git-ai-control 0.4.0 → 0.4.3

package/CHANGELOG.md

@@ -2,6 +2,34 @@
 
 本项目的所有重要变更都会记录在此文件中。
 
+## [0.4.3] - 2026-07-30
+
+### 修复
+
+- 使用 Linux Node.js 24 环境重新生成前端锁文件，补齐 Linux 发布环境所需的 `@emnapi/core` 和 `@emnapi/runtime` 可选依赖，修复 `npm ci` 跨平台解析失败。
+
+## [0.4.2] - 2026-07-30
+
+### 修复
+
+- 重建前端依赖锁文件，补齐 WASM 可选依赖的嵌套锁定版本，修复发布工作流中 `npm ci` 失败的问题。
+
+## [0.4.1] - 2026-07-30
+
+### 新增
+
+- 新增敏感内容脱敏和 Agent / 模型治理独立插件；
+- 仓库插件支持多个规则、全局兜底、仓库/目录/分支匹配和优先级；
+- Git AI 设置与插件管理均可查看已写入磁盘的 JSON 配置；
+- 插件管理使用统一添加顺序，新插件始终追加到列表最后；
+- 新增 Git AI 启用与暂停状态展示及本机服务复用检测。
+
+### 变更
+
+- `npx git-ai-control` 改为唯一启动命令；已运行时复用现有本机服务，移除 `start` 子命令；
+- 移除本地用量与成本看板、聚合数据文件和相关接口；
+- 配置查看弹窗改为默认全展开的 JSON 树，并加宽显示区域。
+
 ## [0.4.0] - 2026-07-30
 
 ### 新增

package/COMPATIBILITY.md

@@ -2,7 +2,7 @@
 
 ## 操作系统
 
-`npx git-ai-control start` 支持 macOS、Linux 和 Windows，三个系统共用相同的配置格式、过滤运行时和 Web 页面，仅用户级服务管理方式不同：
+`npx git-ai-control` 支持 macOS、Linux 和 Windows，三个系统共用相同的配置格式、过滤运行时和 Web 页面，仅用户级服务管理方式不同：
 
 - macOS：LaunchAgent；
 - Linux：systemd user；
@@ -22,6 +22,7 @@ Linux 需要可用的 systemd 用户会话。Windows 需要当前用户拥有创
 | 仓库级数据类型开关 | 支持 | 不支持 |
 | 仓库、路径、分支字段脱敏 | 支持 | 不支持 |
 | Skill 名称过滤 | 支持 | 不支持 |
+| 敏感内容脱敏、Agent/模型治理 | 支持 | 不支持 |
 | GitHub 仅上传脱敏 Token | 支持 | 官方版当前无法通过外部配置实现 |
 
 安装脚本会检查 `git-ai` 二进制是否包含 `custom_metrics.json` 扩展点：

package/README.md

@@ -15,7 +15,10 @@
 - 可删除仓库信息、项目路径和分支名称等敏感字段；
 - 可将允许上传的 Skill 项目目录替换为固定目录；
 - 可按关键词或正则拦截指定 Skill；
-- 仓库插件与 Skill 过滤插件使用各自独立的配置表单；
+- 仓库插件可按不同主机添加多个，并支持全局兜底、仓库、目录、分支与优先级；
+- 敏感内容脱敏插件可识别 API Key、私钥、凭据，并按规则脱敏或拦截；
+- Agent / 模型治理插件支持允许名单、阻止正则及审计/拦截模式；
+- Skill、敏感内容和 Agent / 模型治理均使用独立配置表单；
 - 支持配置验证、运行状态检查和本机过滤服务重启。
 
 ## 界面截图
@@ -28,7 +31,9 @@
 
 ![仓库插件与 Skill 过滤插件管理页面](docs/images/plugin-management.jpg)
 
-仓库插件可以同时配置匹配主机、允许上传的数据类型和允许保留的字段；Skill 过滤插件使用独立表单管理名称规则。每种插件类型只能配置一次，新增其他类型的插件会增加新的配置模块，不会替换已有模块。
+仓库插件可以同时配置匹配主机、允许上传的数据类型和允许保留的字段，也可以按不同主机添加多个。主机规则重叠时按列表顺序使用第一个命中插件；Skill 过滤插件使用独立表单管理名称规则，且只能添加一次。
+
+插件管理页面按添加顺序展示全部插件。无论添加仓库插件还是独立插件，新卡片都会追加到现有列表最后；这只影响页面展示顺序，不改变仓库规则的匹配优先级。
 
