diff --git a/fastdeploy/golang_router/.claude/skills/troubleshoot/SKILL.md b/fastdeploy/golang_router/.claude/skills/troubleshoot/SKILL.md
index 43ee91a46b1..7f7a5793e91 100644
--- a/fastdeploy/golang_router/.claude/skills/troubleshoot/SKILL.md
+++ b/fastdeploy/golang_router/.claude/skills/troubleshoot/SKILL.md
@@ -24,10 +24,9 @@ IMPORTANT: 执行前务必先读取 references/log_patterns.md 了解日志格
 运行脚本前，Claude 必须按以下顺序向用户确认参数：
 
 ### 1. 日志文件路径
-使用 AskUserQuestion 工具向用户询问日志文件路径。提供常见的默认选项，同时允许用户直接输入自定义路径（支持绝对路径和相对路径）：
+使用 AskUserQuestion 工具向用户询问日志文件路径。提供两个常用快捷选项（客户端会自动提供 Other 自定义输入）：
 - 选项 1: `logs/router.log`（默认）
 - 选项 2: `fd-router.log`（golang_router 根目录）
-- 选项 3: 用户通过 Other 输入自定义路径
 
 **重要规则**：
 - 如果用户已经在消息中明确指定了日志路径，直接使用该路径，跳过询问步骤
@@ -37,11 +36,10 @@ IMPORTANT: 执行前务必先读取 references/log_patterns.md 了解日志格
 如果用户直接确认或未指定路径，使用脚本的自动发现逻辑。
 
 ### 2. 分析范围
-向用户询问分析范围：
-> "请选择分析范围：
-> 1. **全量分析**（默认）— 分析整个日志文件
-> 2. **尾部分析** — 只分析最近数据（可指定行数或时间如 `--tail 5000` 或 `--tail 30m`）
-> 3. **指定时间段** — 分析特定时间范围内的日志"
+必须使用 **AskUserQuestion 的离散选项**（不要只发纯文本编号）：
+- 选项 1: `全量分析（默认）` — 分析整个日志文件
+- 选项 2: `尾部分析` — 只分析最近数据（可指定行数或时间如 `--tail 5000` 或 `--tail 30m`）
+- 选项 3: `指定时间段` — 分析特定时间范围内的日志
 
 如果用户未选择，默认使用全量分析。
 
@@ -54,11 +52,10 @@ IMPORTANT: 执行前务必先读取 references/log_patterns.md 了解日志格
 `--start/--end` 与 `--tail` 互斥。
 
 ### 3. 分析模式
-向用户询问分析模式：
-> "请选择分析模式：
-> 1. **完整分析**（默认）— 运行所有维度（errors + latency + health + cache + load）
-> 2. **单维度/多维度分析** — 选择特定维度（errors / latency / health / cache / load），可选多个
-> 3. **请求追踪** — 追踪特定请求 ID（需提供 ID）"
+必须使用 **AskUserQuestion 的离散选项**（不要只发纯文本编号）：
+- 选项 1: `完整分析（默认）` — 运行所有维度（errors + latency + health + cache + load）
+- 选项 2: `单维度/多维度分析` — 选择特定维度（errors / latency / health / cache / load），可选多个
+- 选项 3: `请求追踪` — 追踪特定请求 ID（需提供 ID）
 
 如果用户未选择，默认使用完整分析。
 
diff --git a/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/analyzers/load.py b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/analyzers/load.py
index 9be82357494..c38b0b80953 100644
--- a/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/analyzers/load.py
+++ b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/analyzers/load.py
@@ -13,9 +13,9 @@
 
 sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
 
-from chart import render_bar, render_sparkline, render_table
 from log_parser import extract_ts, match_select_release, parse_stats_line
 from stats import compute_statistics, time_bucket
+from analyzers.load_report import format_load_report
 
 # ════════════════════════════════════════════════════════════════
 # Counter 异常检测正则
@@ -28,14 +28,21 @@
 TOKEN_PRESERVED_RE = re.compile(rf"token counter preserved.*?{URL_RE}")
 
