@spaceflow/review 5.0.0 → 5.0.2

package/CHANGELOG.md

@@ -1,5 +1,23 @@
 # Changelog
 
+## [5.0.1](https://github.com/Lydanne/spaceflow/compare/@spaceflow/review@5.0.0...@spaceflow/review@5.0.1) (2026-04-14)
+
+### 修复BUG
+
+* **review:** 修正已解决问题统计逻辑,避免与已修复问题重复计数 ([6d191bb](https://github.com/Lydanne/spaceflow/commit/6d191bbd71b8ce6194ce56a8d008dd966173541b))
+
+### 代码重构
+
+* **shared:** 移除 Claude Code 相关配置和文档,统一使用 OpenAI/Gemini/OpenCode ([b0754d6](https://github.com/Lydanne/spaceflow/commit/b0754d64f8991a10708730c191cf0f335e8aa6ca))
+
+### 其他修改
+
+* **cli:** released version 5.0.1 [no ci] ([f3009ea](https://github.com/Lydanne/spaceflow/commit/f3009eaaf26759d708541f54e707882b4927ebbd))
+* **core:** released version 5.0.1 [no ci] ([7904b9b](https://github.com/Lydanne/spaceflow/commit/7904b9bbc619c91a3a4fa1c8ea04d9dc548495ad))
+* **core:** 移除 @anthropic-ai/claude-agent-sdk 依赖及相关 sharp 图像处理包 ([7b50e06](https://github.com/Lydanne/spaceflow/commit/7b50e066df97d408eb63707eb27273592a28d34d))
+* **publish:** released version 5.0.1 [no ci] ([b14849f](https://github.com/Lydanne/spaceflow/commit/b14849ffafd38f230a3e69c2d63c468cb1946388))
+* **shared:** released version 5.0.1 [no ci] ([f70cd93](https://github.com/Lydanne/spaceflow/commit/f70cd930b01ffa09187f66522c8d6b18f835ec53))
+
 ## [4.0.1](https://github.com/Lydanne/spaceflow/compare/@spaceflow/review@4.0.0...@spaceflow/review@4.0.1) (2026-04-13)
 
 ### 新特性

package/README.md

@@ -3,7 +3,7 @@
 [![npm version](https://img.shields.io/npm/v/@spaceflow/review?color=blue)](https://www.npmjs.com/package/@spaceflow/review)
 [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 
-> Spaceflow AI 代码审查扩展，使用 LLM 对 PR 代码进行自动审查。支持 OpenAI、Claude Code、Gemini、Open Code 等多种 LLM 模式。
+> Spaceflow AI 代码审查扩展，使用 LLM 对 PR 代码进行自动审查。支持 OpenAI、Gemini、Open Code 等多种 LLM 模式。
 
 ## 安装
 
@@ -13,7 +13,7 @@ pnpm spaceflow install @spaceflow/review
 
 ## 功能特性
 
-- **多 LLM 支持** — OpenAI、Claude Code、Gemini、Open Code 可选
+- **多 LLM 支持** — OpenAI、Gemini、Open Code 可选
 - **行级评论** — 在 PR 中精确定位问题代码行
 - **增量审查** — 多次运行时自动去重，追踪问题修复状态
 - **删除代码分析** — 评估删除代码可能带来的风险
@@ -63,7 +63,7 @@ spaceflow review -p 123 --fail-on-issues -l openai
 | `--pr-number <number>`            | `-p` | PR 编号                                                      |
 | `--base <ref>`                    | `-b` | 基准分支/tag                                                 |
 | `--head <ref>`                    |      | 目标分支/tag                                                 |
-| `--llm-mode <mode>`               | `-l` | LLM 模式（`openai` / `claude-code` / `gemini` / `open-code`） |
+| `--llm-mode <mode>`               | `-l` | LLM 模式（`openai` / `gemini` / `open-code`） |
 | `--files <files...>`              | `-f` | 仅审查指定文件                                               |
 | `--commits <commits...>`          |      | 仅审查指定 commits                                           |
 | `--includes <patterns...>`        | `-i` | 文件 glob 过滤模式                                           |
@@ -74,7 +74,7 @@ spaceflow review -p 123 --fail-on-issues -l openai
 | `--no-verify-fixes`               |      | 禁用历史问题验证                                             |
 | `--analyze-deletions`             |      | 分析删除代码影响                                             |
 | `--deletion-only`                 |      | 仅执行删除代码分析                                           |
-| `--deletion-analysis-mode <mode>` |      | 删除分析 LLM 模式（`openai` / `claude-code`）               |
+| `--deletion-analysis-mode <mode>` |      | 删除分析 LLM 模式（`openai` / `open-code`）                 |
 | `--generate-description`          |      | 使用 AI 生成 PR 功能描述                                     |
 | `--output-format <format>`        | `-o` | 输出格式（`markdown` / `terminal` / `json`）                 |
 | `--local [mode]`                  |      | 本地审查模式（`uncommitted` / `staged`，默认 `uncommitted`） |
@@ -157,7 +157,7 @@ spaceflow review -p 123 --fail-on-issues -l openai
 | `get_rule_detail`      | 获取单条规则的详细信息                     | `ruleId`                                        |
 | `get_rules_from_dir`   | 从指定目录加载审查规则                     | `dirPath`, `includeExamples?`                   |
 
