feat(growth,kernel,runtime): Embedding 接通 + 自学习自动化 — A线+B线 6 项实现

A线 Embedding 接通:
- A1: MemoryRetriever.set_embedding_client() + GrowthIntegration.configure_embedding()
  + Kernel.set_embedding_client() + viking_configure_embedding 传播到 Kernel
- A2: Skill 路由替换 new_tf_idf_only() 为 EmbeddingAdapter + LlmSkillFallback

B线 自学习自动化:
- B1: evolution_bridge.rs — candidate_to_manifest() (PromptOnly, disabled by default)
- B2: Kernel::generate_and_register_skill() 全链路 (LLM→parse→QualityGate→manifest→persist)
- B3: EvolutionMiddleware 双模式 (auto_mode 跳过注入, 留给 kernel 自动处理)
- B4: QualityGate 加固 (body ≥100字符 + 必须含标题 + 置信度上限 1.0)

验证: 934 tests PASS, 0 failures

crates/zclaw-kernel/src/kernel/skills.rs

@@ -76,4 +76,77 @@ impl Kernel {
         }
         self.skills.execute(&zclaw_types::SkillId::new(id), &ctx, input).await
     }
+
+    /// Generate a skill from an aggregated pattern and register it.
+    ///
+    /// Full pipeline:
+    /// 1. Build LLM prompt from pattern
+    /// 2. Call LLM to get JSON response
+    /// 3. Parse response into SkillCandidate
+    /// 4. Validate through QualityGate (threshold 0.85 for auto-mode)
+    /// 5. Convert to SkillManifest (PromptOnly, disabled by default)
+    /// 6. Persist to disk via SkillRegistry
+    pub async fn generate_and_register_skill(
+        &self,
+        pattern: &zclaw_growth::pattern_aggregator::AggregatedPattern,
+    ) -> Result<String> {
+        // 1. Build prompt
+        let prompt = zclaw_growth::skill_generator::SkillGenerator::build_prompt(pattern);
+
+        // 2. Call LLM
+        let request = zclaw_runtime::driver::CompletionRequest {
+            model: self.driver.provider().to_string(),
+            system: Some("你是技能设计专家，只返回 JSON 格式的技能定义。".to_string()),
+            messages: vec![zclaw_types::Message::user(prompt)],
+            max_tokens: Some(1024),
+            temperature: Some(0.3),
+            stream: false,
+            ..Default::default()
+        };
+
+        let response = self.driver.complete(request).await?;
+        let text = response.content.iter()
+            .filter_map(|block| match block {
+                zclaw_runtime::driver::ContentBlock::Text { text } => Some(text.as_str()),
+                _ => None,
+            })
+            .collect::<Vec<_>>()
+            .join("");
+
+        // 3. Parse into SkillCandidate
+        let candidate = zclaw_growth::skill_generator::SkillGenerator::parse_response(
+            &text, pattern,
+        )?;
+
+        // 4. Validate through QualityGate (higher threshold for auto-generation)
+        let existing_triggers: Vec<String> = self.skills.list().await
+            .into_iter()
+            .flat_map(|m| m.triggers)
+            .collect();
+        let gate = zclaw_growth::quality_gate::QualityGate::new(0.85, existing_triggers);
+        let report = gate.validate_skill(&candidate);
+        if !report.passed {
+            return Err(zclaw_types::ZclawError::ConfigError(format!(
+                "QualityGate rejected: {}", report.issues.join("; ")
+            )));
+        }
+
+        // 5. Convert to SkillManifest (PromptOnly, disabled)
+        let manifest = super::evolution_bridge::candidate_to_manifest(&candidate);
+        let skill_id = manifest.id.to_string();
+
+        // 6. Persist to disk
+        let skills_dir = self.config.skills_dir.as_ref()
+            .ok_or_else(|| zclaw_types::ZclawError::InvalidInput(
+                "Skills directory not configured".into()
+            ))?;
+        self.skills.create_skill(skills_dir, manifest).await?;
+
+        tracing::info!(
+            "[Kernel] Auto-generated skill '{}' (id={}) registered (disabled)",
+            candidate.name, skill_id
+        );
+
+        Ok(skill_id)
+    }
 }