 # Token 事件
-SELECT_TOKENS_RE = re.compile(rf"select worker \(prefill\):\s*{URL_RE},\s*tokens:\s*(\d+)")
-RELEASE_TOKENS_RE = re.compile(rf"release prefill tokens:\s*{URL_RE},\s*tokens:\s*(\d+)")
+SELECT_TOKENS_RE = re.compile(rf"select worker \((\w+)\):\s*{URL_RE},\s*tokens:\s*(\d+)")
+RELEASE_TOKENS_RE = re.compile(rf"release (?:([a-zA-Z_]+)\s+)?tokens:\s*{URL_RE},\s*tokens:\s*(\d+)")
 
 
 def _strip_scheme(url):
     return re.sub(r"^https?://", "", url)
 
 
+def _normalize_worker_type(worker_type):
+    t = (worker_type or "unknown").lower()
+    if t in ("prefill", "decode", "mixed"):
+        return t
+    return "unknown"
+
+
 def parse_counter_anomaly(line):
     """解析 H5 counter 异常行。"""
     ts = extract_ts(line)
@@ -73,7 +80,7 @@ def analyze_load(log_file, tail=None):
         r"counter preserved|cleanup unhealthy|removed counters|counter already|double-release|preserved counters",
         tail,
     )
-    h11_lines = _grep_lines(log_file, r"release prefill tokens", tail)
+    h11_lines = _grep_lines(log_file, r"release (?:[a-zA-Z_]+\s+)?tokens", tail)
 
     # 解析 stats 行
     stats_records = [r for line in h7_lines for r in [parse_stats_line(line)] if r]
@@ -161,12 +168,12 @@ def _analyze_tokens(h3_lines, h11_lines):
     for line in h3_lines:
         m = SELECT_TOKENS_RE.search(line)
         if m:
-            token_alloc[m.group(1)].append(int(m.group(2)))
+            token_alloc[m.group(2)].append(int(m.group(3)))
 
     for line in h11_lines:
         m = RELEASE_TOKENS_RE.search(line)
         if m:
-            token_release[m.group(1)].append(int(m.group(2)))
+            token_release[m.group(2)].append(int(m.group(3)))
 
     result = []
     all_workers = set(token_alloc.keys()) | set(token_release.keys())
@@ -285,135 +292,6 @@ def _diagnose(load_stats, worker_load, anomaly_summary, sr_result, token_stats,
 # ════════════════════════════════════════════════════════════════
 
