Initial commit: CI Code Viewer

.gitignore

+.DS_Store
+__pycache__/
+*.pyc
+*.pyo
+.env

README.md

+# AI Code Review
+
+GitLab CI 集成的 AI 代码审查工具，自动分析代码变更风险并通过飞书通知审查结果。
+
+## 特性
+
+- **三级智能分流**：LLM 自动判定审查级别（skip / standard / deep），节省 API 调用成本
+- **双重安全闸门**：实质代码变更不可跳过 + 安全关键词自动升级为深度审查
+- **多平台预设**：内置 Android / iOS / Frontend 专属审查规则，开箱即用
+- **飞书通知**：审查结果自动推送至飞书群
+- **零侵入接入**：项目只需在 `.gitlab-ci.yml` 中添加几行配置
+
+## 快速接入
+
+### 方式一：include 模板（推荐）
+
+在项目的 `.gitlab-ci.yml` 中添加：
+
+```yaml
+include:
+  - project: 'devops/ai-code-review'
+    file: '/templates/ai-review.yml'
+```
+
+### 方式二：直接引用脚本
+
+```yaml
+stages:
+  - review
+
+ai-code-review:
+  stage: review
+  tags:
+    - code-review
+  variables:
+    GIT_DEPTH: "50"
+    GIT_STRATEGY: fetch
+  script:
+    - echo "Starting AI code review..."
+    - rm -rf /tmp/ai-code-review-${CI_JOB_ID}
+    - git clone --depth 1 https://gitlab-ci-token:${CI_JOB_TOKEN}@git.zmcms.cn/fht/ai-code-review.git /tmp/ai-code-review-${CI_JOB_ID}
+    - python3 /tmp/ai-code-review-${CI_JOB_ID}/ci_review.py
+  after_script:
+    - rm -rf /tmp/ai-code-review-${CI_JOB_ID}
+  allow_failure: true
+```
+
+### 必要配置
+
+在 GitLab **Settings > CI/CD > Variables** 中添加：
+
+| 变量 | 必填 | 说明 |
+|------|------|------|
+| `AI_REVIEW_API_KEY` | ✅ | LLM API 密钥 |
+| `FEISHU_WEBHOOK_URL` | 否 | 飞书机器人 Webhook 地址 |
+
+## 环境变量
+
+### 核心配置
+
+| 变量 | 默认值 | 说明 |
+|------|--------|------|
+| `AI_REVIEW_API_URL` | `https://coding.dashscope.aliyuncs.com/v1/chat/completions` | LLM API 地址 |
+| `AI_REVIEW_API_KEY` | - | LLM API 密钥 |
+| `AI_REVIEW_MODEL` | `qwen3.5-plus` | 深度审查模型 |
+| `AI_REVIEW_TRIAGE_MODEL` | `qwen3-coder-plus` | 分流判定 & 标准审查模型 |
+| `AI_REVIEW_PLATFORM` | - | 平台预设：`android` / `ios` / `frontend` |
+
+### 审查参数
+
+| 变量 | 默认值 | 说明 |
+|------|--------|------|
+| `AI_REVIEW_MAX_DIFF_LINES` | `2000` | 审查 diff 最大行数 |
+| `AI_REVIEW_TRIAGE_MAX_DIFF_LINES` | `300` | 分流判定 diff 最大行数 |
+| `AI_REVIEW_STANDARD_MAX_TOKENS` | `4096` | 标准审查最大 token |
+| `AI_REVIEW_DEEP_MAX_TOKENS` | `8192` | 深度审查最大 token |
+| `AI_REVIEW_STANDARD_TIMEOUT` | `180` | 标准审查超时（秒） |
+| `AI_REVIEW_DEEP_TIMEOUT` | `300` | 深度审查超时（秒） |
+
+### 精细覆盖
+
+| 变量 | 说明 |
+|------|------|
+| `AI_REVIEW_SKIP_PATTERNS` | 跳过文件模式，逗号分隔（如 `*.lock,*.json`） |
+| `AI_REVIEW_FOCUS_AREAS` | 审查关注点，`\|` 分隔（覆盖平台预设） |
+
+## 审查流程
+
+```
+git diff → 文件过滤 → LLM 分流判定 → 审查 → 飞书通知
+                         │
+                    ┌─────┼─────┐
+                    ▼     ▼     ▼
+                  skip  standard deep
+                  (跳过) (轻量模型) (重量模型+thinking)
+                         │     │
+                    安全闸门检查：
+                    • 有实质代码变更 → 不可 skip
+                    • 命中安全关键词 → 升级为 deep
+```
+
+## 项目结构
+
+```
+ai_review/
+├── __init__.py     # 版本号
+├── __main__.py     # python -m ai_review 入口
+├── config.py       # 环境变量、平台预设
+├── git.py          # Git diff 获取、过滤、统计
+├── llm.py          # LLM 调用与 prompt 构建
+├── triage.py       # 分流判定与安全闸门
+├── notify.py       # 飞书通知
+└── main.py         # 主流程编排
+ci_review.py        # 薄包装入口（兼容现有 CI 调用）
+templates/
+└── ai-review.yml   # GitLab CI 模板
+```
+
+## 平台预设
+
+设置 `AI_REVIEW_PLATFORM` 环境变量可启用对应平台的预设规则：
+
+- **android**：关注空指针、内存泄漏、WebView 漏洞等，跳过 `.xml`、`.properties` 等配置文件
+- **ios**：关注强制解包、循环引用、Data Race 等，跳过 `.pbxproj`、`Pods/*` 等文件
+- **frontend**：关注 XSS、状态管理、异步问题等，跳过 `dist/*`、`node_modules/*` 等目录
+
+不设置则使用通用审查规则。