 ## 快速开始
 
@@ -44,10 +49,12 @@ Windows:       %USERPROFILE%\.git-ai\bin\git-ai.exe
 确认文件存在后运行：
 
 ```bash
-npx git-ai-control start
+npx git-ai-control
 ```
 
-该命令会自动下载最新版本、安装或更新本机服务，并打开 Git AI 配置中心。安装过程会：
+首次运行时，该命令会自动下载最新版本、安装本机服务，并打开 Git AI 配置中心；之后检测到已运行的配置中心时，会直接复用现有进程。
+
+首次安装过程会：
 
 1. 识别当前 Git AI 是否支持 `custom_metrics` 扩展；
 2. 将原始上报端点备份到仅当前用户可读的本机配置；
@@ -82,9 +89,18 @@ http://127.0.0.1:38742
 4. 关闭“仓库信息”“项目路径”和“分支名称”；
 5. 保存配置。
 
-同一个仓库插件也可以填写多个主机，每行一个。GitLab、Gitee 和自建 Git 服务不需要单独开发插件。
+同一个仓库插件也可以填写多个主机，每行一个；也可以为不同主机分别添加仓库插件。GitLab、Gitee 和自建 Git 服务不需要单独开发插件。
+
+仓库策略会先匹配特定规则，再匹配“全部仓库”全局规则；特定规则之间优先级数字越大越靠前，同优先级按插件列表顺序处理。仓库、目录、分支同时填写时需要全部命中。
+
+### 3. 配置敏感内容与 Agent
 
-### 3. 配置 Skill 过滤插件
+在“插件管理”中可分别添加以下独立插件：
+
+1. “敏感内容脱敏”：选择作用事件、内置规则和自定义正则；命中后可脱敏或阻止上报；
+2. “Agent / 模型治理”：配置允许的 Agent、模型与阻止正则；先使用“仅审计”确认命中情况，再按需改为阻止。
+
+### 4. 配置 Skill 过滤插件
 
