CORE: added prom metrics on /metrics

src/agent/cover_art.rs

@@ -128,13 +128,21 @@ pub async fn find_best_cover(folder: &Path, audio_files: &[PathBuf]) -> Option<C
         match tokio::fs::read(img_path).await {
             Ok(data) if !data.is_empty() => {
                 let mime = mime_for_image(img_path);
+                crate::metrics::record_agent_cover_lookup("folder", "ok", data.len());
                 return Some(CoverImage {
                     data,
                     mime_type: mime,
                     source: CoverSource::FolderFile(img_path.clone()),
                 });
             }
-            _ => continue,
+            Ok(_) => {
+                crate::metrics::record_agent_cover_lookup("folder", "empty", 0);
+                continue;
+            }
+            Err(_) => {
+                crate::metrics::record_agent_cover_lookup("folder", "error", 0);
+                continue;
+            }
         }
     }
 
@@ -143,10 +151,12 @@ pub async fn find_best_cover(folder: &Path, audio_files: &[PathBuf]) -> Option<C
         let path = audio_path.to_path_buf();
         let result = tokio::task::spawn_blocking(move || extract_embedded_cover(&path)).await;
         if let Ok(Some(cover)) = result {
+            crate::metrics::record_agent_cover_lookup("embedded", "ok", cover.data.len());
             return Some(cover);
         }
     }
 
+    crate::metrics::record_agent_cover_lookup("none", "not_found", 0);
     None
 }
 

src/agent/cover_variants.rs

@@ -74,6 +74,7 @@ fn generate_missing_variants_sync(
 
     let mut created = 0usize;
     for variant in variants {
+        let start = std::time::Instant::now();
         let path = variant_path(original_path, *variant);
         if path.exists() {
             continue;
@@ -88,12 +89,27 @@ fn generate_missing_variants_sync(
             .to_rgb8();
         let mut output = Vec::new();
         let mut encoder = JpegEncoder::new_with_quality(&mut output, variant.quality);
-        encoder.encode(
+        let result = encoder.encode(
             &resized,
             resized.width(),
             resized.height(),
             image::ExtendedColorType::Rgb8,
-        )?;
+        );
+        match result {
+            Ok(()) => crate::metrics::record_agent_cover_variant(
+                variant.name,
+                "ok",
+                start.elapsed(),
+            ),
+            Err(err) => {
+                crate::metrics::record_agent_cover_variant(
+                    variant.name,
+                    "error",
+                    start.elapsed(),
+                );
+                return Err(err.into());
+            }
+        }
         std::fs::write(path, output)?;
         created += 1;
     }

src/agent/normalize.rs

@@ -330,6 +330,7 @@ pub async fn normalize_batch(
     // If over 80% of context limit and more than 1 file, split
     let limit_80 = context_limit * 80 / 100;
     if estimated > limit_80 && files.len() > 1 {
+        crate::metrics::record_agent_llm_split("estimated_context");
         tracing::info!(
             estimated_tokens = estimated,
             context_limit,
@@ -419,6 +420,7 @@ pub async fn normalize_batch(
                 || err_str.contains("length")
                 || err_str.contains("token");
             if is_context_error {
+                crate::metrics::record_agent_llm_split("context_error");
                 tracing::warn!(
                     file_count = files.len(),
                     "LLM error suggests context overflow, splitting batch: {e}"
@@ -466,14 +468,40 @@ pub async fn normalize_batch(
             }
             return Err(e);
         }
-        Err(e) => return Err(e),
+        Err(e) => {
+            crate::metrics::record_agent_llm(
+                llm_model,
+                "error",
+                std::time::Duration::from_millis(duration_ms),
+                0,
+                0,
+                files.len(),
+                Some(estimated),
+            );
+            return Err(e);
+        }
     };
 
     let prompt_tokens = usage.prompt_tokens.unwrap_or(0) as u64;
     let completion_tokens = usage.completion_tokens.unwrap_or(0) as u64;
+    crate::metrics::record_agent_llm(
+        &resp_model,
+        "ok",
+        std::time::Duration::from_millis(duration_ms),
+        prompt_tokens,
+        completion_tokens,
+        files.len(),
+        Some(estimated),
+    );
 
     // Parse batch response
-    let results = parse_batch_response(&response_text, &files)?;
+    let results = match parse_batch_response(&response_text, &files) {
+        Ok(results) => results,
+        Err(error) => {
+            crate::metrics::record_agent_llm_parse_failure(&resp_model);
+            return Err(error);
+        }
+    };
 
     Ok(BatchNormalizeResult {
         results,