-规则搜索目录包括：`review.references` 配置路径、`.claude/skills`、`.cursor/skills`、`review-specs`。
+规则搜索目录包括：`review.references` 配置路径、`.cursor/skills`、`review-specs`。
 
 ## 审查规范格式
 
@@ -437,12 +437,6 @@ jobs:
 | `OPENAI_API_KEY`  | OpenAI API Key  |
 | `OPENAI_MODEL`    | OpenAI 模型名称 |
 
-### Claude Code
-
-| 变量                | 说明             |
-| ------------------- | ---------------- |
-| `ANTHROPIC_API_KEY` | Anthropic API Key |
-
 ### Gemini
 
 | 变量               | 说明           |

package/dist/index.js

@@ -42,9 +42,9 @@ __webpack_require__.d(__webpack_exports__, {
 ;// CONCATENATED MODULE: external "@spaceflow/core"
 
 ;// CONCATENATED MODULE: ./src/locales/zh-cn/review.json
-var review_namespaceObject = JSON.parse('{"description":"代码审查命令，使用 LLM 对 PR 代码进行自动审查","options.dryRun":"仅打印将要执行的操作，不实际提交评论","options.prNumber":"PR 编号，如果不指定则从环境变量获取","options.base":"基准分支/tag，用于比较差异","options.head":"目标分支/tag，用于比较差异","options.includes":"要审查的文件 glob 模式，如 *.ts *.js（可多次指定）。支持变更类型前缀: added|*.ts（仅新增）、modified|*.ts（仅修改）、deleted|*.ts（仅删除）","options.llmMode":"使用的 LLM 模式: claude-code, openai, gemini","options.files":"仅审查指定的文件（空格分隔）","options.commits":"仅审查指定的 commits（空格分隔）","options.verifyFixes":"是否验证历史问题是否已修复（默认从配置文件读取）","options.noVerifyFixes":"禁用历史问题验证","options.verifyConcurrency":"验证历史问题的并发数（默认 10）","options.analyzeDeletions":"分析删除代码可能带来的影响 (true, false, ci, pr, terminal)","options.deletionAnalysisMode":"删除代码分析模式: openai (标准模式) 或 claude-code (Agent 模式，可使用工具)","options.deletionOnly":"仅执行删除代码分析，跳过常规代码审查","options.outputFormat":"输出格式: markdown, terminal, json。不指定则智能选择（PR 用 markdown，终端用 terminal）","options.generateDescription":"使用 AI 生成 PR 功能描述","options.showAll":"显示所有发现的问题，不过滤非变更行的问题","options.flush":"仅刷新状态（同步 reactions、resolved conversations、replies），不执行 LLM 审查","options.eventAction":"PR 事件类型（opened, synchronize, closed 等），closed 时仅收集统计不进行 AI 审查","options.local":"审查本地未提交的代码。模式: \'uncommitted\'（默认，暂存区+工作区）或 \'staged\'（仅暂存区）","options.noLocal":"禁用本地模式，使用分支比较","options.failOnIssues":"存在未解决问题时工作流抛出异常。模式: warn（仅 warn 问题）/ error（仅 error 问题）/ warn&error（warn 或 error 问题）。不加模式等同 error","extensionDescription":"代码审查命令，使用 LLM 对 PR 代码进行自动审查","mcp.serverDescription":"代码审查规则查询服务","mcp.listRules":"获取当前项目的所有代码审查规则，返回规则列表包含 ID、标题、描述、适用的文件扩展名等信息","mcp.getRulesForFile":"获取某个文件应该使用的代码审查规则，根据文件扩展名和 includes 配置过滤","mcp.getRuleDetail":"获取某个规则的完整详情，包括描述、示例代码等","mcp.ruleNotFound":"规则 {{ruleId}} 不存在","mcp.dto.cwd":"项目根目录路径，默认为当前工作目录","mcp.dto.filePath":"文件路径，可以是相对路径或绝对路径","mcp.dto.includeExamples":"是否包含规则示例代码，默认 false","mcp.dto.ruleId":"规则 ID，如 JsTs.Naming.FileName","mcp.getRulesFromDir":"从指定目录加载代码审查规则。读取目录下所有 .md 文件，解析为规则并按规则 ID 去重（后加载的覆盖先加载的）","mcp.dto.dirPath":"包含规则 .md 文件的目录路径，可以是相对路径或绝对路径"}')
+var review_namespaceObject = JSON.parse('{"description":"代码审查命令，使用 LLM 对 PR 代码进行自动审查","options.dryRun":"仅打印将要执行的操作，不实际提交评论","options.prNumber":"PR 编号，如果不指定则从环境变量获取","options.base":"基准分支/tag，用于比较差异","options.head":"目标分支/tag，用于比较差异","options.includes":"要审查的文件 glob 模式，如 *.ts *.js（可多次指定）。支持变更类型前缀: added|*.ts（仅新增）、modified|*.ts（仅修改）、deleted|*.ts（仅删除）","options.llmMode":"使用的 LLM 模式: openai, gemini, open-code","options.files":"仅审查指定的文件（空格分隔）","options.commits":"仅审查指定的 commits（空格分隔）","options.verifyFixes":"是否验证历史问题是否已修复（默认从配置文件读取）","options.noVerifyFixes":"禁用历史问题验证","options.verifyConcurrency":"验证历史问题的并发数（默认 10）","options.analyzeDeletions":"分析删除代码可能带来的影响 (true, false, ci, pr, terminal)","options.deletionAnalysisMode":"删除代码分析模式: openai (标准模式) 或 open-code (Agent 模式，可使用工具)","options.deletionOnly":"仅执行删除代码分析，跳过常规代码审查","options.outputFormat":"输出格式: markdown, terminal, json。不指定则智能选择（PR 用 markdown，终端用 terminal）","options.generateDescription":"使用 AI 生成 PR 功能描述","options.showAll":"显示所有发现的问题，不过滤非变更行的问题","options.flush":"仅刷新状态（同步 reactions、resolved conversations、replies），不执行 LLM 审查","options.eventAction":"PR 事件类型（opened, synchronize, closed 等），closed 时仅收集统计不进行 AI 审查","options.local":"审查本地未提交的代码。模式: \'uncommitted\'（默认，暂存区+工作区）或 \'staged\'（仅暂存区）","options.noLocal":"禁用本地模式，使用分支比较","options.failOnIssues":"存在未解决问题时工作流抛出异常。模式: warn（仅 warn 问题）/ error（仅 error 问题）/ warn&error（warn 或 error 问题）。不加模式等同 error","extensionDescription":"代码审查命令，使用 LLM 对 PR 代码进行自动审查","mcp.serverDescription":"代码审查规则查询服务","mcp.listRules":"获取当前项目的所有代码审查规则，返回规则列表包含 ID、标题、描述、适用的文件扩展名等信息","mcp.getRulesForFile":"获取某个文件应该使用的代码审查规则，根据文件扩展名和 includes 配置过滤","mcp.getRuleDetail":"获取某个规则的完整详情，包括描述、示例代码等","mcp.ruleNotFound":"规则 {{ruleId}} 不存在","mcp.dto.cwd":"项目根目录路径，默认为当前工作目录","mcp.dto.filePath":"文件路径，可以是相对路径或绝对路径","mcp.dto.includeExamples":"是否包含规则示例代码，默认 false","mcp.dto.ruleId":"规则 ID，如 JsTs.Naming.FileName","mcp.getRulesFromDir":"从指定目录加载代码审查规则。读取目录下所有 .md 文件，解析为规则并按规则 ID 去重（后加载的覆盖先加载的）","mcp.dto.dirPath":"包含规则 .md 文件的目录路径，可以是相对路径或绝对路径"}')
 ;// CONCATENATED MODULE: ./src/locales/en/review.json
-var en_review_namespaceObject = JSON.parse('{"description":"Code review command using LLM for automated PR review","options.dryRun":"Only print actions without posting comments","options.prNumber":"PR number, auto-detected from env if not specified","options.base":"Base branch/tag for diff comparison","options.head":"Head branch/tag for diff comparison","options.includes":"File glob patterns to review, e.g. *.ts *.js (can be specified multiple times). Supports change-type prefixes: added|*.ts (new files only), modified|*.ts (modified only), deleted|*.ts (deleted only)","options.llmMode":"LLM mode: claude-code, openai, gemini","options.files":"Only review specified files (space-separated)","options.commits":"Only review specified commits (space-separated)","options.verifyFixes":"Verify if historical issues are fixed (default from config)","options.noVerifyFixes":"Disable historical issue verification","options.verifyConcurrency":"Concurrency for verifying