 需要按名称拦截 Skill 时：
 
@@ -95,7 +111,7 @@ http://127.0.0.1:38742
 
 Skill 过滤插件只负责名称规则。仓库匹配、上传类型和目录替换仍由仓库插件负责。
 
-### 4. 验证配置
+### 5. 验证配置
 
 点击页面中的“验证配置”。结果会通过右上角 Toast 显示，不会占用配置表单空间。
 
@@ -119,6 +135,8 @@ Git AI
   -> 拦截，或脱敏后转发至原上报服务
 ```
 
+两个页签中的“查看配置”会重新读取当前磁盘内容，不会显示未保存的表单修改。Git AI 原生配置只展示本页面安全管理的字段，插件策略展示完整已写入策略。
+
 控制页面保存时只更新：
 
 - `~/.git-ai/config.json`：Git AI 原生设置；

package/bin/git-ai-control.js

@@ -5,6 +5,7 @@ import path from "node:path"
 import {fileURLToPath} from "node:url"
 
 import {browserCommand} from "../scripts/platform-services.mjs"
+import {controlPanelIsRunning} from "../scripts/install.mjs"
 
 const command = process.argv[2]
 const packageRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..")
@@ -14,17 +15,17 @@ function printHelp() {
   console.log(`Git AI Control Panel
 
 用法：
-  git-ai-control start    安装或更新本机服务，并打开配置页面
+  git-ai-control          首次安装服务；已运行时复用现有进程并打开配置页面
   git-ai-control uninstall 卸载本机服务并恢复上报端点
   git-ai-control --help   显示帮助`)
 }
 
-if (command === "--help" || command === "-h" || !command) {
+if (command === "--help" || command === "-h") {
   printHelp()
   process.exit(0)
 }
 
-if (!["start", "uninstall"].includes(command)) {
+if (command && command !== "uninstall") {
   console.error(`未知命令：${command}`)
   printHelp()
   process.exit(1)
@@ -36,10 +37,15 @@ try {
     await uninstall()
     process.exit(0)
   }
-  const {install} = await import("../scripts/install.mjs")
-  await install()
+  const reused = await controlPanelIsRunning()
+  if (!reused) {
+    const {install} = await import("../scripts/install.mjs")
+    await install()
+  } else {
+    console.log("检测到已运行的 Git AI 配置中心，复用现有进程")
+  }
 } catch (error) {
-  console.error(`${command === "start" ? "安装" : "卸载"}失败：${error.message}`)
+  console.error(`${command === "uninstall" ? "卸载" : "启动"}失败：${error.message}`)
   process.exit(1)
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "git-ai-control",
-  "version": "0.4.0",
+  "version": "0.4.3",
   "description": "Git AI 本地配置与上传隐私控制面板",
   "type": "module",
   "bin": {

package/plugins/upload-filter/plugin_filter_runtime.py

@@ -10,6 +10,7 @@ import subprocess
 import sys
 import urllib.error
 import urllib.request
+from fnmatch import fnmatchcase
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from pathlib import Path
 from urllib.parse import urlsplit
@@ -86,10 +87,26 @@ BRANCH_KEYS = {
     "baseref",
 }
 NORMALIZED_PATH_KEYS = {re.sub(r"[^a-z0-9]", "", key.lower()) for key in PATH_KEYS}
+AGENT_KEYS = {"agent", "agentname", "agentid", "agenttype", "tool", "toolname"}
+MODEL_KEYS = {"model", "modelname", "modelid", "llmmodel"}
+SENSITIVE_PATTERNS = {
+    "api_key": re.compile(
+        r"(?i)\b(?:sk|rk|ghp|github_pat|glpat|xox[baprs]|akia)[a-z0-9_=-]{12,}\b"
+    ),
+    "private_key": re.compile(
+        r"-----BEGIN [A-Z ]*PRIVATE KEY-----[\s\S]*?-----END [A-Z ]*PRIVATE KEY-----"
+    ),
+    "credential": re.compile(
+        r"(?i)\b(?:password|passwd|secret|api[_-]?key)\s*[:=]\s*([^\s,;\]\}]{4,})"
+    ),
+    "email": re.compile(r"(?i)\b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,}\b"),
+    "phone": re.compile(r"(?<!\d)(?:\+?\d[\s-]?){7,14}\d(?!\d)"),
+}
 DROP = object()
 
 DEFAULT_POLICY_CONFIG = {
-    "version": 1,
+    "version": 2,
+    "git_ai_enabled": True,
     "default_allow_unmatched": True,
     "skill_policy": {
         "installed": True,
@@ -120,7 +137,14 @@ DEFAULT_POLICY_CONFIG = {
             "id": "github",
             "name": "GitHub 仓库",
             "enabled": True,
-            "match": {"hosts": ["github.com"]},
+            "priority": 100,
+            "match": {
+                "all_repositories": False,
+                "hosts": ["github.com"],
+                "repositories": [],
+                "directories": [],
+                "branches": [],
+            },
             "fixed_project_directory": "",
             "allow": {
                 "token": False,
@@ -138,6 +162,37 @@ DEFAULT_POLICY_CONFIG = {
             },
         }
     ],
+    "plugin_order": ["repository:github", "skill_filter"],
+    "sensitive_data_policy": {
+        "installed": False,
+        "enabled": False,
+        "action": "redact",
+        "event_types": {
+            "token": True,
+            "skill": True,
+            "commit": True,
+            "checkpoint": True,
+            "agent": True,
+            "prompt_duration": True,
+            "prompt_report": True,
+        },
+        "built_in_rules": {
+            "api_key": True,
+            "private_key": True,