ai_review/__init__.py

+__version__ = "1.0.0"

ai_review/__main__.py

+from ai_review.main import main
+
+main()

ai_review/config.py

+# -*- coding: utf-8 -*-
+"""所有配置通过环境变量注入，不依赖项目内任何文件。"""
+
+import os
+
+# ─── 配置（全部从环境变量读取） ──────────────────────────────
+API_URL = os.environ.get("AI_REVIEW_API_URL", "https://coding.dashscope.aliyuncs.com/v1/chat/completions")
+API_KEY = os.environ.get("AI_REVIEW_API_KEY", "")
+MODEL = os.environ.get("AI_REVIEW_MODEL", "qwen3.5-plus")
+TRIAGE_MODEL = os.environ.get("AI_REVIEW_TRIAGE_MODEL", "qwen3-coder-plus")
+FEISHU_WEBHOOK = os.environ.get("FEISHU_WEBHOOK_URL", "")
+
+MAX_DIFF_LINES = int(os.environ.get("AI_REVIEW_MAX_DIFF_LINES", "2000"))
+STANDARD_MAX_TOKENS = int(os.environ.get("AI_REVIEW_STANDARD_MAX_TOKENS", "4096"))
+DEEP_MAX_TOKENS = int(os.environ.get("AI_REVIEW_DEEP_MAX_TOKENS", "8192"))
+STANDARD_TIMEOUT = int(os.environ.get("AI_REVIEW_STANDARD_TIMEOUT", "180"))
+DEEP_TIMEOUT = int(os.environ.get("AI_REVIEW_DEEP_TIMEOUT", "300"))
+
+# ─── 平台预设（android / ios / frontend，留空或其他值则使用通用配置） ───
+PLATFORM = os.environ.get("AI_REVIEW_PLATFORM", "").strip().lower()
+
+_PLATFORM_PROFILES = {
+    "android": {
+        "skip": "*.lock,*.json,*.toml,*.xml,*.properties,*.pro,*.txt,*.md",
+        "focus": [
+            "安全漏洞（硬编码密钥、SQL注入、XSS、WebView漏洞等）",
+            "空指针风险（NullPointerException）",
+            "内存泄漏（Activity/Fragment/Context泄漏）",
+            "线程安全问题（Handler/AsyncTask/协程misuse）",
+            "资源未关闭（Cursor/Stream/数据库连接等）",
+            "逻辑错误",
+            "性能问题（主线程阻塞、频繁GC、过度绘制等）",
+        ],
+    },
+    "ios": {
+        "skip": "*.lock,*.json,*.toml,*.xml,*.txt,*.md,"
+                "*.pbxproj,*.xcscheme,*.xcworkspacedata,Pods/*,*.xib.nib",
+        "focus": [
+            "安全漏洞（硬编码密钥、ATS配置、Keychain误用等）",
+            "空值风险（强制解包 Force Unwrap、隐式解包）",
+            "内存泄漏（循环引用、delegate强引用、闭包捕获self）",
+            "线程安全问题（Data Race、非主线程更新UI）",
+            "资源未关闭（FileHandle/URLSession/Timer未invalidate）",
+            "逻辑错误",
+            "性能问题（主线程阻塞、离屏渲染、AutoLayout性能等）",
+        ],
+    },
+    "frontend": {
+        "skip": "*.lock,*.json,*.md,*.txt,*.svg,*.png,*.jpg,*.ico,"
+                "*.css,*.scss,*.less,*.map,dist/*,node_modules/*,.next/*,build/*",
+        "focus": [
+            "安全漏洞（XSS、硬编码密钥/Token、innerHTML滥用、eval使用、CSRF等）",
+            "类型安全问题（any滥用、类型断言不当、缺少空值检查）",
+            "内存泄漏（未清理的定时器/事件监听/订阅、组件卸载未清理副作用）",
+            "异步问题（未处理的Promise rejection、竞态条件、async/await误用）",
+            "状态管理问题（不必要的重渲染、状态更新逻辑错误、props drilling）",
+            "逻辑错误",
+            "性能问题（大列表未虚拟化、频繁重渲染、bundle体积过大、图片未优化等）",
+        ],
+    },
+}
+
+_profile = _PLATFORM_PROFILES.get(PLATFORM, {})
+_default_skip = _profile.get("skip",
+    "*.lock,*.json,*.toml,*.xml,*.properties,*.pro,*.txt,*.md,"
+    "*.pbxproj,*.xcscheme,*.xcworkspacedata,Pods/*")
+_default_focus = _profile.get("focus", [
+    "安全漏洞（硬编码密钥、SQL注入、XSS等）",
+    "空指针/空值风险",
+    "内存泄漏（循环引用、资源未释放等）",
+    "线程安全问题",
+    "资源未关闭（文件句柄/数据库连接/网络会话等）",
+    "逻辑错误",
+    "性能问题（主线程阻塞、不必要的重复计算等）",
+])
+
+# 环境变量仍可精细覆盖（优先级最高）
+SKIP_PATTERNS = os.environ.get("AI_REVIEW_SKIP_PATTERNS", _default_skip).split(",")
+FOCUS_AREAS = os.environ.get("AI_REVIEW_FOCUS_AREAS", "").split("|") \
+    if os.environ.get("AI_REVIEW_FOCUS_AREAS") else _default_focus
+
+
+def format_duration(seconds):
+    if seconds < 60:
+        return f"{seconds:.1f}s"
+    return f"{int(seconds // 60)}m{int(seconds % 60)}s"