historical issues (default 10)","options.analyzeDeletions":"Analyze impact of deleted code (true, false, ci, pr, terminal)","options.deletionAnalysisMode":"Deletion analysis mode: openai (standard) or claude-code (Agent mode with tools)","options.deletionOnly":"Only run deletion analysis, skip regular code review","options.outputFormat":"Output format: markdown, terminal, json. Auto-selected if not specified (markdown for PR, terminal for CLI)","options.generateDescription":"Generate PR description using AI","options.showAll":"Show all issues found, including those on unchanged lines","options.flush":"Only sync status (reactions, resolved conversations, replies), skip LLM review","options.eventAction":"PR event type (opened, synchronize, closed, etc.), closed only collects stats without AI review","options.local":"Review local uncommitted code. Mode: \'uncommitted\' (default, staged + working) or \'staged\' (only staged)","options.noLocal":"Disable local mode, use branch comparison instead","options.failOnIssues":"Fail the workflow when unresolved issues exist. Mode: warn (warn-level only) / error (error-level only) / warn+error (either warn or error). Passing no mode is equivalent to error","extensionDescription":"Code review command using LLM for automated PR review","mcp.serverDescription":"Code review rules query service","mcp.listRules":"List all code review rules for the current project, returning rule list with ID, title, description, applicable file extensions, etc.","mcp.getRulesForFile":"Get applicable code review rules for a specific file, filtered by file extension and includes configuration","mcp.getRuleDetail":"Get full details of a specific rule, including description, example code, etc.","mcp.ruleNotFound":"Rule {{ruleId}} not found","mcp.dto.cwd":"Project root directory path, defaults to current working directory","mcp.dto.filePath":"File path, can be relative or absolute","mcp.dto.includeExamples":"Whether to include rule example code, defaults to false","mcp.dto.ruleId":"Rule ID, e.g. JsTs.Naming.FileName","mcp.getRulesFromDir":"Load code review rules from a specific directory. Reads all .md files in the directory, parses them into rules, and deduplicates by rule ID (later rules override earlier ones)","mcp.dto.dirPath":"Directory path containing rule .md files, can be relative or absolute"}')
+var en_review_namespaceObject = JSON.parse('{"description":"Code review command using LLM for automated PR review","options.dryRun":"Only print actions without posting comments","options.prNumber":"PR number, auto-detected from env if not specified","options.base":"Base branch/tag for diff comparison","options.head":"Head branch/tag for diff comparison","options.includes":"File glob patterns to review, e.g. *.ts *.js (can be specified multiple times). Supports change-type prefixes: added|*.ts (new files only), modified|*.ts (modified only), deleted|*.ts (deleted only)","options.llmMode":"LLM mode: openai, gemini, open-code","options.files":"Only review specified files (space-separated)","options.commits":"Only review specified commits (space-separated)","options.verifyFixes":"Verify if historical issues are fixed (default from config)","options.noVerifyFixes":"Disable historical issue verification","options.verifyConcurrency":"Concurrency for verifying historical issues (default 10)","options.analyzeDeletions":"Analyze impact of deleted code (true, false, ci, pr, terminal)","options.deletionAnalysisMode":"Deletion analysis mode: openai (standard) or open-code (Agent mode with tools)","options.deletionOnly":"Only run deletion analysis, skip regular code review","options.outputFormat":"Output format: markdown, terminal, json. Auto-selected if not specified (markdown for PR, terminal for CLI)","options.generateDescription":"Generate PR description using AI","options.showAll":"Show all issues found, including those on unchanged lines","options.flush":"Only sync status (reactions, resolved conversations, replies), skip LLM review","options.eventAction":"PR event type (opened, synchronize, closed, etc.), closed only collects stats without AI review","options.local":"Review local uncommitted code. Mode: \'uncommitted\' (default, staged + working) or \'staged\' (only staged)","options.noLocal":"Disable local mode, use branch comparison instead","options.failOnIssues":"Fail the workflow when unresolved issues exist. Mode: warn (warn-level only) / error (error-level only) / warn+error (either warn or error). Passing no mode is equivalent to error","extensionDescription":"Code review command using LLM for automated PR review","mcp.serverDescription":"Code review rules query service","mcp.listRules":"List all code review rules for the current project, returning rule list with ID, title, description, applicable file extensions, etc.","mcp.getRulesForFile":"Get applicable code review rules for a specific file, filtered by file extension and includes configuration","mcp.getRuleDetail":"Get full details of a specific rule, including description, example code, etc.","mcp.ruleNotFound":"Rule {{ruleId}} not found","mcp.dto.cwd":"Project root directory path, defaults to current working directory","mcp.dto.filePath":"File path, can be relative or absolute","mcp.dto.includeExamples":"Whether to include rule example code, defaults to false","mcp.dto.ruleId":"Rule ID, e.g. JsTs.Naming.FileName","mcp.getRulesFromDir":"Load code review rules from a specific directory. Reads all .md files in the directory, parses them into rules, and deduplicates by rule ID (later rules override earlier ones)","mcp.dto.dirPath":"Directory path containing rule .md files, can be relative or absolute"}')
 ;// CONCATENATED MODULE: ./src/locales/index.ts
 
