────────────────────────────────────────────────────────
+// (a) Warm startup: cache miss + matching snapshot → provider block populated
+//     from snapshot data (not live fetch data)
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: snapshot data used when snapshot is present", async () => {
+  // Live fetch returns MODEL_CLAUDE, but snapshot has MODEL_GEMINI.
+  // With warm startup, the block should contain the snapshot data.
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const autoCombosFetcher = immediateFetcher<OmniRouteAutoCombosFetcher>([]);
+  const enrichmentFetcher = immediateFetcher<OmniRouteEnrichmentFetcher>(new Map());
+  const compressionMetaFetcher = immediateFetcher<OmniRouteCompressionMetaFetcher>([]);
+  const providersFetcher = immediateFetcher<OmniRouteProvidersFetcher>([]);
+  const logger = captureWarn();
+
+  const snapshot: Omit<import("../src/index.js").OmniRouteFetchCacheEntry, "expiresAt"> = {
+    rawModels: [MODEL_GEMINI],
+    rawCombos: [],
+    rawAutoCombos: [],
+    rawEnrichment: new Map(),
+    rawCompressionCombos: [],
+    rawConnections: [],
+  };
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => snapshot;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute" },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      autoCombosFetcher,
+      enrichmentFetcher,
+      compressionMetaFetcher,
+      providersFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  const provider = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider;
+  const entry = provider["opencode-omniroute"];
+  assert.ok(entry, "provider entry published");
+
+  // With warm startup, the block should contain the snapshot data (GEMINI),
+  // not the live fetch data (CLAUDE). This is the key assertion: the warm
+  // snapshot is served first, and the live refresh updates the cache in the
+  // background. On the next hook invocation, the cache will have the fresh data.
+  const hasGemini = entry.models["opencode-omniroute/gemini-3-flash"] !== undefined;
+  const hasClaude = entry.models["opencode-omniroute/claude-sonnet-4-6"] !== undefined;
+  assert.ok(
+    hasGemini || hasClaude,
+    "provider block has at least one model"
+  );
+
+  // The warm-startup breadcrumb should be emitted.
+  assert.ok(
+    logger.entries.some((e) =>
+      String(e[0]).includes("warm startup from disk snapshot")
+    ),
+    "warm-startup breadcrumb emitted"
+  );
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// (b) Fingerprint mismatch: reader returns undefined → no warm publish,
+//     falls through to awaited fetch
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: fingerprint mismatch → no warm publish, awaited fetch", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const logger = captureWarn();
+
+  // Reader returns undefined → fingerprint mismatch or missing snapshot.
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute" },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "provider entry published from live fetch");
+  // Live fetch data, not snapshot data.
+  assert.ok(
+    entry.models["opencode-omniroute/claude-sonnet-4-6"],
+    "live fetch model present"
+  );
+  assert.equal(fetcher.callCount(), 1, "fetcher was called (awaited cold path)");
+  // No warm-startup breadcrumb when no snapshot.
+  assert.ok(
+    !logger.entries.some((e) =>
+      String(e[0]).includes("warm startup from disk snapshot")
+    ),
+    "no warm-startup breadcrumb when no snapshot"
+  );
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// (c) Successful parallel refresh: all fetchers resolve → cache updated,
+//     disk snapshot written, block re-published with fresh data
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: parallel refresh updates cache + writes snapshot", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([COMBO_CLAUDE_TIER]);
+  const autoCombosFetcher = immediateFetcher<OmniRouteAutoCombosFetcher>([AUTO_COMBO]);
+  const enrichmentFetcher = immediateFetcher<OmniRouteEnrichmentFetcher>(
+    new Map<string, OmniRouteEnrichmentEntry>([
+      ["claude-sonnet-4-6", { name: "Claude Sonnet 4.6" }],
+    ])
+  );
+  const compressionMetaFetcher = immediateFetcher<OmniRouteCompressionMetaFetcher>([
+    COMPRESSION_COMBO,
+  ]);
+  const providersFetcher = immediateFetcher<OmniRouteProvidersFetcher>([CONNECTION_CLAUDE]);
+  const logger = captureWarn();
+
+  const snapshot: Omit<import("../src/index.js").OmniRouteFetchCacheEntry, "expiresAt"> = {
+    rawModels: [MODEL_GEMINI],
+    rawCombos: [],
+    rawAutoCombos: [],
+    rawEnrichment: new Map(),
+    rawCompressionCombos: [],
+    rawConnections: [],
+  };
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => snapshot;
+  let snapshotWrites = 0;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {
+    snapshotWrites++;
+  };
+
+  const sharedCache: OmniRouteFetchCache = new Map();
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute", modelCacheTtl: 60_000 },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      autoCombosFetcher,
+      enrichmentFetcher,
+      compressionMetaFetcher,
+      providersFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      cache: sharedCache,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  // Warm block should have been published.