ai_review/git.py

+# -*- coding: utf-8 -*-
+"""Git diff 获取、过滤、统计。"""
+
+import fnmatch
+import os
+import subprocess
+
+from ai_review.config import SKIP_PATTERNS
+
+
+def get_ci_diff():
+    before_sha = os.environ.get("CI_COMMIT_BEFORE_SHA", "")
+    commit_sha = os.environ.get("CI_COMMIT_SHA", "HEAD")
+
+    if not before_sha or before_sha == "0000000000000000000000000000000000000000":
+        default_branch = os.environ.get("CI_DEFAULT_BRANCH", "master")
+        subprocess.run(["git", "fetch", "origin", default_branch], capture_output=True)
+        before_sha = f"origin/{default_branch}"
+
+    cmd = ["git", "diff", before_sha, commit_sha, "--diff-filter=ACMR", "--no-color"]
+    result = subprocess.run(cmd, capture_output=True, text=True)
+
+    if result.returncode != 0:
+        print(f"git diff 失败: {result.stderr}")
+        cmd = ["git", "diff", "HEAD~1", "HEAD", "--diff-filter=ACMR", "--no-color"]
+        result = subprocess.run(cmd, capture_output=True, text=True)
+
+    return filter_diff(result.stdout)
+
+
+def filter_diff(diff_text):
+    if not diff_text:
+        return ""
+    filtered_lines = []
+    skip_current = False
+    for line in diff_text.splitlines(True):
+        if line.startswith("diff --git"):
+            parts = line.split(" b/")
+            if len(parts) > 1:
+                filepath = parts[-1].strip()
+                skip_current = any(
+                    fnmatch.fnmatch(filepath, p.strip()) or fnmatch.fnmatch(os.path.basename(filepath), p.strip())
+                    for p in SKIP_PATTERNS
+                )
+            else:
+                skip_current = False
+        if not skip_current:
+            filtered_lines.append(line)
+    return "".join(filtered_lines)
+
+
+def truncate_diff(diff_text, max_lines):
+    lines = diff_text.splitlines(True)
+    if len(lines) <= max_lines:
+        return diff_text, False
+    truncated = "".join(lines[:max_lines])
+    truncated += f"\n\n... [已截断，共 {len(lines)} 行，仅展示前 {max_lines} 行] ...\n"
+    return truncated, True
+
+
+def diff_stats(diff_text):
+    lines = diff_text.splitlines()
+    added = sum(1 for l in lines if l.startswith("+") and not l.startswith("+++"))
+    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

