fix: 修正 logo 中 N 和 G 字母造型

N 添加对角线笔画（█▄ █），G 添加内横杠（█ ▀█），
避免与 O 字母造型雷同。同步更新 ui.ts 中的硬编码 wordmark。

packages/llm/test/provider/anthropic-messages-cache.recorded.test.ts

@@ -0,0 +1,53 @@
+import { describe, expect } from "bun:test"
+import { Effect } from "effect"
+import { CacheHint, LLM } from "../../src"
+import { LLMClient } from "../../src/route"
+import * as Anthropic from "../../src/providers/anthropic"
+import { LARGE_CACHEABLE_SYSTEM } from "../recorded-scenarios"
+import { recordedTests } from "../recorded-test"
+
+const model = Anthropic.configure({
+  apiKey: process.env.ANTHROPIC_API_KEY ?? "fixture",
+}).model("claude-haiku-4-5-20251001")
+
+// Two identical generations in a row. The first call writes the prefix into
+// Anthropic's cache; the second should report a cache read against the same
+// prefix. Cassette captures both interactions in order.
+const cacheRequest = LLM.request({
+  id: "recorded_anthropic_cache",
+  model,
+  system: [{ type: "text", text: LARGE_CACHEABLE_SYSTEM, cache: new CacheHint({ type: "ephemeral" }) }],
+  prompt: "Say hi.",
+  // Manual hint on the system part is the only marker we want here — skip the
+  // auto-policy's latest-user-message breakpoint so the cassette body matches.
+  cache: "none",
+  generation: { maxTokens: 16, temperature: 0 },
+})
+
+const recorded = recordedTests({
+  prefix: "anthropic-messages-cache",
+  provider: "anthropic",
+  protocol: "anthropic-messages",
+  requires: ["ANTHROPIC_API_KEY"],
+  // Two identical requests in one cassette — replay walks the cassette in
+  // recording order so the second call replays the cached-hit interaction.
+  options: {
+    redact: { allowRequestHeaders: ["anthropic-version"] },
+  },
+})
+
+describe("Anthropic Messages cache recorded", () => {
+  recorded.effect.with("writes then reads cache_control on identical second call", { tags: ["cache"] }, () =>
+    Effect.gen(function* () {
+      const first = yield* LLMClient.generate(cacheRequest)
+      // The first call may write the cache (cacheWriteInputTokens > 0) or it
+      // may be a fresh miss (both fields 0) depending on whether the prefix is
+      // already warm on Anthropic's side. The assertion that matters is that
+      // the SECOND call reports a non-zero cache read.
+      expect(first.usage?.cacheReadInputTokens ?? 0).toBeGreaterThanOrEqual(0)
+
+      const second = yield* LLMClient.generate(cacheRequest)
+      expect(second.usage?.cacheReadInputTokens ?? 0).toBeGreaterThan(0)
+    }),
+  )
+})