feat(sight): observe-only LLM cache-hit shadow analyzer

src/agentsight/src/bin/cli/trace.rs

@@ -22,6 +22,11 @@ pub struct TraceCommand {
     #[structopt(long)]
     pub enable_filewatch: bool,
 
+    /// Enable the observe-only LLM cache-hit shadow analyzer (measures would-be
+    /// cache hits and key precision on real traffic; never serves)
+    #[structopt(long)]
+    pub enable_cache_analysis: bool,
+
     /// Path to JSON configuration file
     #[structopt(short, long, default_value = "/etc/agentsight/config.json")]
     pub config: String,
@@ -72,6 +77,7 @@ impl TraceCommand {
         let config = AgentsightConfig::new()
             .set_verbose(self.verbose)
             .set_enable_filewatch(self.enable_filewatch)
+            .set_enable_cache_analysis(self.enable_cache_analysis)
             .set_config_path(std::path::PathBuf::from(&self.config));
 
         // Create AgentSight (auto-attaches probes and starts polling)

src/agentsight/src/config.rs

@@ -445,6 +445,9 @@ pub struct AgentsightConfig {
     /// Obtain via `stat -c %i /sys/fs/cgroup/<path>` (v2) or
     /// `stat -c %i /sys/fs/cgroup/memory/<path>` (v1).
     pub cgroup_ids: Vec<u64>,
+    /// Enable the observe-only LLM cache-hit shadow analyzer (measures would-be
+    /// cache hits + key precision on real traffic; never serves).
+    pub enable_cache_analysis: bool,
     /// TCP capture targets for plain HTTP capture (empty = disabled).
     /// Each entry specifies destination IP, port, or both.
     pub tcp_targets: Vec<TcpTarget>,
@@ -525,6 +528,7 @@ impl Default for AgentsightConfig {
             enable_filewatch: false,
             cgroup_filter_enabled: false,
             cgroup_ids: Vec::new(),
+            enable_cache_analysis: false,
             tcp_targets: Vec::new(),
 
             // HTTP/Aggregation defaults
@@ -630,6 +634,12 @@ impl AgentsightConfig {
         self
     }
 
+    /// Set enable_cache_analysis
+    pub fn set_enable_cache_analysis(mut self, enable: bool) -> Self {
+        self.enable_cache_analysis = enable;
+        self
+    }
+
     /// Set connection capacity
     pub fn set_connection_capacity(mut self, capacity: usize) -> Self {
         self.connection_capacity = capacity;

src/agentsight/src/genai/builder.rs

@@ -1186,7 +1186,7 @@ impl GenAIBuilder {
     ///
     /// [Tue 2026-03-31 17:19 GMT+8] 用户实际输入
     /// ```
-    fn strip_user_query_prefix(text: &str) -> String {
+    pub(crate) fn strip_user_query_prefix(text: &str) -> String {
         // 查找最后一个 [timestamp] 模式，取其后的内容
         // 格式: [Day YYYY-MM-DD HH:MM TZ] 或 [Day, DD Mon YYYY HH:MM:SS TZ]
         if let Some(pos) = text.rfind(']') {