ai_review/llm.py

+# -*- coding: utf-8 -*-
+"""LLM 调用与 prompt 构建。"""
+
+import json
+import urllib.request
+import urllib.error
+
+from ai_review.config import API_URL, API_KEY, MODEL, FOCUS_AREAS
+
+
+def build_standard_prompt(diff_text):
+    focus_text = "\n".join(f"  - {a}" for a in FOCUS_AREAS)
+    system_msg = "你是一位资深开发工程师，专门负责代码审查。请用中文回复，结论要可执行、可定位。"
+    user_msg = f"""请审查以下 git diff 代码变更，重点关注以下风险点：
+{focus_text}
+
+审查规则：
+- 只基于 diff 中可见内容下结论，不要猜测未展示代码
+- 每个问题必须给出定位（文件路径 + 尽量精确的行号范围）
+- 每个问题必须包含：风险原因、触发条件、修复建议
+- 无法确认的问题标记为"需补充上下文"，不要当作高风险定论
+- 优先发现真实风险，避免泛泛建议
+
+请严格按以下格式回复：
+
+## 🔴 高风险
+（必须修复；如无写"无"）
+- [标题]
+    - 位置: path:line
+    - 问题: ...
+    - 风险: ...
+    - 建议: ...
+
+## 🟡 中风险
+（建议修复；如无写"无"）
+- [标题]
+    - 位置: path:line
+    - 问题: ...
+    - 风险: ...
+    - 建议: ...
+
+## 🟢 低风险/建议
+（可优化项；如无写"无"）
+
+## 📋 总结
+（一句话总结 + 风险等级：✅ 安全 / ⚠️ 需关注 / 🚨 有风险）
+
+## ✅ 建议动作
+- [ ] 必须立即修复（最多 3 条）
+- [ ] 可在后续迭代修复（最多 3 条）
+
+如果没有发现明显问题，直接回复"✅ 代码审查通过，未发现明显风险点。"
+
+---
+以下是 diff 内容：
+
+```diff
+{diff_text}
+```"""
+    return system_msg, user_msg
+
+
+def build_deep_prompt(diff_text):
+    focus_text = "\n".join(f"  - {a}" for a in FOCUS_AREAS)
+    system_msg = (
+        "你是一位资深安全与架构工程师，专门负责深度代码审查。"
+        "请充分运用推理能力，逐步分析代码变更的安全性、逻辑正确性和架构影响。"
+        "请用中文回复，结论要可执行、可定位。"
+    )
+    user_msg = f"""请对以下 git diff 代码变更进行深度审查，重点关注以下风险点：
+{focus_text}
+  - 跨文件联动影响（接口变更、数据流变化）
+  - 安全攻击面评估（输入校验、权限控制、敏感数据处理）
+  - 并发与线程安全（竞态条件、死锁、数据竞争）
+
+深度审查规则：
+- 只基于 diff 中可见内容下结论，不要猜测未展示代码
+- 每个问题必须给出定位（文件路径 + 尽量精确的行号范围）
+- 每个问题必须包含：风险原因、触发条件、影响范围、修复建议
+- 无法确认的问题标记为"需补充上下文"，不要当作高风险定论
+- 优先发现真实风险，避免泛泛建议
+- 请逐步推理：先理解变更意图，再分析可能的风险路径
+- 对安全相关变更，请考虑潜在的攻击向量和利用方式
+
+请严格按以下格式回复：
+
+## 🔴 高风险
+（必须修复；如无写"无"）
+- [标题]
+    - 位置: path:line
+    - 问题: ...
+    - 风险: ...
+    - 影响范围: ...
+    - 建议: ...
+
+## 🟡 中风险
+（建议修复；如无写"无"）
+- [标题]
+    - 位置: path:line
+    - 问题: ...
+    - 风险: ...
+    - 影响范围: ...
+    - 建议: ...
+
+## 🟢 低风险/建议
+（可优化项；如无写"无"）
+
+## 🔍 深度分析
+- 变更意图: （一句话概括此次变更的目的）
+- 影响面: （列出可能受影响的模块/功能）
+- 安全评估: （是否引入新的攻击面或安全风险）
+
+## 📋 总结
+（一句话总结 + 风险等级：✅ 安全 / ⚠️ 需关注 / 🚨 有风险）
+
+## ✅ 建议动作
+- [ ] 必须立即修复（最多 3 条）
+- [ ] 可在后续迭代修复（最多 3 条）
+
+如果没有发现明显问题，直接回复"✅ 代码审查通过，未发现明显风险点。"
+
+---
+以下是 diff 内容：
+
+```diff
+{diff_text}
+```"""
+    return system_msg, user_msg
+
+
+def call_llm(system_msg, user_msg, max_tokens=4096, timeout=300, model=None, enable_thinking=False):
+    payload = {
+        "model": model or MODEL,
+        "messages": [
+            {"role": "system", "content": system_msg},
+            {"role": "user", "content": user_msg}
+        ],
+        "temperature": 0.1,
+        "max_tokens": max_tokens
+    }
+    if enable_thinking:
+        payload["enable_thinking"] = True
+        payload["thinking_budget"] = 8192
+    headers = {
+        "Content-Type": "application/json",
+        "Authorization": f"Bearer {API_KEY}"
+    }
+    data = json.dumps(payload).encode("utf-8")
+    req = urllib.request.Request(API_URL, data=data, headers=headers, method="POST")
+
+    try:
+        with urllib.request.urlopen(req, timeout=timeout) as resp:
+            result = json.loads(resp.read().decode("utf-8"))
+            choices = result.get("choices", [])
+            if not choices:
+                return "⚠️ API 返回了空结果，可能触发了内容安全策略。"
+            return choices[0].get("message", {}).get("content", "⚠️ API 返回格式异常")
+    except urllib.error.HTTPError as e:
+        body = e.read().decode("utf-8") if e.fp else ""
+        return f"❌ API 请求失败 (HTTP {e.code}): {body}"
+    except urllib.error.URLError as e:
+        return f"❌ 网络错误: {e.reason}"
+    except Exception as e:
+        return f"❌ 调用失败: {e}"