 
@@ -2252,9 +2252,9 @@ function generateIssueKey(issue) {
     const validIssue = issues.filter((i)=>i.valid !== "false");
     const validTotal = validIssue.length;
     const fixed = validIssue.filter((i)=>i.fixed).length;
-    const resolved = validIssue.filter((i)=>i.resolved && !i.fixed).length;
+    const resolved = validIssue.filter((i)=>i.resolved).length;
     const invalid = total - validTotal;
-    const pending = validTotal - fixed - resolved;
+    const pending = validTotal - validIssue.filter((i)=>i.fixed || i.resolved).length;
     const fixRate = validTotal > 0 ? Math.round(fixed / validTotal * 100 * 10) / 10 : 0;
     const resolveRate = validTotal > 0 ? Math.round(resolved / validTotal * 100 * 10) / 10 : 0;
     return {
@@ -3131,10 +3131,7 @@ function generateIssueKey(issue) {
         if (round > 1) {
             const prevIssues = allIssues.filter((i)=>i.round === round - 1);
             if (prevIssues.length > 0) {
-                const prevFixed = prevIssues.filter((i)=>i.fixed).length;
-                const prevResolved = prevIssues.filter((i)=>i.resolved && !i.fixed).length;
-                const prevInvalid = prevIssues.filter((i)=>i.valid === "false" && !i.fixed && !i.resolved).length;
-                const prevPending = prevIssues.length - prevFixed - prevResolved - prevInvalid;
+                const { fixed: prevFixed, resolved: prevResolved, invalid: prevInvalid, pending: prevPending } = calculateIssueStats(prevIssues);
                 parts.push("");
                 parts.push(`<details><summary>📊 Round ${round - 1} 回顾 (${prevIssues.length} 个问题)</summary>\n`);
                 parts.push(`| 状态 | 数量 |`);
@@ -3153,7 +3150,6 @@ function generateIssueKey(issue) {
 ;// CONCATENATED MODULE: ./src/review.config.ts
 
