feat: add cross-file symbol reference context to LLM prompts

Introduces `find_symbol_references` in `ai_review/git.py` to detect deleted symbols within a diff and search for remaining references across the repository using `git grep`. This provides the LLM with factual context to determine if callers have been properly updated, reducing "hallucinated" risks regarding unsynchronized changes.

Key changes:
- Implements heuristic-based symbol extraction for deleted lines (supporting Kotlin, Java, Swift, JS/TS, etc.).
- Filters out references within the files already modified in the current diff.
- Appends a formatted "Symbol Reference Context" block to both standard and deep review prompts.
- Limits analysis to the first 8 symbols and 20 references per symbol to manage prompt size.

ai_review/git.py

 # -*- coding: utf-8 -*-
-"""Git diff 获取、过滤、统计。"""
+"""Git diff 获取、过滤、统计、跨文件引用检索。"""
 
 import fnmatch
 import os
+import re
 import subprocess
 
 from ai_review.config import SKIP_PATTERNS
@@ -64,3 +65,106 @@ def diff_stats(diff_text):
     removed = sum(1 for l in lines if l.startswith("-") and not l.startswith("---"))
     files = sum(1 for l in lines if l.startswith("diff --git"))
     return files, added, removed
+
+
+# ─── 跨文件符号引用检索 ────────────────────────────────────────
+# 让 LLM 不再"假设"调用方有没有同步清理，而是直接看 git grep 结果。
+
+# 常见关键字 / 通用类型，避免被误识别为符号
+_REF_KEYWORDS = {
+    "if", "else", "for", "while", "return", "true", "false", "null", "this",
+    "new", "var", "val", "let", "const", "fun", "def", "class", "object",
+    "import", "package", "private", "public", "protected", "internal", "static",
+    "abstract", "final", "open", "override", "suspend", "inline", "data",
+    "void", "int", "long", "float", "double", "bool", "boolean", "string",
+    "self", "init", "deinit", "lazy", "lateinit", "function", "interface",
+    "enum", "struct", "trait", "extends", "implements", "throws", "throw",
+    "try", "catch", "finally", "switch", "case", "default", "break", "continue",
+    "operator", "infix", "companion", "sealed", "annotation", "typealias",
+}
+
+# 多语言"删除行可能是定义"的启发式
+_DEF_PATTERNS = [
+    # Kotlin/Java/Swift/Scala: fun/def/class/interface/object/enum/struct NAME
+    re.compile(r"\b(?:fun|def|class|interface|object|enum|struct|trait)\s+([A-Za-z_][A-Za-z0-9_]*)"),
+    # Java/C 风格方法: [modifiers] returnType NAME(
+    re.compile(r"\b(?:public|private|protected|internal|static|final|abstract|override|open)\s+(?:\w+\s+)*([A-Za-z_][A-Za-z0-9_]*)\s*\("),
+    # JS/TS: function NAME / const NAME = / let NAME =
+    re.compile(r"\b(?:function|const|let)\s+([A-Za-z_][A-Za-z0-9_]*)"),
+    # Kotlin 顶层属性: val NAME / var NAME
+    re.compile(r"\b(?:val|var)\s+([A-Za-z_][A-Za-z0-9_]*)"),
+]
+
+_MAX_REFS_PER_SYMBOL = 20
+_MAX_SYMBOLS = 8
+_MIN_SYMBOL_LEN = 4
+
+
+def _extract_removed_symbols(diff_text):
+    """从 diff 删除行里抽出可能的符号名（保持出现顺序）。"""
+    seen = set()
+    ordered = []
+    for line in diff_text.splitlines():
+        if not line.startswith("-") or line.startswith("---"):
+            continue
+        content = line[1:].strip()
+        if not content:
+            continue
+        for pat in _DEF_PATTERNS:
+            for m in pat.finditer(content):
+                name = m.group(1)
+                if (len(name) < _MIN_SYMBOL_LEN
+                        or name.lower() in _REF_KEYWORDS
+                        or name in seen):
+                    continue
+                seen.add(name)
+                ordered.append(name)
+    return ordered
+
+
+def find_symbol_references(diff_text, project_root="."):
+    """识别 diff 里被删除的符号，在项目其它文件中 grep 引用情况。
+
+    返回一段可拼到 prompt 末尾的 Markdown 文本；diff 里没有可分析的删除符号
+    时返回空串。grep 的结果会过滤掉 diff 自身改动的文件（本就在 diff 里）。
+    """
+    symbols = _extract_removed_symbols(diff_text)
+    if not symbols:
+        return ""
+    symbols = symbols[:_MAX_SYMBOLS]
+
+    changed_files = set()
+    for line in diff_text.splitlines():
+        if line.startswith("diff --git"):
+            parts = line.split(" b/")
+            if len(parts) > 1:
+                changed_files.add(parts[-1].strip())
+
+    blocks = []
+    for sym in symbols:
+        try:
+            result = subprocess.run(
+                ["git", "grep", "-n", "-F", sym],
+                capture_output=True, text=True, cwd=project_root,
+                timeout=10,
+            )
+        except (subprocess.TimeoutExpired, FileNotFoundError):
+            continue
+
+        external = []
+        for ln in result.stdout.splitlines():
+            file_part = ln.split(":", 1)[0]
+            if file_part in changed_files:
+                continue
+            external.append(ln)
+
+        if not external:
+            blocks.append(f"### `{sym}`\n  （在其它文件中未找到引用，调用方应已清理）")
+        else:
+            shown = external[:_MAX_REFS_PER_SYMBOL]
+            lines = [f"### `{sym}`"] + [f"  {ln}" for ln in shown]
+            if len(external) > _MAX_REFS_PER_SYMBOL:
+                lines.append(f"  ... 还有 {len(external) - _MAX_REFS_PER_SYMBOL} 处")
+            blocks.append("\n".join(lines))
+
+    return "\n\n".join(blocks)