ai_review/main.py

+# -*- coding: utf-8 -*-
+"""主流程编排。"""
+
+import os
+import sys
+import time
+
+from ai_review.config import (
+    API_KEY, MODEL, TRIAGE_MODEL,
+    MAX_DIFF_LINES, STANDARD_MAX_TOKENS, DEEP_MAX_TOKENS,
+    STANDARD_TIMEOUT, DEEP_TIMEOUT,
+    format_duration,
+)
+from ai_review.git import get_ci_diff, truncate_diff, diff_stats
+from ai_review.llm import call_llm, build_standard_prompt, build_deep_prompt
+from ai_review.triage import (
+    build_triage_prompt, parse_triage_decision,
+    has_meaningful_code_change, security_keyword_gate,
+)
+from ai_review.notify import send_feishu
+
+
+def main():
+    total_start = time.time()
+
+    if not API_KEY:
+        print("错误: 未设置 AI_REVIEW_API_KEY 环境变量")
+        print("请在 GitLab Settings > CI/CD > Variables 中添加。")
+        sys.exit(0)
+
+    branch = os.environ.get("CI_COMMIT_BRANCH", os.environ.get("CI_COMMIT_REF_NAME", "unknown"))
+    committer = os.environ.get("CI_COMMIT_AUTHOR", os.environ.get("GITLAB_USER_NAME", "unknown"))
+    commit_msg = os.environ.get("CI_COMMIT_MESSAGE", "")[:100]
+    project_name = os.environ.get("CI_PROJECT_NAME", "unknown")
+    commit_sha_short = os.environ.get("CI_COMMIT_SHORT_SHA", "")
+    pipeline_url = os.environ.get("CI_PIPELINE_URL", "")
+
+    print(f"🤖 AI 代码审查 - {project_name}/{branch}")
+    print(f"   提交者: {committer}")
+    print(f"   提交信息: {commit_msg}")
+
+    t_diff = time.time()
+    diff = get_ci_diff()
+    diff_elapsed = time.time() - t_diff
+    print(f"   ⏱️ 阶段耗时 - 获取Diff: {format_duration(diff_elapsed)}")
+    if not diff.strip():
+        print("没有需要审查的代码变更，跳过。")
+        t_feishu = time.time()
+        send_feishu(
+            f"🤖 代码审查 - {project_name}/{branch}",
+            [
+                f"提交者: {committer}",
+                f"提交: {commit_sha_short} {commit_msg}",
+                f"分流模型: {TRIAGE_MODEL}",
+                "",
+                "✅ 没有需要审查的代码变更。",
+            ]
+        )
+        print(f"   ⏱️ 阶段耗时 - 飞书通知: {format_duration(time.time() - t_feishu)}")
+        print(f"   ⏱️ 总耗时: {format_duration(time.time() - total_start)}")
+        return
+
+    files, added, removed = diff_stats(diff)
+    print(f"   变更文件: {files} 个, +{added} -{removed} 行")
+
+    analysis_start = time.time()
+
+    # 先由 LLM 做快速分流，决定是否需要完整审查
+    triage_max_lines = int(os.environ.get("AI_REVIEW_TRIAGE_MAX_DIFF_LINES", "300"))
+    triage_diff, _ = truncate_diff(diff, triage_max_lines)
+    print("   ⏳ 正在调用大模型分流判定...")
+    t_triage = time.time()
+    triage_system_msg, triage_user_msg = build_triage_prompt(triage_diff)
+    triage_result = call_llm(
+        triage_system_msg,
+        triage_user_msg,
+        max_tokens=256,
+        timeout=120,
+        model=TRIAGE_MODEL,
+    )
+    triage_elapsed = time.time() - t_triage
+    print(f"   ⏱️ 阶段耗时 - 分流判定: {format_duration(triage_elapsed)}")
+    triage_decision, triage_reason, triage_category = parse_triage_decision(triage_result)
+
+    # 安全闸：只要检测到实质代码变更，就不允许 skip。
+    if triage_decision == "skip" and has_meaningful_code_change(triage_diff):
+        triage_decision = "standard"
+        triage_reason = "检测到实质代码变更，强制标准审查"
+
+    # 安全关键词闸：standard 命中安全关键词则提升为 deep
+    if triage_decision == "standard" and security_keyword_gate(diff):
+        triage_decision = "deep"
+        triage_reason = "检测到安全敏感关键词，提升为深度审查"
+
+    review_level = triage_decision  # skip / standard / deep