+            "credential": True,
+            "email": False,
+            "phone": False,
+        },
+        "custom_patterns": [],
+    },
+    "agent_model_policy": {
+        "installed": False,
+        "enabled": False,
+        "mode": "audit",
+        "allowed_agents": [],
+        "allowed_models": [],
+        "blocked_patterns": [],
+    },
 }
 
 
@@ -226,6 +281,42 @@ def iter_path_text(value):
             yield from iter_path_text(child)
 
 
+def iter_repository_text(value):
+    if isinstance(value, dict):
+        for key, child in value.items():
+            if is_repository_key(key):
+                yield from iter_all_text(child)
+            else:
+                yield from iter_repository_text(child)
+    elif isinstance(value, list):
+        for child in value:
+            yield from iter_repository_text(child)
+
+
+def iter_branch_text(value):
+    if isinstance(value, dict):
+        for key, child in value.items():
+            if is_branch_key(key):
+                yield from iter_all_text(child)
+            else:
+                yield from iter_branch_text(child)
+    elif isinstance(value, list):
+        for child in value:
+            yield from iter_branch_text(child)
+
+
+def iter_named_text(value, keys: set[str]):
+    if isinstance(value, dict):
+        for key, child in value.items():
+            if normalize_key(key) in keys:
+                yield from iter_all_text(child)
+            else:
+                yield from iter_named_text(child, keys)
+    elif isinstance(value, list):
+        for child in value:
+            yield from iter_named_text(child, keys)
+
+
 def read_remote_text(path: str) -> str:
     if not isinstance(path, str) or not path:
         return ""
@@ -255,7 +346,7 @@ def load_policy_config() -> tuple[dict, str]:
     try:
         with POLICY_CONFIG_PATH.open("r", encoding="utf-8") as handle:
             config = json.load(handle)
-        if not isinstance(config, dict) or config.get("version") != 1:
+        if not isinstance(config, dict) or config.get("version") not in {1, 2}:
             raise ValueError("unsupported policy version")
         if not isinstance(config.get("plugins"), list):
             raise ValueError("plugins must be a list")
@@ -273,24 +364,59 @@ def host_needles(plugin: dict) -> list[str]:
     ]
 
 
+def selector_matches(pattern: str, candidates) -> bool:
+    normalized_pattern = str(pattern).strip().lower()
+    if not normalized_pattern:
+        return False
+    for candidate in candidates:
+        normalized_candidate = str(candidate).lower()
+        if fnmatchcase(normalized_candidate, normalized_pattern):
+            return True
+        if "*" not in normalized_pattern and "?" not in normalized_pattern and normalized_pattern in normalized_candidate:
+            return True
+    return False
+
+
+def selectors_match(patterns, candidates) -> bool:
+    if not patterns:
+        return True
+    return any(selector_matches(pattern, candidates) for pattern in patterns)
+
+
 def plugin_matches(plugin: dict, payload, raw_body: bytes) -> bool:
+    match = plugin.get("match", {})
+    if match.get("all_repositories", False):
+        return True
     needles = host_needles(plugin)
-    if not needles:
+    repositories = match.get("repositories", [])
+    directories = match.get("directories", [])
+    branches = match.get("branches", [])
+    if not any((needles, repositories, directories, branches)):
         return False
     raw_text = raw_body.decode("utf-8", errors="ignore").lower()
-    if any(needle in raw_text for needle in needles):
-        return True
-    if any(needle in text.lower() for text in iter_all_text(payload) for needle in needles):
-        return True
-    for path in iter_path_text(payload):
-        remote_text = read_remote_text(path).lower()
-        if any(needle in remote_text for needle in needles):
-            return True
-    return False
+    path_values = list(iter_path_text(payload))
+    remote_values = [read_remote_text(path) for path in path_values]
+    host_values = [raw_text, *iter_all_text(payload), *remote_values]
+    repository_values = [*iter_repository_text(payload), *remote_values]
+    return (
+        selectors_match(needles, host_values)
+        and selectors_match(repositories, repository_values)
+        and selectors_match(directories, path_values)
+        and selectors_match(branches, iter_branch_text(payload))
+    )
+
+
+def plugin_sort_key(item) -> tuple[int, int, int]:
+    index, plugin = item