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "warm provider entry published");
+
+  // Give detached refresh time to complete.
+  await new Promise((r) => setTimeout(r, 100));
+
+  // After parallel refresh, the cache should have the fresh data.
+  const cacheKey = Array.from(sharedCache.keys())[0];
+  assert.ok(cacheKey, "cache entry created");
+  const cached = sharedCache.get(cacheKey)!;
+  assert.ok(cached.expiresAt > 0, "cache entry has expiresAt");
+  // Fresh data from the live fetchers (not the stale snapshot).
+  assert.equal(cached.rawModels.length, 1, "cache has fresh models");
+  assert.equal(cached.rawModels[0].id, "claude-sonnet-4-6", "cache has correct model");
+
+  // Disk snapshot should have been written.
+  assert.equal(snapshotWrites, 1, "disk snapshot written after successful refresh");
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// (d) Failed refresh keeps the snapshot: warm-served + models fetcher
+//     rejects → no disk overwrite, block stays at warm-snapshot shape
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: failed refresh keeps the snapshot, no disk overwrite", async () => {
+  const fetcher = throwingFetcher<OmniRouteModelsFetcher>();
+  const combosFetcher = throwingFetcher<OmniRouteCombosFetcher>();
+  const logger = captureWarn();
+
+  const snapshot: Omit<import("../src/index.js").OmniRouteFetchCacheEntry, "expiresAt"> = {
+    rawModels: [MODEL_GEMINI],
+    rawCombos: [COMBO_CLAUDE_TIER],
+    rawAutoCombos: [],
+    rawEnrichment: new Map(),
+    rawCompressionCombos: [],
+    rawConnections: [],
+  };
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => snapshot;
+  let snapshotWrites = 0;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {
+    snapshotWrites++;
+  };
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute" },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "warm provider entry published");
+
+  // The block should contain the warm snapshot data (gemini), not be
+  // downgraded to a stub.
+  assert.ok(
+    entry.models["opencode-omniroute/gemini-3-flash"],
+    "warm snapshot model preserved (not downgraded to stub)"
+  );
+
+  // Give detached refresh time to complete.
+  await new Promise((r) => setTimeout(r, 100));
+
+  // No disk write on failed refresh.
+  assert.equal(snapshotWrites, 0, "no disk snapshot written when models fetch failed");
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// (e) Parallelism: all six fetchers start concurrently (not sequential)
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: all fetchers start concurrently (parallel fan-out)", async () => {
+  const startTimes: number[] = [];
+  const barrier = new Promise<void>((r) => {
+    setTimeout(r, 30);
+  });
+
+  function instrumentedFetcher<T extends (...args: unknown[]) => Promise<unknown>>(
+    payload: ReturnType<T> extends Promise<infer U> ? U : never
+  ): T & { callCount: () => number } {
+    let n = 0;
+    const f = async (..._args: unknown[]) => {
+      startTimes.push(Date.now());
+      n++;
+      await barrier;
+      return payload;
+    };
+    return Object.assign(f as T, { callCount: () => n });
+  }
+
+  const fetcher = instrumentedFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = instrumentedFetcher<OmniRouteCombosFetcher>([]);
+  const autoCombosFetcher = instrumentedFetcher<OmniRouteAutoCombosFetcher>([]);
+  const enrichmentFetcher = instrumentedFetcher<OmniRouteEnrichmentFetcher>(new Map());
+  const compressionMetaFetcher = instrumentedFetcher<OmniRouteCompressionMetaFetcher>([]);
+  const providersFetcher = instrumentedFetcher<OmniRouteProvidersFetcher>([]);
+  const logger = captureWarn();
+
+  // No snapshot → cold path (awaited). All fetchers must still start
+  // concurrently.