 /** LLM 模式 schema（与 core 中的 LLMMode 保持一致） */ const llmModeSchema = z["enum"]([
-    "claude-code",
     "openai",
     "gemini",
     "open-code"
@@ -3308,7 +3304,7 @@ function generateIssueKey(issue) {
  * 支持的格式：标题末尾 [/review -l openai -v 2]
  *
  * 支持的参数：
- * - `-l, --llm-mode <mode>`: LLM 模式 (claude-code, openai, gemini)
+ * - `-l, --llm-mode <mode>`: LLM 模式 (openai, gemini, open-code)
  * - `-v, --verbose [level]`: 详细输出级别 (1 或 2)
  * - `-d, --dry-run`: 仅打印不执行
  * - `-i, --includes <pattern>`: 文件过滤模式
@@ -3438,15 +3434,15 @@ function generateIssueKey(issue) {
 }
 function isValidLLMMode(value) {
     return [
-        "claude-code",
         "openai",
-        "gemini"
+        "gemini",
+        "open-code"
     ].includes(value);
 }
 function isValidDeletionAnalysisMode(value) {
     return [
         "openai",
-        "claude-code"
+        "open-code"
     ].includes(value);
 }
 function isValidAnalyzeDeletionsMode(value) {
@@ -5711,7 +5707,6 @@ class ReviewLlmProcessor {
 
 
 
-
 class ReviewService {
     gitProvider;
     config;
@@ -6191,28 +6186,6 @@ class ReviewService {
         }
         return result;
     }
-    /**
-   * 确保 Claude CLI 已安装
-   */ async ensureClaudeCli(ci) {
-        try {
-            execSync("claude --version", {
-                stdio: "ignore"
-            });
-        } catch  {
-            if (ci) {
-                throw new Error("Claude CLI 未安装。CI 环境请在 workflow 中预装: npm install -g @anthropic-ai/claude-code");
-            }
-            console.log("🔧 Claude CLI 未安装，正在安装...");
-            try {
-                execSync("npm install -g @anthropic-ai/claude-code", {
-                    stdio: "inherit"
-                });
-                console.log("✅ Claude CLI 安装完成");
-            } catch (installError) {
-                throw new Error(`Claude CLI 安装失败: ${installError instanceof Error ? installError.message : String(installError)}`);
-            }
-        }
-    }
 }
 