 
-def format_load_report(result):
-    """将分析结果格式化为终端报告。"""
-    sections = ["## 负载与计数器分析", ""]
-    sections.append(f'  {result["summary"]}')
-    sections.append("")
-
-    if result["diagnoses"]:
-        sections.append("### 诊断")
-        sections.append("")
-        for d in result["diagnoses"]:
-            sections.append(f'  [{d["severity"]}] [{d["source_layer"]}] {d["message"]}')
-        sections.append("")
-
-    # 负载概览
-    ls = result.get("load_stats", {})
-    if ls:
-        sections.append("### 负载概览 (total_running)")
-        sections.append("")
-        sections.append(
-            f'  mean={ls.get("mean",0)}  p50={ls.get("p50",0)}  p90={ls.get("p90",0)}  '
-            f'p99={ls.get("p99",0)}  max={ls.get("max",0)}  stddev={ls.get("stddev",0)}'
-        )
-        sections.append("")
-
-    # Per-Worker 负载
-    if result["worker_load"]:
-        sections.append("### Per-Worker 负载")
-        sections.append("")
-        bar_data = [
-            {"label": w["worker"][:25], "value": min(100, w["avg_running"] * 5), "count": w["avg_running"]}
-            for w in result["worker_load"]
-        ]
-        sections.append(render_bar(bar_data, show_count=True))
-        sections.append("")
-
-    # 负载趋势
-    if result["load_trend"] and len(result["load_trend"]) > 1:
-        sections.append("### 负载趋势")
-        sections.append("")
-        sections.append(
-            render_sparkline(
-                result["load_trend"], value_field="total_running_mean", title="Total Running", y_label="req"
-            )
-        )
-        sections.append("")
-
-    # Counter 异常
-    if result["counter_anomalies"]:
-        sections.append("### 计数器异常")
-        sections.append("")
-        for a in result["counter_anomalies"]:
-            workers_str = ", ".join(f'{_strip_scheme(w)}({c})' for w, c in a["workers"].items())
-            sections.append(f'  {a["type"]}: {a["total"]} 次 [{workers_str}]')
-        sections.append("")
-
-    id_cov = result.get("select_release", {}).get("id_coverage", {})
-    if id_cov:
-        sections.append("### 请求标识覆盖（基于 select 近似请求数）")
-        sections.append("")
-        sections.append(
-            "  total={total} | with_request_id={with_rid} | without_request_id={without_rid} | "
-            "with_alt_id={with_alt} | without_any_id={without_any}".format(
-                total=id_cov.get("total_requests_estimated", 0),
-                with_rid=id_cov.get("with_request_id", 0),
-                without_rid=id_cov.get("without_request_id", 0),
-                with_alt=id_cov.get("with_alt_id", 0),
-                without_any=id_cov.get("without_any_id", 0),
-            )
-        )
-        if id_cov.get("without_any_id", 0) > 0:
-            sections.append("  ℹ 无 request/session/trace/req_id 时，不做退化匹配，仅统计为 untracked。")
-        sections.append("")
-
-    # Select/Release 匹配
-    sr = result.get("select_release", {})
-    if sr.get("per_worker"):
-        sections.append("### Select/Release 匹配")
-        sections.append("")
-        id_cov = sr.get("id_coverage", {})
-        no_correlatable_id = (id_cov.get("with_request_id", 0) + id_cov.get("with_alt_id", 0)) == 0
-        table_data = []
-        for w_url, pw in sorted(sr["per_worker"].items()):
-            delta_display = "N/A" if no_correlatable_id else str(pw["delta"])
-            table_data.append(
-                {
-                    "Worker": _strip_scheme(w_url),
-                    "Select": str(pw["selects"]),
-                    "Release": str(pw["releases"]),
-                    "Delta": delta_display,
-                }
-            )
-        sections.append(
-            render_table(
-                table_data,
-                columns=["Worker", "Select", "Release", "Delta"],
-                right_align={"Select", "Release", "Delta"},
-            )
-        )
-        sections.append("")
-        if no_correlatable_id:
-            sections.append("  ℹ 当前样本无可关联 ID，Delta 不用于请求泄漏结论。")
-            sections.append("")
-
-    if sr.get("unmatched_selects"):
-        sections.append(f'  ⚠ {len(sr["unmatched_selects"])} 个未匹配 select（疑似请求卡住）')
-        for u in sr["unmatched_selects"][:5]:
-            sections.append(f'    [{u.get("select_ts","")}] {_strip_scheme(u["worker"])} ({u["type"]})')
-        sections.append("")
-
-    if sr.get("untracked_selects"):
-        sections.append(f'  ℹ {len(sr["untracked_selects"])} 个 select 缺少可关联 ID，未参与卡住判定')
-        for u in sr["untracked_selects"][:5]:
-            sections.append(f'    [{u.get("select_ts","")}] {_strip_scheme(u["worker"])} ({u["type"]})')
-        sections.append("")
-
-    # Token 统计
-    if result.get("token_stats"):
-        sections.append("### Token 计数器")
-        sections.append("")
-        sections.append(
-            render_table(
-                result["token_stats"],
-                columns=["worker", "alloc_count", "alloc_avg", "release_count"],
-                right_align={"alloc_count", "alloc_avg", "release_count"},
-            )
-        )
-        sections.append("")
-
-    return "\n".join(sections)
 