+    match = plugin.get("match", {})
+    global_rule = bool(match.get("all_repositories", False))
+    priority = plugin.get("priority", 100)
+    return (1 if global_rule else 0, -priority if isinstance(priority, int) else -100, index)
 
 
 def find_matching_plugin(config: dict, payload, raw_body: bytes) -> dict | None:
-    for plugin in config.get("plugins", []):
+    plugins = sorted(enumerate(config.get("plugins", [])), key=plugin_sort_key)
+    for _, plugin in plugins:
         if plugin.get("enabled", True) and plugin_matches(plugin, payload, raw_body):
             return plugin
     return None
@@ -353,6 +479,73 @@ def sanitize_fields(value, fields: dict, hosts: list[str]):
     return value
 
 
+def sensitive_policy_matches(policy: dict, event_type: str | None) -> bool:
+    return bool(
+        policy.get("installed", False)
+        and policy.get("enabled", False)
+        and policy.get("event_types", {}).get(event_type, False)
+    )
+
+
+def redact_sensitive_content(value, policy: dict) -> tuple[object, list[str]]:
+    patterns = [
+        (name, pattern)
+        for name, pattern in SENSITIVE_PATTERNS.items()
+        if policy.get("built_in_rules", {}).get(name, False)
+    ]
+    for index, source in enumerate(policy.get("custom_patterns", []), start=1):
+        try:
+            patterns.append((f"custom_{index}", re.compile(str(source), re.IGNORECASE)))
+        except re.error:
+            continue
+
+    findings: list[str] = []
+
+    def clean(current):
+        if isinstance(current, dict):
+            return {key: clean(child) for key, child in current.items()}
+        if isinstance(current, list):
+            return [clean(child) for child in current]
+        if not isinstance(current, str):
+            return current
+        sanitized = current
+        for name, pattern in patterns:
+            sanitized, count = pattern.subn(f"[已脱敏:{name}]", sanitized)
+            if count:
+                findings.append(name)
+        return sanitized
+
+    return clean(value), sorted(set(findings))
+
+
+def matches_allowlist(value: str, patterns) -> bool:
+    return any(selector_matches(pattern, [value]) for pattern in patterns)
+
+
+def governance_findings(payload, policy: dict) -> list[str]:
+    if not policy.get("installed", False) or not policy.get("enabled", False):
+        return []
+    agents = sorted(set(iter_named_text(payload, AGENT_KEYS)))
+    models = sorted(set(iter_named_text(payload, MODEL_KEYS)))
+    compiled_patterns = []
+    for source in policy.get("blocked_patterns", []):
+        try:
+            compiled_patterns.append(re.compile(str(source), re.IGNORECASE))
+        except re.error:
+            continue
+    findings = []
+    for label, values, allowlist in (
+        ("agent", agents, policy.get("allowed_agents", [])),
+        ("model", models, policy.get("allowed_models", [])),
+    ):
+        for value in values:
+            if any(pattern.search(value) for pattern in compiled_patterns):
+                findings.append(f"blocked_{label}")
+            elif allowlist and not matches_allowlist(value, allowlist):
+                findings.append(f"unapproved_{label}")
+    return sorted(set(findings))
+
+
 def evaluate_request(
     endpoint_path: str,
     payload,
@@ -366,6 +559,14 @@ def evaluate_request(
     if raw_body is None:
         raw_body = json.dumps(payload, ensure_ascii=False).encode("utf-8")
     event_type = EVENT_TYPES.get(endpoint_path)
+    if not config.get("git_ai_enabled", True):
+        return {
+            "blocked": True,
+            "reason": "git_ai_paused",
+            "payload": payload,
+            "plugin": None,
+            "config_error": config_error,
+        }
     plugin = find_matching_plugin(config, payload, raw_body)
     plugin_id = plugin.get("id") if plugin else None
 
@@ -419,12 +620,46 @@ def evaluate_request(
                 "config_error": config_error,
             }
 
+    sensitive_findings: list[str] = []
+    sensitive_policy = config.get("sensitive_data_policy", {})
+    if sensitive_policy_matches(sensitive_policy, event_type):
+        transformed, sensitive_findings = redact_sensitive_content(
+            transformed, sensitive_policy
+        )
+        if sensitive_findings and sensitive_policy.get("action") == "block":
+            return {
+                "blocked": True,
+                "reason": "sensitive_content",
+                "payload": payload,
+                "plugin": plugin_id,
+                "config_error": config_error,