 ;// CONCATENATED MODULE: ./src/issue-verify.service.ts
@@ -6505,11 +6478,8 @@ class DeletionImpactService {
         // 3. 根据分析模式选择不同的分析方法
         const analysisMode = context.analysisMode ?? "openai";
         let result;
-        if ([
-            "claude-code",
-            "open-code"
-        ].includes(analysisMode)) {
-            // Claude Agent 模式：使用工具主动探索代码库
+        if (analysisMode === "open-code") {
+            // Agent 模式：使用工具主动探索代码库
             result = await this.analyzeWithAgent(analysisMode, deletedBlocks, references, verbose);
         } else {
             // OpenAI 模式：标准 chat completion
@@ -6887,8 +6857,8 @@ class DeletionImpactService {
         }
     }
     /**
-   * 使用 Claude Agent 模式分析删除代码的影响
-   * Claude Agent 可以使用工具主动探索代码库，分析更深入
+   * 使用 Agent 模式分析删除代码的影响
+   * Agent 可以使用工具主动探索代码库，分析更深入
    */ async analyzeWithAgent(analysisMode, deletedBlocks, references, verbose) {
         const llmJsonPut = new LlmJsonPut(DELETION_IMPACT_SCHEMA);
         const { systemPrompt, userPrompt } = buildDeletionImpactAgentPrompt({
@@ -7110,7 +7080,6 @@ const getRulesFromDirInputSchema = z.object({
     // 忽略配置读取错误
     }
     const defaultDirs = [
-        join(cwd, ".claude", "skills"),
         join(cwd, ".cursor", "skills"),
         join(cwd, "review-specs")
     ];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@spaceflow/review",
-  "version": "5.0.0",
+  "version": "5.0.2",
   "description": "Spaceflow 代码审查插件，使用 LLM 对 PR 代码进行自动审查",
   "license": "MIT",
   "author": "Lydanne",
@@ -25,10 +25,10 @@
     "@vitest/coverage-v8": "^4.0.18",
     "unplugin-swc": "^1.5.9",
     "vitest": "^4.0.18",
-    "@spaceflow/cli": "5.0.0"
+    "@spaceflow/cli": "5.0.1"
   },
   "peerDependencies": {
-    "@spaceflow/core": "5.0.0"
+    "@spaceflow/core": "5.0.1"
   },
   "spaceflow": {
     "type": "flow",

package/src/README.md

@@ -23,7 +23,7 @@ review/src/
 review extension
 ├── GitProviderService     # Git 平台适配器
 ├── GitSdkService          # Git 命令封装
-├── LlmProxyService        # LLM 统一代理（OpenAI / Claude）
+├── LlmProxyService        # LLM 统一代理（OpenAI / OpenCode）
 ├── LoggerService          # 日志系统
 ├── StorageService         # 缓存存储
 └── ParallelService        # 并行执行
@@ -44,7 +44,7 @@ CLI 命令入口，支持以下参数：
 | `--head <ref>`                    |      | 目标分支/tag                          |
 | `--verbose [level]`               | `-v` | 详细输出 (1: 过程日志, 2: 含提示词)   |
 | `--includes <patterns...>`        | `-i` | 文件 glob 过滤模式                    |
-| `--llm-mode <mode>`               | `-l` | LLM 模式: claude-code, openai, gemini |
+| `--llm-mode <mode>`               | `-l` | LLM 模式: openai, gemini, open-code |
 | `--files <files...>`              | `-f` | 仅审查指定文件                        |
 | `--commits <commits...>`          |      | 仅审查指定 commits                    |
 | `--verify-fixes`                  |      | 验证历史问题是否已修复                |
@@ -52,7 +52,7 @@ CLI 命令入口，支持以下参数：
 | `--verify-concurrency <n>`        |      | 验证并发数（默认 10）                 |
 | `--analyze-deletions`             |      | 分析删除代码影响                      |
 | `--deletion-only`                 |      | 仅执行删除代码分析                    |
-| `--deletion-analysis-mode <mode>` |      | 删除分析模式: openai, claude-code     |
+| `--deletion-analysis-mode <mode>` |      | 删除分析模式: openai, open-code     |
 | `--output-format <format>`        | `-o` | 输出格式: markdown, terminal, json    |
 | `--generate-description`          |      | 使用 AI 生成 PR 功能描述              |
 