 
 # ════════════════════════════════════════════════════════════════
diff --git a/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/analyzers/load_report.py b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/analyzers/load_report.py
new file mode 100644
index 00000000000..86ba1f0d94f
--- /dev/null
+++ b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/analyzers/load_report.py
@@ -0,0 +1,242 @@
+#!/usr/bin/env python3
+"""Load report formatter."""
+
+from chart import render_bar, render_sparkline, render_table
+
+
+def _strip_scheme(url):
+    import re
+    return re.sub(r"^https?://", "", url)
+
+
+def format_load_report(result):
+    """将分析结果格式化为终端报告。
+
+    Returns:
+        tuple: (summary_text, detail_text)
+    """
+    sections = ["## 负载与计数器分析", ""]
+    sections.append(f'  {result["summary"]}')
+    sections.append("")
+    detail_sections = ["# 负载与计数器详情", ""]
+    detail_sections.append(f'总结: {result["summary"]}')
+    detail_sections.append("")
+
+    if result["diagnoses"]:
+        sections.append("### 诊断")
+        sections.append("")
+        for d in result["diagnoses"]:
+            sections.append(f'  [{d["severity"]}] [{d["source_layer"]}] {d["message"]}')
+        sections.append("")
+        detail_sections.append("## 诊断")
+        detail_sections.append("")
+        for d in result["diagnoses"]:
+            detail_sections.append(f'[{d["severity"]}] [{d["source_layer"]}] {d["message"]}')
+        detail_sections.append("")
+
+    # 负载概览
+    ls = result.get("load_stats", {})
+    if ls:
+        sections.append("### 负载概览 (total_running)")
+        sections.append("")
+        sections.append(
+            f'  mean={ls.get("mean",0)}  p50={ls.get("p50",0)}  p90={ls.get("p90",0)}  '
+            f'p99={ls.get("p99",0)}  max={ls.get("max",0)}  stddev={ls.get("stddev",0)}'
+        )
+        sections.append("")
+
+    # Per-Worker 负载
+    if result["worker_load"]:
+        sections.append("### Per-Worker 负载")
+        sections.append("")
+        bar_data = [
+            {"label": w["worker"][:25], "value": min(100, w["avg_running"] * 5), "count": w["avg_running"]}
+            for w in result["worker_load"]
+        ]
+        sections.append(render_bar(bar_data, show_count=True))
+        sections.append("")
+
+    # 负载趋势
+    if result["load_trend"] and len(result["load_trend"]) > 1:
+        sections.append("### 负载趋势")
+        sections.append("")
+        sections.append(
+            render_sparkline(
+                result["load_trend"], value_field="total_running_mean", title="Total Running", y_label="req"
+            )
+        )
+        sections.append("")
+
+    # Counter 异常
+    if result["counter_anomalies"]:
+        sections.append("### 计数器异常")
+        sections.append("")
+        for a in result["counter_anomalies"]:
+            workers_str = ", ".join(f'{_strip_scheme(w)}({c})' for w, c in a["workers"].items())
+            sections.append(f'  {a["type"]}: {a["total"]} 次 [{workers_str}]')
+        sections.append("")
+        detail_sections.append("## 计数器异常")
+        detail_sections.append("")
+        for a in result["counter_anomalies"]:
+            workers_str = ", ".join(f'{_strip_scheme(w)}({c})' for w, c in a["workers"].items())
+            detail_sections.append(f'- {a["type"]}: {a["total"]} 次 [{workers_str}]')
+        detail_sections.append("")
+
+    # 按 prefill / decode / mixed 分类统计
+    type_summary = result.get("select_release", {}).get("type_summary", {})
+    if type_summary:
+        sections.append("### 按类型统计（prefill / decode / mixed）")
+        sections.append("")
+        type_rows = []
+        for t in ("prefill", "decode", "mixed", "unknown"):
+            s = type_summary.get(t)
+            if not s:
+                continue
+            token_display = "-"
+            if t == "prefill":
+                token_display = f'{s.get("token_selects",0)}/{s.get("token_releases",0)}'
+            elif t == "mixed" and (s.get("token_selects", 0) > 0 or s.get("token_releases", 0) > 0):
