fix(runtime,hands): 4项根因修复 — URL编码/Browser桩/定时解析/LLM超时

1. researcher.rs: url_encode() chars→bytes，修复中文搜索URL编码
   (U+533B→%533B 改为 UTF-8 %E5%8C%BB)
2. browser.rs: WebDriver不可用时返回明确错误而非静默成功，
   防止LLM误以为操作已完成
3. nl_schedule.rs: 新增相对延迟解析(秒后/分钟后/小时后)，
   避免fallback到LLM幻觉cron
4. 4个LLM driver: 移除http1_only()防reqwest解码错误，
   超时120s→300s适配工具调用链，Anthropic裸Client::new()补全配置

crates/zclaw-hands/src/hands/browser.rs

@@ -246,23 +246,29 @@ impl Hand for BrowserHand {
     }
 
     async fn execute(&self, _context: &HandContext, input: Value) -> Result<HandResult> {
-        // Parse the action
         let action: BrowserAction = match serde_json::from_value(input) {
             Ok(a) => a,
             Err(e) => return Ok(HandResult::error(format!("Invalid action: {}", e))),
         };
 
-        // Browser automation executes on the frontend via BrowserHandCard.
-        // Return the parsed action with a clear message so the LLM can inform
-        // the user and the frontend can pick it up via Tauri events.
         let action_type = action.action_name();
         let summary = action.summary();
 
+        // Check if WebDriver is available
+        if !self.check_webdriver() {
+            return Ok(HandResult::error(format!(
+                "浏览器操作「{}」无法执行：未检测到 WebDriver (ChromeDriver/GeckoDriver)。请先启动 WebDriver 服务。",
+                summary
+            )));
+        }
+
+        // WebDriver is running — delegate to frontend BrowserHandCard.
+        // The frontend manages the Fantoccini session lifecycle.
         Ok(HandResult::success(serde_json::json!({
             "action": action_type,
             "status": "delegated_to_frontend",
-            "message": format!("浏览器操作「{}」已委托给前端执行。请在 HandsPanel 中查看执行结果。", summary),
-            "details": format!("{} — 需要 WebDriver 会话，由前端 BrowserHandCard 管理。", summary),
+            "message": format!("浏览器操作「{}」已发送到前端执行。WebDriver 已就绪。", summary),
+            "details": format!("{} — 由前端 BrowserHandCard 通过 Fantoccini 执行。", summary),
         })))
     }
 
@@ -512,12 +518,16 @@ mod tests {
         assert!(!sequence.stop_on_error);
         assert_eq!(sequence.steps.len(), 1);
 
-        // Execute the navigate step
+        // Execute the navigate step — without WebDriver running, should report error
         let action_json = serde_json::to_value(&sequence.steps[0]).expect("serialize step");
         let result = hand.execute(&ctx, action_json).await.expect("execute");
-        assert!(result.success);
-        assert_eq!(result.output["action"], "navigate");
-        assert_eq!(result.output["status"], "delegated_to_frontend");
+        // In test env no WebDriver is running, so we get an error about missing WebDriver
+        if result.success {
+            assert_eq!(result.output["action"], "navigate");
+            assert_eq!(result.output["status"], "delegated_to_frontend");
+        } else {
+            assert!(result.error.as_deref().unwrap_or("").contains("WebDriver"));
+        }
     }
 
     #[tokio::test]
@@ -533,11 +543,18 @@ mod tests {
 
         assert_eq!(sequence.steps.len(), 4);
 
-        // Verify each step can execute
+        // Verify each step can parse and execute (or report missing WebDriver)
         for (i, step) in sequence.steps.iter().enumerate() {
             let action_json = serde_json::to_value(step).expect("serialize step");
             let result = hand.execute(&ctx, action_json).await.expect("execute step");
-            assert!(result.success, "Step {} failed: {:?}", i, result.error);
+            // Without WebDriver, all steps should report the error cleanly
+            if !result.success {
+                assert!(
+                    result.error.as_deref().unwrap_or("").contains("WebDriver"),
+                    "Step {} unexpected error: {:?}",
+                    i, result.error
+                );
+            }
         }
     }
 

crates/zclaw-hands/src/hands/researcher.rs

@@ -562,12 +562,14 @@ impl Hand for ResearcherHand {
     }
 }
 
-/// URL encoding helper (simple implementation)
+/// URL encoding helper — encodes each UTF-8 byte, not Unicode code points.
 fn url_encode(s: &str) -> String {
-    s.chars()
-        .map(|c| match c {
-            'A'..='Z' | 'a'..='z' | '0'..='9' | '-' | '_' | '.' | '~' => c.to_string(),
-            _ => format!("%{:02X}", c as u32),
+    s.bytes()
+        .map(|b| match b {
+            b'A'..=b'Z' | b'a'..=b'z' | b'0'..=b'9' | b'-' | b'_' | b'.' | b'~' => {
+                (b as char).to_string()
+            }
+            _ => format!("%{:02X}", b),
         })
         .collect()
 }
@@ -706,9 +708,13 @@ mod tests {
 
     #[test]
     fn test_url_encode_chinese() {
+        // "医" = UTF-8 bytes E5 8C BB → must produce %E5%8C%BB, not %533B
+        let encoded = url_encode("医");
+        assert_eq!(encoded, "%E5%8C%BB");
+
+        // Full phrase: "中文" = E4 B8 AD E6 96 87
         let encoded = url_encode("中文搜索");
-        assert!(encoded.contains("%"));
-        // Chinese chars should be percent-encoded
+        assert_eq!(&encoded[0..9], "%E4%B8%AD");
         assert!(!encoded.contains("中文"));
     }