+    print(f"   📌 分流决策: {review_level} | 类型: {triage_category} | 原因: {triage_reason}")
+
+    if triage_decision == "skip":
+        review_result = f"✅ 代码审查通过，LLM 判定可跳过完整审查：{triage_reason}"
+        print(f"   ✅ 分流判定：跳过完整审查（{triage_reason}）")
+    elif triage_decision == "standard":
+        diff, was_truncated = truncate_diff(diff, MAX_DIFF_LINES)
+        if was_truncated:
+            print(f"   ⚠️ diff 过长，已截断至 {MAX_DIFF_LINES} 行")
+
+        print(f"   ⏳ 正在调用标准审查 ({TRIAGE_MODEL})...")
+        t_review = time.time()
+        system_msg, user_msg = build_standard_prompt(diff)
+        review_result = call_llm(
+            system_msg, user_msg,
+            max_tokens=STANDARD_MAX_TOKENS,
+            timeout=STANDARD_TIMEOUT,
+            model=TRIAGE_MODEL,
+        )
+        review_elapsed = time.time() - t_review
+        print(f"   ⏱️ 阶段耗时 - 标准审查: {format_duration(review_elapsed)}")
+    else:  # deep
+        diff, was_truncated = truncate_diff(diff, MAX_DIFF_LINES)
+        if was_truncated:
+            print(f"   ⚠️ diff 过长，已截断至 {MAX_DIFF_LINES} 行")
+
+        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)
+        review_result = call_llm(
+            system_msg, user_msg,
+            max_tokens=DEEP_MAX_TOKENS,
+            timeout=DEEP_TIMEOUT,
+            model=MODEL,
+            enable_thinking=True,
+        )
+        review_elapsed = time.time() - t_review
+        print(f"   ⏱️ 阶段耗时 - 深度审查: {format_duration(review_elapsed)}")
+
+    elapsed = time.time() - analysis_start
+    elapsed_str = format_duration(elapsed)
+    print(f"   ⏱️ 阶段耗时 - AI分析总计: {elapsed_str}")
+
+    print("\n" + "=" * 60)
+    print("📋 审查结果：")
+    print("=" * 60)
+    print(review_result)
+    print("=" * 60)
+
+    has_high_risk = "🔴" in review_result or "高风险" in review_result
+    is_pass = "✅" in review_result and not has_high_risk
+    risk_emoji = "✅" if is_pass else ("🚨" if has_high_risk else "⚠️")
+
+    # 构造 diff 链接
+    project_url = os.environ.get("CI_PROJECT_URL", "")
+    commit_sha_full = os.environ.get("CI_COMMIT_SHA", "")
+    diff_link = f"{project_url}/commit/{commit_sha_full}" if project_url and commit_sha_full else ""
+
+    _level_labels = {
+        "skip": "跳过",
+        "standard": f"标准审查 ({TRIAGE_MODEL})",
+        "deep": f"深度审查 ({MODEL}, thinking)",
+    }
+    review_level_label = _level_labels.get(review_level, review_level)
+
+    feishu_title = f"🤖 代码审查报告 - {project_name}/{branch} {risk_emoji}"
+    feishu_lines = [
+        f"提交者: {committer}",
+        f"分支: {branch}",
+        f"提交: {commit_sha_short} {commit_msg}",
+        f"变更: {files} 个文件, +{added} -{removed} 行",
+        f"分流模型: {TRIAGE_MODEL}",
+        f"审查层级: {review_level_label}",
+        f"变更类型: {triage_category}",
+        f"分析耗时: {elapsed_str}",
+    ]
+    if diff_link:
+        feishu_lines.append(f"Diff: {diff_link}")
+    if pipeline_url:
+        feishu_lines.append(f"流水线: {pipeline_url}")
+    feishu_lines.append("")
+    feishu_lines.append("────── 代码审查结果 ──────")
+    feishu_lines.append("")
+    for line in review_result.splitlines():
+        feishu_lines.append(line)
+
+    t_feishu = time.time()
+    send_feishu(feishu_title, feishu_lines)
+    print(f"   ⏱️ 阶段耗时 - 飞书通知: {format_duration(time.time() - t_feishu)}")
+    print(f"   ⏱️ 总耗时: {format_duration(time.time() - total_start)}")
+    print("\n✅ 审查完成。")