+                token_display = f'{s.get("token_selects",0)}/{s.get("token_releases",0)}'
+            type_rows.append(
+                {
+                    "type": t,
+                    "counter(S/R)": f'{s.get("counter_selects",0)}/{s.get("counter_releases",0)}',
+                    "token(S/R)": token_display,
+                }
+            )
+        if type_rows:
+            sections.append(render_table(type_rows, columns=["type", "counter(S/R)", "token(S/R)"]))
+            sections.append("")
+            sections.append("  说明: prefill/mixed 的 token-select 同时表示 request counter + token counter 增加；decode 仅 request counter。")
+            sections.append("")
+            detail_sections.append("## 按类型统计")
+            detail_sections.append("")
+            detail_sections.append(render_table(type_rows, columns=["type", "counter(S/R)", "token(S/R)"]))
+            detail_sections.append("")
+
+    id_cov = result.get("select_release", {}).get("id_coverage", {})
+    if id_cov:
+        sections.append("### 请求标识覆盖（基于 select 近似请求数）")
+        sections.append("")
+        sections.append(
+            "  total={total} | with_request_id={with_rid} | without_request_id={without_rid} | "
+            "with_alt_id={with_alt} | without_any_id={without_any}".format(
+                total=id_cov.get("total_requests_estimated", 0),
+                with_rid=id_cov.get("with_request_id", 0),
+                without_rid=id_cov.get("without_request_id", 0),
+                with_alt=id_cov.get("with_alt_id", 0),
+                without_any=id_cov.get("without_any_id", 0),
+            )
+        )
+        if id_cov.get("without_any_id", 0) > 0:
+            sections.append("  ℹ 无 request/session/trace/req_id 时，不做退化匹配，仅统计为 untracked。")
+        sections.append("  字段说明: total=select 事件总数估算；with_request_id=含 request_id；without_request_id=不含 request_id；with_alt_id=含 req_id/trace_id/session_id；without_any_id=四类 ID 都缺失。")
+        sections.append("")
+        detail_sections.append("## 请求标识覆盖字段说明")
+        detail_sections.append("")
+        detail_sections.append(
+            "- total: select 事件总数（近似请求数）\n"
+            "- with_request_id: 携带 request_id 的 select 数\n"
+            "- without_request_id: 未携带 request_id 的 select 数\n"
+            "- with_alt_id: 无 request_id 但携带 req_id/trace_id/session_id 的 select 数\n"
+            "- without_any_id: 四类 ID 都没有，无法做请求级关联"
+        )
+        detail_sections.append("")
+
+    # Select/Release 匹配
+    sr = result.get("select_release", {})
+    if sr.get("per_worker"):
+        sections.append("### Select/Release 匹配")
+        sections.append("")
+        id_cov = sr.get("id_coverage", {})
+        no_correlatable_id = (id_cov.get("with_request_id", 0) + id_cov.get("with_alt_id", 0)) == 0
+        table_data = []
+        for w_url, pw in sorted(sr["per_worker"].items()):
+            delta_display = "N/A" if no_correlatable_id else str(pw["delta"])
+            table_data.append(
+                {
+                    "Worker": _strip_scheme(w_url),
+                    "ReqSelect": str(pw["selects"]),
+                    "ReqRelease": str(pw["releases"]),
+                    "ReqDelta": delta_display,
+                    "TokenSelect": str(pw.get("token_selects", 0)),
+                    "TokenRelease": str(pw.get("token_releases", 0)),
+                }
+            )
+        sections.append(
+            render_table(
+                table_data,
+                columns=["Worker", "ReqSelect", "ReqRelease", "ReqDelta", "TokenSelect", "TokenRelease"],
+                right_align={"ReqSelect", "ReqRelease", "ReqDelta", "TokenSelect", "TokenRelease"},
+            )
+        )
+        sections.append("")
+        if no_correlatable_id:
+            sections.append("  ℹ 当前样本无可关联 ID，Delta 不用于请求泄漏结论。")
+            sections.append("")
+        sections.append("  说明: TokenSelect 按 worker type 统计（prefill + mixed 的 select 都计入），不依赖日志里是否出现 tokens 字段。")
+        sections.append("")