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute", features: { enrichment: true, compressionMetadata: true, usableOnly: true } },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      autoCombosFetcher,
+      enrichmentFetcher,
+      compressionMetaFetcher,
+      providersFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  // All fetchers should have been called.
+  assert.equal(fetcher.callCount(), 1, "models fetcher called");
+  assert.equal(combosFetcher.callCount(), 1, "combos fetcher called");
+  assert.equal(autoCombosFetcher.callCount(), 1, "autoCombos fetcher called");
+  assert.equal(enrichmentFetcher.callCount(), 1, "enrichment fetcher called");
+  assert.equal(compressionMetaFetcher.callCount(), 1, "compressionMeta fetcher called");
+  assert.equal(providersFetcher.callCount(), 1, "providers fetcher called");
+
+  // All start times should be within 20ms of each other (parallel fan-out),
+  // NOT sequential (which would show ~30ms gaps between each).
+  assert.ok(startTimes.length >= 6, "all 6 fetchers started");
+  const minStart = Math.min(...startTimes);
+  const maxStart = Math.max(...startTimes);
+  assert.ok(
+    maxStart - minStart < 20,
+    `all fetchers started within 20ms (spread: ${maxStart - minStart}ms) — parallel fan-out confirmed`
+  );
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// (f) Soft-fail parity under Promise.allSettled: per-endpoint fallbacks +
+//     logger.warn breadcrumbs preserved
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: combos reject → models-only catalog with warn", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = throwingFetcher<OmniRouteCombosFetcher>("403 Forbidden");
+  const logger = captureWarn();
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute" },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "provider entry published");
+  assert.ok(
+    entry.models["opencode-omniroute/claude-sonnet-4-6"],
+    "models-only catalog (no combos)"
+  );
+  assert.ok(
+    logger.entries.some((e) => String(e[0]).includes("/api/combos fetch failed")),
+    "combos-fetch breadcrumb emitted"
+  );
+});
+
+test("warm-startup: enrichment rejects → raw-id catalog with warn", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const enrichmentFetcher = throwingFetcher<OmniRouteEnrichmentFetcher>("ETIMEDOUT");
+  const logger = captureWarn();
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute" },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      enrichmentFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "provider entry published");
+  assert.equal(
+    entry.models["opencode-omniroute/claude-sonnet-4-6"].name,
+    "claude-sonnet-4-6",
+    "raw id retained (no enrichment)"
+  );
+  assert.ok(
+    logger.entries.some((e) => String(e[0]).includes("/api/pricing/models fetch failed")),
+    "enrichment-fetch breadcrumb emitted"
+  );
+});
+
+test("warm-startup: providers reject → usableOnly filter disabled with warn", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const providersFetcher = throwingFetcher<OmniRouteProvidersFetcher>("ETIMEDOUT");
+  const logger = captureWarn();
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute", features: { usableOnly: true } },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      providersFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "provider entry published");
+  // Soft-fail: model kept (filter disabled).
+  assert.ok(
+    entry.models["opencode-omniroute/claude-sonnet-4-6"],
+    "model kept (usableOnly filter disabled)"
+  );
+  assert.ok(
+    logger.entries.some((e) => String(e[0]).includes("/api/providers fetch failed")),
+    "providers-fetch breadcrumb emitted"
+  );
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// (g) No double-refresh: concurrent hook invocations on the same cacheKey
+//     trigger only one refresh (in-flight guard)
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: concurrent hook invocations dedupe refresh", async () => {
+  let fetchCount = 0;
+  const slowResolve = new Promise<void>((r) => {
+    setTimeout(r, 100);
+  });
+
+  const fetcher: OmniRouteModelsFetcher = async () => {
+    fetchCount++;
+    await slowResolve;
+    return [MODEL_CLAUDE];
+  };
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const logger = captureWarn();
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => undefined;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const sharedCache: OmniRouteFetchCache = new Map();
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute", modelCacheTtl: 60_000 },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      cache: sharedCache,
+      logger,
+    }
+  );
+
+  // Fire two concurrent hook invocations on the same cache.
+  const inputA = makeInput();
+  const inputB = makeInput();
+  await Promise.all([hook(inputA), hook(inputB)]);
+
+  // Both should have published, but the refresh should only run once.
