fix(cache): strengthen prompt cache breakpoint injection

Honor both the optimizer master switch and the cache-injection sub-switch before mutating native optimizer requests.

Use the available four-breakpoint budget across tools, system content, the latest cacheable message, and an older user anchor for long tool-heavy conversations.

Preserve caller-owned breakpoint limits, avoid thinking blocks as cache targets, normalize configured TTLs, and add regression coverage for disabled optimization and long histories.

src-tauri/src/proxy/cache_injector.rs

@@ -7,7 +7,7 @@ use serde_json::{json, Value};
 
 /// 在请求体关键位置注入 cache_control 断点
 pub fn inject(body: &mut Value, config: &OptimizerConfig) {
-    if !config.cache_injection {
+    if !config.enabled || !config.cache_injection {
         return;
     }
 
@@ -89,24 +89,29 @@ pub fn inject(body: &mut Value, config: &OptimizerConfig) {
     // user/tool_result 结束；只标 assistant 会让最新稳定前缀无法命中缓存。
     if budget > 0 {
         if let Some(messages) = body.get_mut("messages").and_then(|m| m.as_array_mut()) {
-            'messages: for message in messages.iter_mut().rev() {
+            for message in messages.iter_mut().rev() {
-                if let Some(content) = message.get_mut("content").and_then(|c| c.as_array_mut()) {
+                if inject_message_breakpoint(message, &config.cache_ttl) {
-                    if let Some(block) = content.iter_mut().rev().find(|block| {
+                    budget -= 1;
-                        !matches!(
+                    injected.push("msgs-latest");
-                            block.get("type").and_then(Value::as_str),
+                    break;
-                            Some("thinking" | "redacted_thinking")
+                }
-                        )
+            }
-                    }) {
+
-                        if block.get("cache_control").is_none() {
+            // (d) A second, older user anchor helps long tool-result turns where
-                            if let Some(object) = block.as_object_mut() {
+            // the stable prefix falls outside Anthropic's 20-block lookback from
-                                object.insert(
+            // the newest breakpoint. Keep this best-effort and inside the 4-BP cap.
-                                    "cache_control".to_string(),
+            if budget > 0 && messages.len() >= 4 {
-                                    make_cache_control(&config.cache_ttl),
+                let mut user_count = 0;
-                                );
+                for message in messages.iter_mut().rev() {
-                                injected.push("msgs");
+                    if message.get("role").and_then(Value::as_str) != Some("user") {
-                            }
+                        continue;
+                    }
+                    user_count += 1;
+                    if user_count == 2 {
+                        if inject_message_breakpoint(message, &config.cache_ttl) {
+                            injected.push("msgs-prior-user");
                         }
-                        break 'messages;
+                        break;
                     }
                 }
             }
@@ -121,6 +126,28 @@ pub fn inject(body: &mut Value, config: &OptimizerConfig) {
     );
 }
 
+fn inject_message_breakpoint(message: &mut Value, ttl: &str) -> bool {
+    let Some(content) = message.get_mut("content").and_then(Value::as_array_mut) else {
+        return false;
+    };
+    let Some(block) = content.iter_mut().rev().find(|block| {
+        !matches!(
+            block.get("type").and_then(Value::as_str),
+            Some("thinking" | "redacted_thinking")
+        )
+    }) else {
+        return false;
+    };
+    if block.get("cache_control").is_some() {
+        return false;
+    }
+    let Some(object) = block.as_object_mut() else {
+        return false;
+    };
+    object.insert("cache_control".to_string(), make_cache_control(ttl));
+    true
+}
+
 fn make_cache_control(ttl: &str) -> Value {
     if ttl == "5m" {
         json!({"type": "ephemeral"})
@@ -244,6 +271,30 @@ mod tests {
             .is_some());
     }
 
+    #[test]
+    fn test_long_history_uses_fourth_prior_user_breakpoint() {
+        let mut body = json!({
+            "model":"test",
+            "tools":[{"name":"tool1"}],
+            "system":[{"type":"text","text":"sys"}],
+            "messages":[
+                {"role":"user","content":[{"type":"text","text":"first"}]},
+                {"role":"assistant","content":[{"type":"text","text":"answer"}]},
+                {"role":"user","content":[{"type":"tool_result","tool_use_id":"c1","content":"result"}]},
+                {"role":"assistant","content":[{"type":"text","text":"latest"}]}
+            ]
+        });
+
+        inject(&mut body, &default_config());
+        assert_eq!(count_existing(&body), 4);
+        assert!(body["messages"][0]["content"][0]
+            .get("cache_control")
+            .is_some());
+        assert!(body["messages"][3]["content"][0]
+            .get("cache_control")
+            .is_some());
+    }
+
     #[test]
     fn test_existing_four_breakpoints_only_upgrades_ttl() {
         let mut body = json!({
@@ -379,6 +430,24 @@ mod tests {
         assert_eq!(body, original);
     }
 
+    #[test]
+    fn test_optimizer_disabled_no_change() {
+        let config = OptimizerConfig {
+            enabled: false,
+            cache_injection: true,
+            ..default_config()
+        };
+        let mut body = json!({
+            "model":"test",
+            "tools":[{"name":"tool1"}],
+            "messages":[{"role":"user","content":[{"type":"text","text":"hi"}]}]
+        });
+        let original = body.clone();
+
+        inject(&mut body, &config);
+        assert_eq!(body, original);
+    }
+
     #[test]
     fn test_skip_thinking_blocks_in_assistant() {
         let mut body = json!({