+        detail_sections.append("## Select/Release Per-Worker")
+        detail_sections.append("")
+        detail_sections.append(
+            render_table(
+                table_data,
+                columns=["Worker", "ReqSelect", "ReqRelease", "ReqDelta", "TokenSelect", "TokenRelease"],
+                right_align={"ReqSelect", "ReqRelease", "ReqDelta", "TokenSelect", "TokenRelease"},
+            )
+        )
+        detail_sections.append("")
+
+    if sr.get("unmatched_selects"):
+        sections.append(f'  ⚠ {len(sr["unmatched_selects"])} 个未匹配 select（疑似请求卡住）')
+        sections.append("  解释: 出现 request select，但在 request release 口径下找不到匹配。可能是请求卡住、日志缺失、或窗口外释放。")
+        for u in sr["unmatched_selects"][:3]:
+            sections.append(f'    [{u.get("select_ts","")}] {_strip_scheme(u["worker"])} ({u["type"]})')
+        sections.append("  > 完整列表见: [details/load_select_release.md](details/load_select_release.md)")
+        sections.append("")
+        detail_sections.append("## 未匹配 select（完整）")
+        detail_sections.append("")
+        for u in sr["unmatched_selects"]:
+            detail_sections.append(
+                f'- [{u.get("select_ts","")}] worker={_strip_scheme(u["worker"])} type={u["type"]} note={u.get("note","")}'
+            )
+        detail_sections.append("")
+
+    if sr.get("untracked_selects"):
+        sections.append(f'  ℹ {len(sr["untracked_selects"])} 个 select 缺少可关联 ID，未参与卡住判定')
+        for u in sr["untracked_selects"][:3]:
+            sections.append(f'    [{u.get("select_ts","")}] {_strip_scheme(u["worker"])} ({u["type"]})')
+        sections.append("  > 完整列表见: [details/load_select_release.md](details/load_select_release.md)")
+        sections.append("")
+        detail_sections.append("## Untracked selects（缺少可关联 ID）")
+        detail_sections.append("")
+        for u in sr["untracked_selects"]:
+            detail_sections.append(
+                f'- [{u.get("select_ts","")}] worker={_strip_scheme(u["worker"])} type={u["type"]} note={u.get("note","")}'
+            )
+        detail_sections.append("")
+
+    if sr.get("failed_selects"):
+        sections.append(f'  ⚠ Failed to select: {len(sr["failed_selects"])} 次')
+        sections.append("  解释: 路由在该时刻未能选出可用 worker，通常意味着可用池不足或健康状态异常。")
+        sections.append("")
+        detail_sections.append("## Failed to select")
+        detail_sections.append("")
+        for f in sr["failed_selects"]:
+            detail_sections.append(f'- [{f.get("ts","")}] line={f.get("line","")}')
+        detail_sections.append("")
+
+    # Token 统计
+    if result.get("token_stats"):
+        sections.append("### Token 计数器")
+        sections.append("")
+        sections.append(
+            render_table(
+                result["token_stats"],
+                columns=["worker", "alloc_count", "alloc_avg", "release_count"],
+                right_align={"alloc_count", "alloc_avg", "release_count"},
+            )
+        )
+        sections.append("")
+
+    return "\n".join(sections), "\n".join(detail_sections)
diff --git a/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/log_parser.py b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/log_parser.py
index 44f5cdebd94..200f976f2ff 100644
--- a/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/log_parser.py
+++ b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/log_parser.py
@@ -466,8 +466,8 @@ def parse_error_line(line):
 SELECT_RE = re.compile(rf"select worker\s*(?:\((\w+)\))?:\s*({URL_RE})")
 RELEASE_RE = re.compile(rf"release worker\s*(?:\((\w+)\))?:\s*({URL_RE})")
 FAILED_SELECT_RE = re.compile(r"Failed to select")
-SELECT_TOKENS_RE = re.compile(rf"select worker \(prefill\):\s*({URL_RE}),\s*tokens:\s*(\d+)")
-RELEASE_TOKENS_RE = re.compile(rf"release prefill tokens:\s*({URL_RE}),\s*tokens:\s*(\d+)")
+SELECT_TOKENS_RE = re.compile(rf"select worker \((\w+)\):\s*({URL_RE}),\s*tokens:\s*(\d+)")
+RELEASE_TOKENS_RE = re.compile(rf"release (?:([a-zA-Z_]+)\s+)?tokens:\s*({URL_RE}),\s*tokens:\s*(\d+)")
 