ai_review/main.py

@@ -11,7 +11,7 @@ from ai_review.config import (
     STANDARD_TIMEOUT, DEEP_TIMEOUT,
     format_duration,
 )
-from ai_review.git import get_ci_diff, truncate_diff, diff_stats
+from ai_review.git import get_ci_diff, truncate_diff, diff_stats, find_symbol_references
 from ai_review.llm import call_llm, build_standard_prompt, build_deep_prompt
 from ai_review.triage import (
     build_triage_prompt, parse_triage_decision,
@@ -20,6 +20,15 @@ from ai_review.triage import (
 from ai_review.notify import send_feishu
 
 
+_REFS_PROMPT_HEADER = (
+    "\n\n## 项目内符号引用上下文\n\n"
+    "以下是 diff 中被删除符号在当前代码库其它位置的引用情况，请优先依据这些事实"
+    "判断调用方是否同步处理：若某符号"
+    "「在其它文件中未找到引用」，请直接放行不要列为"
+    "需要补充上下文的风险；若仍有引用，请明确指出哪些文件未同步并定为高/中风险。\n\n"
+)
+
+
 def main():
     total_start = time.time()
 
@@ -99,9 +108,15 @@ def main():
         if was_truncated:
             print(f"   ⚠️ diff 过长，已截断至 {MAX_DIFF_LINES} 行")
 
+        refs_block = find_symbol_references(diff)
+        if refs_block:
+            print(f"   🔍 已附加 {refs_block.count('### `')} 个符号的项目引用上下文")
+
         print(f"   ⏳ 正在调用标准审查 ({TRIAGE_MODEL})...")
         t_review = time.time()
         system_msg, user_msg = build_standard_prompt(diff)
+        if refs_block:
+            user_msg += _REFS_PROMPT_HEADER + refs_block
         review_result, review_model_used = call_llm(
             system_msg, user_msg,
             max_tokens=STANDARD_MAX_TOKENS,
@@ -115,9 +130,15 @@ def main():
         if was_truncated:
             print(f"   ⚠️ diff 过长，已截断至 {MAX_DIFF_LINES} 行")
 
+        refs_block = find_symbol_references(diff)
+        if refs_block:
+            print(f"   🔍 已附加 {refs_block.count('### `')} 个符号的项目引用上下文")
+
         print(f"   ⏳ 正在调用深度审查 ({MODEL}, thinking mode)...")
         t_review = time.time()
         system_msg, user_msg = build_deep_prompt(diff)
+        if refs_block:
+            user_msg += _REFS_PROMPT_HEADER + refs_block
         review_result, review_model_used = call_llm(
             system_msg, user_msg,
             max_tokens=DEEP_MAX_TOKENS,