+                "findings": sensitive_findings,
+            }
+
+    governance_findings_list = governance_findings(
+        transformed, config.get("agent_model_policy", {})
+    )
+    agent_model_policy = config.get("agent_model_policy", {})
+    if governance_findings_list and agent_model_policy.get("mode") == "block":
+        return {
+            "blocked": True,
+            "reason": "agent_model_policy",
+            "payload": payload,
+            "plugin": plugin_id,
+            "config_error": config_error,
+            "findings": governance_findings_list,
+        }
+
     return {
         "blocked": False,
         "reason": "",
         "payload": transformed,
         "plugin": plugin_id,
         "config_error": config_error,
+        "findings": {
+            "sensitive": sensitive_findings,
+            "agent_model": governance_findings_list,
+        },
     }
 
 
@@ -450,6 +685,10 @@ class Handler(BaseHTTPRequestHandler):
                 {
                     "ok": not bool(config_error or upstream_error),
                     "mode": "plugins",
+                    "gitAiEnabled": config.get("git_ai_enabled", True),
+                    "gitAiStatus": "enabled"
+                    if config.get("git_ai_enabled", True)
+                    else "paused",
                     "policyVersion": config.get("version"),
                     "plugins": [
                         plugin.get("id")
@@ -466,15 +705,7 @@ class Handler(BaseHTTPRequestHandler):
 
     def do_POST(self):
         endpoint_path = urlsplit(self.path).path
-        upstreams, upstream_error = load_upstreams()
-        if upstream_error:
-            self.send_json(
-                503,
-                {"code": -1, "data": None, "message": upstream_error},
-            )
-            return
-        upstream_url = upstreams.get(endpoint_path)
-        if not upstream_url:
+        if endpoint_path not in EVENT_TYPES:
             self.send_json(404, {"code": -1, "data": None, "message": "unknown endpoint"})
             return
 
@@ -491,6 +722,18 @@ class Handler(BaseHTTPRequestHandler):
             self.send_filtered(endpoint_path, decision["reason"], decision.get("plugin"))
             return
 
+        upstreams, upstream_error = load_upstreams()
+        if upstream_error:
+            self.send_json(
+                503,
+                {"code": -1, "data": None, "message": upstream_error},
+            )
+            return
+        upstream_url = upstreams.get(endpoint_path)
+        if not upstream_url:
+            self.send_json(404, {"code": -1, "data": None, "message": "unknown endpoint"})
+            return
+
         body = json.dumps(
             decision["payload"],
             ensure_ascii=False,

package/policy.example.json

@@ -1,5 +1,6 @@
 {
-  "version": 1,
+  "version": 2,
+  "git_ai_enabled": true,
   "default_allow_unmatched": true,
   "skill_policy": {
     "installed": true,
@@ -28,10 +29,13 @@
       "id": "github",
       "name": "GitHub 仓库",
       "enabled": true,
+      "priority": 100,
       "match": {
-        "hosts": [
-          "github.com"
-        ]
+        "all_repositories": false,
+        "hosts": ["github.com"],
+        "repositories": [],
+        "directories": [],
+        "branches": []
       },
       "fixed_project_directory": "",
       "allow": {
@@ -49,5 +53,39 @@
         "repository": false
       }
     }
-  ]
+  ],
+  "plugin_order": [
+    "repository:github",
+    "skill_filter"
+  ],
+  "sensitive_data_policy": {
+    "installed": false,
+    "enabled": false,
+    "action": "redact",
+    "event_types": {
+      "agent": true,
+      "checkpoint": true,
+      "commit": true,
+      "prompt_duration": true,
+      "prompt_report": true,
+      "skill": true,
+      "token": true
+    },
+    "built_in_rules": {
+      "api_key": true,
+      "credential": true,
+      "email": false,
+      "phone": false,
+      "private_key": true
+    },
+    "custom_patterns": []
+  },
+  "agent_model_policy": {
+    "installed": false,
+    "enabled": false,
+    "mode": "audit",
+    "allowed_agents": [],
+    "allowed_models": [],
+    "blocked_patterns": []
+  }
 }

package/scripts/check.mjs

@@ -36,6 +36,21 @@ if (
   process.exit(1)
 }
 
+if (
+  !run(python, [
+    ...prefix,
+    "-m",
+    "unittest",
+    "discover",
+    "-s",
+    "tests",
+    "-p",
+    "test_*.py",
+  ])
+) {
+  process.exit(1)
+}
+
 for (const file of [
   "bin/git-ai-control.js",
   "scripts/check.mjs",