 
 def _parse_ts_safe(ts):
@@ -493,6 +493,14 @@ def _select_match_key(tags):
     return (None, None)
 
 
+def _normalize_worker_type(worker_type):
+    """归一化 worker type。"""
+    t = (worker_type or "unknown").lower()
+    if t in ("prefill", "decode", "mixed"):
+        return t
+    return "unknown"
+
+
 def match_select_release(lines, fallback_window_s=120):
     """匹配 select/release worker 事件对。
 
@@ -516,10 +524,10 @@ def match_select_release(lines, fallback_window_s=120):
             selects.append(
                 {
                     "ts": ts,
-                    "worker": tm.group(1),
-                    "type": "prefill",
+                    "worker": tm.group(2),
+                    "type": _normalize_worker_type(tm.group(1)),
                     "tags": tags,
-                    "tokens": int(tm.group(2)),
+                    "tokens": int(tm.group(3)),
                     "line": line_no,
                 }
             )
@@ -528,13 +536,14 @@ def match_select_release(lines, fallback_window_s=120):
         # Token-bearing release
         trm = RELEASE_TOKENS_RE.search(line)
         if trm:
+            token_type = trm.group(1) or "prefill"
             releases.append(
                 {
                     "ts": ts,
-                    "worker": trm.group(1),
-                    "type": "prefill_tokens",
+                    "worker": trm.group(2),
+                    "type": f'{_normalize_worker_type(token_type)}_tokens',
                     "tags": tags,
-                    "tokens": int(trm.group(2)),
+                    "tokens": int(trm.group(3)),
                     "line": line_no,
                 }
             )
@@ -546,7 +555,7 @@ def match_select_release(lines, fallback_window_s=120):
                 {
                     "ts": ts,
                     "worker": sm.group(2),
-                    "type": sm.group(1) or "unknown",
+                    "type": _normalize_worker_type(sm.group(1)),
                     "tags": tags,
                     "tokens": None,
                     "line": line_no,
@@ -560,7 +569,7 @@ def match_select_release(lines, fallback_window_s=120):
                 {
                     "ts": ts,
                     "worker": rm.group(2),
-                    "type": rm.group(1) or "unknown",
+                    "type": _normalize_worker_type(rm.group(1)),
                     "tags": tags,
                     "tokens": None,
                     "line": line_no,
@@ -576,8 +585,11 @@ def match_select_release(lines, fallback_window_s=120):
     unmatched_selects = []
     release_used = set()
 
+    # 请求生命周期匹配只使用 request counter release（排除 token release）
+    counter_release_indexes = [i for i, r in enumerate(releases) if not str(r.get("type", "")).endswith("_tokens")]
     release_by_key = defaultdict(list)
-    for i, r in enumerate(releases):
+    for i in counter_release_indexes:
+        r = releases[i]
         _, key = _select_match_key(r.get("tags", {}))
         if key:
             release_by_key[key].append(i)
@@ -639,7 +651,8 @@ def match_select_release(lines, fallback_window_s=120):
         sdt = _parse_ts_safe(s["ts"])
         best_idx = None
         best_delta = None
-        for ri, r in enumerate(releases):
+        for ri in counter_release_indexes:
+            r = releases[ri]
             if ri in release_used:
                 continue
             if r.get("worker") != s.get("worker"):
@@ -679,12 +692,19 @@ def match_select_release(lines, fallback_window_s=120):
                 }
             )
 