ai_review/notify.py

+# -*- coding: utf-8 -*-
+"""飞书 Webhook 通知。"""
+
+import json
+import urllib.request
+
+from ai_review.config import FEISHU_WEBHOOK
+
+
+def send_feishu(title, content_lines):
+    if not FEISHU_WEBHOOK:
+        print("未配置 FEISHU_WEBHOOK_URL，跳过飞书通知。")
+        return
+
+    elements = []
+    for line in content_lines:
+        if isinstance(line, list):
+            elements.append(line)
+        else:
+            elements.append([{"tag": "text", "text": line}])
+
+    payload = {
+        "msg_type": "post",
+        "content": {
+            "post": {
+                "zh_cn": {
+                    "title": title,
+                    "content": elements
+                }
+            }
+        }
+    }
+
+    data = json.dumps(payload).encode("utf-8")
+    req = urllib.request.Request(
+        FEISHU_WEBHOOK,
+        data=data,
+        headers={"Content-Type": "application/json"},
+        method="POST"
+    )
+
+    try:
+        with urllib.request.urlopen(req, timeout=15) as resp:
+            result = json.loads(resp.read().decode("utf-8"))
+            if result.get("code") == 0 or result.get("StatusCode") == 0:
+                print("✅ 飞书通知发送成功")
+            else:
+                print(f"⚠️ 飞书通知返回: {result}")
+    except Exception as e:
+        print(f"⚠️ 飞书通知发送失败: {e}")