@@ -176,7 +176,7 @@ analyzeDeletionImpact()
 │ 2. 查找代码引用关系 (git grep)       │
 │ 3. 调用 LLM 分析影响                 │
 │    - OpenAI 模式：标准 chat          │
-│    - Claude Agent 模式：可用工具     │
+│    - OpenCode 模式：可用工具        │
 └─────────────────────────────────────┘
 ```
 

package/src/deletion-impact.service.spec.ts

@@ -3,10 +3,6 @@ import { DeletionImpactService } from "./deletion-impact.service";
 import * as child_process from "child_process";
 import { EventEmitter } from "events";
 
-vi.mock("@anthropic-ai/claude-agent-sdk", () => ({
-  query: vi.fn(),
-}));
-
 vi.mock("child_process");
 
 describe("DeletionImpactService", () => {
@@ -520,7 +516,7 @@ diff --git a/file2.ts b/file2.ts
 
       const blocks = [{ file: "test.ts", startLine: 1, endLine: 1, content: "const x = 1;" }];
       const refs = new Map<string, string[]>();
-      const result = await (service as any).analyzeWithAgent("claude-code", blocks, refs);
+      const result = await (service as any).analyzeWithAgent("open-code", blocks, refs);
       expect(result.summary).toBe("agent ok");
     });
 
@@ -533,7 +529,7 @@ diff --git a/file2.ts b/file2.ts
       const consoleSpy = vi.spyOn(console, "error").mockImplementation(() => {});
       const blocks = [{ file: "test.ts", startLine: 1, endLine: 1, content: "const x = 1;" }];
       const refs = new Map<string, string[]>();
-      const result = await (service as any).analyzeWithAgent("claude-code", blocks, refs);
+      const result = await (service as any).analyzeWithAgent("open-code", blocks, refs);
       expect(result.impacts).toHaveLength(0);
       consoleSpy.mockRestore();
     });
@@ -546,7 +542,7 @@ diff --git a/file2.ts b/file2.ts
       const consoleSpy = vi.spyOn(console, "error").mockImplementation(() => {});
       const blocks = [{ file: "test.ts", startLine: 1, endLine: 1, content: "const x = 1;" }];
       const refs = new Map<string, string[]>();
-      const result = await (service as any).analyzeWithAgent("claude-code", blocks, refs);
+      const result = await (service as any).analyzeWithAgent("open-code", blocks, refs);
       expect(result.summary).toBe("Agent 调用失败");
       consoleSpy.mockRestore();
     });
@@ -559,7 +555,7 @@ diff --git a/file2.ts b/file2.ts
       const consoleSpy = vi.spyOn(console, "error").mockImplementation(() => {});
       const blocks = [{ file: "test.ts", startLine: 1, endLine: 1, content: "const x = 1;" }];
       const refs = new Map<string, string[]>();
-      const result = await (service as any).analyzeWithAgent("claude-code", blocks, refs);
+      const result = await (service as any).analyzeWithAgent("open-code", blocks, refs);
       expect(result.summary).toBe("Agent 调用失败");
       consoleSpy.mockRestore();
     });
@@ -572,7 +568,7 @@ diff --git a/file2.ts b/file2.ts
 
       const blocks = [{ file: "test.ts", startLine: 1, endLine: 1, content: "const x = 1;" }];
       const refs = new Map<string, string[]>();
-      const result = await (service as any).analyzeWithAgent("claude-code", blocks, refs);
+      const result = await (service as any).analyzeWithAgent("open-code", blocks, refs);
       expect(result.summary).toBe("分析返回格式无效");
     });
 
@@ -587,7 +583,7 @@ diff --git a/file2.ts b/file2.ts
 
       const blocks = [{ file: "test.ts", startLine: 1, endLine: 1, content: "const x = 1;" }];
       const refs = new Map<string, string[]>();
-      const result = await (service as any).analyzeWithAgent("claude-code", blocks, refs, 2);
+      const result = await (service as any).analyzeWithAgent("open-code", blocks, refs, 2);
       expect(result.summary).toBe("ok");
     });
   });
@@ -898,8 +894,8 @@ diff --git a/file2.ts b/file2.ts
   });
 
   describe("analyzeDeletionImpact - agent mode", () => {
-    it("should use agent mode for claude-code", async () => {
-      const context = { owner: "o", repo: "r", prNumber: 1, analysisMode: "claude-code" as const };
+    it("should use agent mode for open-code", async () => {
+      const context = { owner: "o", repo: "r", prNumber: 1, analysisMode: "open-code" as const };
       gitProvider.getPullRequestFiles.mockResolvedValue([
         {
           filename: "test.ts",

package/src/deletion-impact.service.ts

@@ -37,7 +37,7 @@ export interface DeletionAnalysisContext {
   headRef?: string;
   /** diff 来源：provider-api 使用 Git Provider API，git-diff 使用本地 git 命令（两点语法） */