-    # Per-worker summary
-    per_worker = defaultdict(lambda: {"selects": 0, "releases": 0})
+    # Per-worker summary（按 worker type 统计，不依赖日志中的 tokens 字段）
+    # 规则：prefill/mixed 的 select 均计入 token_selects。
+    per_worker = defaultdict(lambda: {"selects": 0, "releases": 0, "token_selects": 0, "token_releases": 0})
     for s in selects:
+        s_type = _normalize_worker_type(s.get("type"))
         per_worker[s["worker"]]["selects"] += 1
+        if s_type in ("prefill", "mixed"):
+            per_worker[s["worker"]]["token_selects"] += 1
     for r in releases:
-        per_worker[r["worker"]]["releases"] += 1
+        if str(r.get("type", "")).endswith("_tokens"):
+            per_worker[r["worker"]]["token_releases"] += 1
+        else:
+            per_worker[r["worker"]]["releases"] += 1
 
     pw_result = {}
     for w, counts in per_worker.items():
@@ -692,7 +712,30 @@ def match_select_release(lines, fallback_window_s=120):
             "selects": counts["selects"],
             "releases": counts["releases"],
             "delta": counts["selects"] - counts["releases"],
+            "token_selects": counts["token_selects"],
+            "token_releases": counts["token_releases"],
+        }
+
+    # 按 worker type 分类统计（prefill/decode/mixed）
+    type_summary = defaultdict(
+        lambda: {
+            "counter_selects": 0,
+            "counter_releases": 0,
+            "token_selects": 0,
+            "token_releases": 0,
         }
+    )
+    for s in selects:
+        s_type = _normalize_worker_type(s.get("type"))
+        type_summary[s_type]["counter_selects"] += 1
+        if s_type in ("prefill", "mixed"):
+            type_summary[s_type]["token_selects"] += 1
+    for r in releases:
+        r_type = _normalize_worker_type(str(r.get("type", "")).replace("_tokens", ""))
+        if str(r.get("type", "")).endswith("_tokens"):
+            type_summary[r_type]["token_releases"] += 1
+        else:
+            type_summary[r_type]["counter_releases"] += 1
 
     return {
         "matched": matched,
@@ -707,6 +750,7 @@ def match_select_release(lines, fallback_window_s=120):
             "with_alt_id": with_alt_id,
             "without_any_id": without_any_id,
         },
+        "type_summary": dict(type_summary),
     }
 
 
diff --git a/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/troubleshoot.py b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/troubleshoot.py
index 30b9df0f443..a818d31150f 100644
--- a/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/troubleshoot.py
+++ b/fastdeploy/golang_router/.claude/skills/troubleshoot/scripts/troubleshoot.py
@@ -144,10 +144,11 @@ def format_full_report(results, status, status_reason):
             report_text: 主报告文本（总结 + 可视化）
             details: dict 包含需要拆分到独立文件的详情数据
                 - 'health_events': str 或 None
+                - 'load_select_release': str 或 None
                 - 'trace_files': {trace_id: text} 或 {}
     """
     parts = []
-    details = {"health_events": None, "trace_files": {}}
+    details = {"health_events": None, "load_select_release": None, "trace_files": {}}
 
     # 状态行
     parts.append(f"STATUS: {status} — {status_reason}")
@@ -168,7 +169,10 @@ def format_full_report(results, status, status_reason):
             details["health_events"] = detail
 
     if "load" in results:
-        parts.append(format_load_report(results["load"]))
+        summary, detail = format_load_report(results["load"])
+        parts.append(summary)
+        if detail:
+            details["load_select_release"] = detail
 
     if "cache" in results:
         parts.append(format_cache_report(results["cache"]))
@@ -208,6 +212,11 @@ def save_detailed_report(report_text, output_dir, details=None):
             with open(health_path, "w", encoding="utf-8") as f:
                 f.write(details["health_events"])
 
+        if details.get("load_select_release"):
+            load_path = os.path.join(detail_dir, "load_select_release.md")
+            with open(load_path, "w", encoding="utf-8") as f:
+                f.write(details["load_select_release"])
+
         for trace_id, trace_text in details.get("trace_files", {}).items():
             safe_id = trace_id.replace("/", "_")
             trace_path = os.path.join(detail_dir, f"trace_{safe_id}.md")