ai_review/triage.py

+# -*- coding: utf-8 -*-
+"""分流判定：决定审查级别（skip / standard / deep）。"""
+
+import json
+import re
+
+
+def build_triage_prompt(diff_text):
+    system_msg = "你是代码审查分流器。你的默认策略是保守：有任何不确定即 deep。"
+    user_msg = f"""请判断下面的 git diff 应该进行哪种级别的审查。
+
+三级决策规则：
+1. decision=skip: 仅当变更为纯空白/空行/注释/格式化，且不影响行为时
+2. decision=standard: 简单重构（重命名、提取方法）、单文件bugfix、配置项变更、UI/样式调整、测试代码修改
+3. decision=deep: 以下任一情况：
+   - 安全相关（密钥、认证、权限、加密、SQL、XSS、CSRF）
+   - 多文件联动逻辑变更（≥3个文件有逻辑关联）
+   - 新增或修改依赖
+   - 核心业务逻辑改动（支付、用户数据、数据库schema）
+   - 并发/线程安全相关
+   - 不确定归属哪个级别
+
+变更类型 category 取值：refactor / bugfix / feature / security / config / dependency / test / other
+
+只输出 JSON，不要输出其他内容：
+{{"decision":"skip|standard|deep","reason":"一句话原因","category":"变更类型","evidence":"最多20字说明依据"}}
+
+```diff
+{diff_text}
+```"""
+    return system_msg, user_msg
+
+
+def parse_triage_decision(content):
+    if not content:
+        return "deep", "分流结果为空，默认深度审查", "unknown"
+
+    match = re.search(r"\{[\s\S]*\}", content)
+    raw_json = match.group(0) if match else content
+
+    try:
+        data = json.loads(raw_json)
+        decision = str(data.get("decision", "deep")).strip().lower()
+        reason = str(data.get("reason", "")).strip() or "未提供原因"
+        category = str(data.get("category", "other")).strip().lower()
+        if decision not in ("skip", "standard", "deep"):
+            return "deep", "分流结果非法，默认深度审查", category
+        return decision, reason, category
+    except Exception:
+        return "deep", "分流结果无法解析，默认深度审查", "unknown"
+
+
+def has_meaningful_code_change(diff_text):
+    """保守判定：只要存在非空白、非注释的增删行，就认为是实质代码变更。"""
+    comment_prefixes = ("#", "//", "/*", "*", "*/", "--")
+    for line in diff_text.splitlines():
+        if line.startswith("+++") or line.startswith("---"):
+            continue
+        if not (line.startswith("+") or line.startswith("-")):
+            continue
+
+        content = line[1:].strip()
+        if not content:
+            continue
+        if content.startswith(comment_prefixes):
+            continue
+
+        return True
+    return False
+
+
+def security_keyword_gate(diff_text):
+    """检测 diff 中是否包含安全敏感关键词，用于将 standard 提升为 deep。"""
+    pattern = re.compile(
+        r"(password|passwd|secret|private.?key|credential|"
+        r"access.?key|api.?key|auth.?token|bearer|"
+        r"encrypt|decrypt|cipher|hmac|"
+        r"injection|xss|csrf|ssrf|"
+        r"eval\s*\(|exec\s*\(|innerHTML|dangerouslySetInnerHTML|"
+        r"subprocess|os\.system|shell.?exec|"
+        r"chmod|chown|setuid|sudo)",
+        re.IGNORECASE
+    )
+    for line in diff_text.splitlines():
+        if not (line.startswith("+") or line.startswith("-")):
+            continue
+        if line.startswith("+++") or line.startswith("---"):
+            continue
+        if pattern.search(line):
+            return True
+    return False

ci_review.py

+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+GitLab CI AI Code Review + 飞书通知
+薄包装入口：兼容现有 CI 调用方式 `python3 ci_review.py`
+实际逻辑已拆分至 ai_review/ 包。
+"""
+
+from ai_review.main import main
+
+if __name__ == "__main__":
+    main()

templates/ai-review.yml

+# AI Code Review CI 模板
+# 其他项目通过 include 引用此文件即可自动接入 AI 代码审查
+#
+# 使用方法：在项目的 .gitlab-ci.yml 中添加：
+#   include:
+#     - project: 'devops/ai-code-review'
+#       file: '/templates/ai-review.yml'
+
+stages:
+  - review
+
+ai-code-review:
+  stage: review
+  tags:
+    - code-review
+  variables:
+    GIT_DEPTH: "50"
+    GIT_STRATEGY: fetch
+  before_script:
+    # 从中央仓库拉取审查脚本（使用 CI_JOB_TOKEN 自动鉴权，无需额外配置）
+    - rm -rf /tmp/ai-code-review-${CI_JOB_ID}
+    - git clone --depth 1 https://gitlab-ci-token:${CI_JOB_TOKEN}@git.zmcms.cn/fht/ai-code-review.git /tmp/ai-code-review-${CI_JOB_ID}
+  script:
+    - python3 /tmp/ai-code-review-${CI_JOB_ID}/ci_review.py || python /tmp/ai-code-review-${CI_JOB_ID}/ci_review.py
+  after_script:
+    - rm -rf /tmp/ai-code-review-${CI_JOB_ID}
+  allow_failure: true
+  rules:
+    - if: $CI_PIPELINE_SOURCE == "push"