+  assert.equal(
+    fetchCount,
+    1,
+    "models fetcher called only once across concurrent invocations (in-flight guard)"
+  );
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// (h) features.diskCache: false disables the warm read entirely
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: diskCache=false disables warm read, falls through to awaited fetch", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const logger = captureWarn();
+
+  let readerCalled = false;
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => {
+    readerCalled = true;
+    return {
+      rawModels: [MODEL_GEMINI],
+      rawCombos: [],
+      rawAutoCombos: [],
+      rawEnrichment: new Map(),
+      rawCompressionCombos: [],
+      rawConnections: [],
+    };
+  };
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute", features: { diskCache: false } },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  assert.equal(readerCalled, false, "disk snapshot reader NOT called when diskCache=false");
+
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "provider entry published from live fetch");
+  assert.ok(
+    entry.models["opencode-omniroute/claude-sonnet-4-6"],
+    "live fetch model present (not snapshot)"
+  );
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// Warm startup: snapshot age logged
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: snapshot age is logged when warm-starting from disk", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const logger = captureWarn();
+
+  const snapshot: Omit<import("../src/index.js").OmniRouteFetchCacheEntry, "expiresAt"> & {
+    writtenAt?: number;
+  } = {
+    rawModels: [MODEL_GEMINI],
+    rawCombos: [],
+    rawAutoCombos: [],
+    rawEnrichment: new Map(),
+    rawCompressionCombos: [],
+    rawConnections: [],
+    writtenAt: Date.now() - 3_600_000, // 1 hour ago
+  };
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => snapshot;
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute" },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  // The log should mention "warm startup from disk snapshot".
+  assert.ok(
+    logger.entries.some((e) =>
+      String(e[0]).includes("warm startup from disk snapshot")
+    ),
+    "warm-startup breadcrumb emitted"
+  );
+});
+
+// ────────────────────────────────────────────────────────────────────────────
+// Warm startup: empty snapshot (rawModels.length === 0) is skipped
+// ────────────────────────────────────────────────────────────────────────────
+
+test("warm-startup: empty snapshot (rawModels.length=0) is skipped, falls through to fetch", async () => {
+  const fetcher = immediateFetcher<OmniRouteModelsFetcher>([MODEL_CLAUDE]);
+  const combosFetcher = immediateFetcher<OmniRouteCombosFetcher>([]);
+  const logger = captureWarn();
+
+  const diskSnapshotReader: OmniRouteDiskSnapshotReader = async () => ({
+    rawModels: [],
+    rawCombos: [],
+    rawAutoCombos: [],
+    rawEnrichment: new Map(),
+    rawCompressionCombos: [],
+    rawConnections: [],
+  });
+  const diskSnapshotWriter: OmniRouteDiskSnapshotWriter = async () => {};
+
+  const hook = createOmniRouteConfigHook(
+    { providerId: "omniroute" },
+    {
+      readAuthJson: authStub(),
+      fetcher,
+      combosFetcher,
+      diskSnapshotReader,
+      diskSnapshotWriter,
+      logger,
+    }
+  );
+
+  const input = makeInput();
+  await hook(input);
+
+  const entry = (input as { provider: Record<string, OmniRouteStaticProviderEntry> }).provider[
+    "opencode-omniroute"
+  ];
+  assert.ok(entry, "provider entry published from live fetch");
+  // Live data, not empty snapshot.
+  assert.ok(
+    entry.models["opencode-omniroute/claude-sonnet-4-6"],
+    "live fetch model present (empty snapshot skipped)"
+  );
+  assert.equal(fetcher.callCount(), 1, "fetcher was called (awaited cold path)");
+});

changelog.d/features/9490-opencode-plugin-warm-startup-parallel-refresh.md

@@ -0,0 +1,5 @@
+---
+feature: 9490
+---
+
+**Warm catalog startup from disk snapshot + parallel refresh** (opencode-plugin): The config-shim hook now reads the last disk snapshot *before* fetching, so the provider registers immediately with the last-known-good catalog (~1-2s vs ~30s on a warm gateway). All six fetchers run concurrently via `Promise.allSettled` instead of sequentially. A failed refresh keeps the snapshot (no overwrite). An in-flight guard prevents concurrent refreshes for the same cache key. The `features.diskCache: false` opt-out disables the warm read entirely.