test(growth,runtime,skills): 深度验证测试 Phase 1-2 — 20 个新测试

- MockLlmDriver 基础设施 (zclaw-runtime/src/test_util.rs)
- 经验闭环 E-01~06: 累积/溢出/反序列化/跨行业/并发/阈值
- Embedding 管道 EM-01~08: 路由/降级/维度不匹配/空查询/CJK/LLM Fallback/热更新
- Skill 执行 SK-01~03: 工具传递/纯 Prompt/锁竞争

crates/zclaw-growth/tests/memory_embedding_test.rs

@@ -0,0 +1,143 @@
+//! Memory embedding tests (EM-07 ~ EM-08)
+//!
+//! Validates memory retrieval with embedding enhancement and configuration hot-update.
+
+use std::sync::Arc;
+use async_trait::async_trait;
+use zclaw_growth::{
+    EmbeddingClient, MemoryEntry, MemoryRetriever, MemoryType, SqliteStorage, VikingAdapter,
+};
+use zclaw_types::AgentId;
+
+/// Mock embedding client that returns deterministic 128-dim vectors.
+struct MockEmbeddingClient {
+    dim: usize,
+}
+
+impl MockEmbeddingClient {
+    fn new() -> Self {
+        Self { dim: 128 }
+    }
+}
+
+#[async_trait::async_trait]
+impl EmbeddingClient for MockEmbeddingClient {
+    async fn embed(&self, text: &str) -> Result<Vec<f32>, String> {
+        let mut vec = vec![0.0f32; self.dim];
+        for (i, b) in text.as_bytes().iter().enumerate() {
+            vec[i % self.dim] += (*b as f32) / 255.0;
+        }
+        let norm: f32 = vec.iter().map(|v| v * v).sum::<f32>().sqrt().max(1e-8);
+        for v in vec.iter_mut() {
+            *v /= norm;
+        }
+        Ok(vec)
+    }
+
+    fn is_available(&self) -> bool {
+        true
+    }
+}
+
+/// EM-07: Memory retrieval with embedding enhancement.
+#[tokio::test]
+async fn em07_memory_retrieval_embedding_enhanced() {
+    let storage = Arc::new(SqliteStorage::in_memory().await);
+    let adapter = Arc::new(VikingAdapter::new(storage));
+
+    let agent_id = AgentId::new();
+
+    // Store 20 mixed Chinese/English memories
+    let entries = vec![
+        ("pref-theme", MemoryType::Preference, "用户偏好深色模式"),
+        ("pref-language", MemoryType::Preference, "用户使用中文沟通"),
+        ("know-rust", MemoryType::Knowledge, "Rust async programming with tokio"),
+        ("know-python", MemoryType::Knowledge, "Python data science with pandas"),
+        ("exp-report", MemoryType::Experience, "月度报表生成经验：使用Excel宏自动化"),
+        ("know-react", MemoryType::Knowledge, "React hooks patterns"),
+        ("pref-editor", MemoryType::Preference, "偏好 VS Code 编辑器"),
+        ("exp-schedule", MemoryType::Experience, "排班冲突解决方案：协商调换"),
+        ("know-sql", MemoryType::Knowledge, "SQL query optimization techniques"),
+        ("exp-deploy", MemoryType::Experience, "部署失败经验：端口冲突检测"),
+        ("know-docker", MemoryType::Knowledge, "Docker container networking"),
+        ("pref-font", MemoryType::Preference, "字体大小偏好 14px"),
+        ("know-tokio", MemoryType::Knowledge, "Tokio runtime configuration"),
+        ("exp-review", MemoryType::Experience, "代码审查经验：关注错误处理"),
+        ("know-git", MemoryType::Knowledge, "Git rebase vs merge strategies"),
+        ("exp-perf", MemoryType::Experience, "性能优化经验：数据库索引"),
+        ("pref-timezone", MemoryType::Preference, "时区 UTC+8"),
+        ("know-linux", MemoryType::Knowledge, "Linux system administration basics"),
+        ("exp-test", MemoryType::Experience, "测试经验：TDD方法论实践"),
+        ("know-api", MemoryType::Knowledge, "RESTful API design principles"),
+    ];
+
+    for (key, mtype, content) in &entries {
+        let entry = MemoryEntry::new(
+            &agent_id.to_string(),
+            *mtype,
+            key,
+            content.to_string(),
+        );
+        adapter.store(&entry).await.unwrap();
+    }
+
+    // Create retriever with embedding
+    let retriever = MemoryRetriever::new(adapter);
+    retriever.set_embedding_client(Arc::new(MockEmbeddingClient::new()));
+
+    // Retrieve memories about user preferences
+    let result = retriever
+        .retrieve(&agent_id, "我之前说过什么偏好？")
+        .await
+        .unwrap();
+
+    let total =
+        result.knowledge.len() + result.preferences.len() + result.experience.len();
+    assert!(
+        total > 0,
+        "embedding-enhanced retrieval should find memories"
+    );
+
+    assert!(
+        result.preferences.len() > 0,
+        "should find preference memories"
+    );
+}
+
+/// EM-08: Embedding configuration hot update — no panic, no disruption.
+#[tokio::test]
+async fn em08_embedding_hot_update() {
+    let storage = Arc::new(SqliteStorage::in_memory().await);
+    let adapter = Arc::new(VikingAdapter::new(storage));
+
+    let agent_id = AgentId::new();
+
+    // Store a memory
+    let entry = MemoryEntry::new(
+        &agent_id.to_string(),
+        MemoryType::Knowledge,
+        "rust-async",
+        "Tokio runtime uses work-stealing scheduler".to_string(),
+    );
+    adapter.store(&entry).await.unwrap();
+
+    // Start without embedding
+    let retriever = MemoryRetriever::new(adapter);
+
+    // Retrieve without embedding — should not panic
+    let _result_before = retriever
+        .retrieve(&agent_id, "async runtime")
+        .await
+        .unwrap();
+
+    // Hot-update with embedding — should not disrupt ongoing operations
+    retriever.set_embedding_client(Arc::new(MockEmbeddingClient::new()));
+
+    // Retrieve with embedding — should not panic
+    let _result_after = retriever
+        .retrieve(&agent_id, "async runtime")
+        .await
+        .unwrap();
+
+    // Key assertion: hot-update does not panic or disrupt
+}