   diffSource?: DeletionDiffSource;
-  /** 分析模式：openai 使用标准模式，claude-agent 使用 Agent 模式（可使用工具） */
+  /** 分析模式：openai 使用标准模式，open-code 使用 Agent 模式（可使用工具） */
   analysisMode?: LLMMode;
   /** 文件过滤 glob 模式，与 review.includes 一致 */
   includes?: string[];
@@ -156,8 +156,8 @@ export class DeletionImpactService {
     const analysisMode = context.analysisMode ?? "openai";
     let result: DeletionImpactResult;
 
-    if (["claude-code", "open-code"].includes(analysisMode)) {
-      // Claude Agent 模式：使用工具主动探索代码库
+    if (analysisMode === "open-code") {
+      // Agent 模式：使用工具主动探索代码库
       result = await this.analyzeWithAgent(analysisMode, deletedBlocks, references, verbose);
     } else {
       // OpenAI 模式：标准 chat completion
@@ -584,8 +584,8 @@ export class DeletionImpactService {
   }
 
   /**
-   * 使用 Claude Agent 模式分析删除代码的影响
-   * Claude Agent 可以使用工具主动探索代码库，分析更深入
+   * 使用 Agent 模式分析删除代码的影响
+   * Agent 可以使用工具主动探索代码库，分析更深入
    */
   protected async analyzeWithAgent(
     analysisMode: LLMMode,

package/src/issue-verify.service.spec.ts

@@ -2,10 +2,6 @@ import { vi } from "vitest";
 import { ReviewIssue, FileContentsMap } from "./review-spec";
 import { IssueVerifyService } from "./issue-verify.service";
 
-vi.mock("@anthropic-ai/claude-agent-sdk", () => ({
-  query: vi.fn(),
-}));
-
 describe("IssueVerifyService", () => {
   let service: IssueVerifyService;
   let llmProxyService: any;

package/src/locales/en/review.json

@@ -5,14 +5,14 @@
   "options.base": "Base branch/tag for diff comparison",
   "options.head": "Head branch/tag for diff comparison",
   "options.includes": "File glob patterns to review, e.g. *.ts *.js (can be specified multiple times). Supports change-type prefixes: added|*.ts (new files only), modified|*.ts (modified only), deleted|*.ts (deleted only)",
-  "options.llmMode": "LLM mode: claude-code, openai, gemini",
+  "options.llmMode": "LLM mode: openai, gemini, open-code",
   "options.files": "Only review specified files (space-separated)",
   "options.commits": "Only review specified commits (space-separated)",
   "options.verifyFixes": "Verify if historical issues are fixed (default from config)",
   "options.noVerifyFixes": "Disable historical issue verification",
   "options.verifyConcurrency": "Concurrency for verifying historical issues (default 10)",
   "options.analyzeDeletions": "Analyze impact of deleted code (true, false, ci, pr, terminal)",
-  "options.deletionAnalysisMode": "Deletion analysis mode: openai (standard) or claude-code (Agent mode with tools)",
+  "options.deletionAnalysisMode": "Deletion analysis mode: openai (standard) or open-code (Agent mode with tools)",
   "options.deletionOnly": "Only run deletion analysis, skip regular code review",
   "options.outputFormat": "Output format: markdown, terminal, json. Auto-selected if not specified (markdown for PR, terminal for CLI)",
   "options.generateDescription": "Generate PR description using AI",

package/src/locales/zh-cn/review.json

@@ -5,14 +5,14 @@
   "options.base": "基准分支/tag，用于比较差异",
   "options.head": "目标分支/tag，用于比较差异",
   "options.includes": "要审查的文件 glob 模式，如 *.ts *.js（可多次指定）。支持变更类型前缀: added|*.ts（仅新增）、modified|*.ts（仅修改）、deleted|*.ts（仅删除）",
-  "options.llmMode": "使用的 LLM 模式: claude-code, openai, gemini",
+  "options.llmMode": "使用的 LLM 模式: openai, gemini, open-code",
   "options.files": "仅审查指定的文件（空格分隔）",
   "options.commits": "仅审查指定的 commits（空格分隔）",
   "options.verifyFixes": "是否验证历史问题是否已修复（默认从配置文件读取）",
   "options.noVerifyFixes": "禁用历史问题验证",
   "options.verifyConcurrency": "验证历史问题的并发数（默认 10）",
   "options.analyzeDeletions": "分析删除代码可能带来的影响 (true, false, ci, pr, terminal)",
-  "options.deletionAnalysisMode": "删除代码分析模式: openai (标准模式) 或 claude-code (Agent 模式，可使用工具)",
+  "options.deletionAnalysisMode": "删除代码分析模式: openai (标准模式) 或 open-code (Agent 模式，可使用工具)",
   "options.deletionOnly": "仅执行删除代码分析，跳过常规代码审查",
   "options.outputFormat": "输出格式: markdown, terminal, json。不指定则智能选择（PR 用 markdown，终端用 terminal）",
   "options.generateDescription": "使用 AI 生成 PR 功能描述",

package/src/mcp/index.ts

@@ -50,11 +50,7 @@ async function getSpecDirs(cwd: string, ctx: SpaceflowContext): Promise<string[]
   } catch {
     // 忽略配置读取错误
   }
-  const defaultDirs = [
-    join(cwd, ".claude", "skills"),
-    join(cwd, ".cursor", "skills"),
-    join(cwd, "review-specs"),
-  ];
+  const defaultDirs = [join(cwd, ".cursor", "skills"), join(cwd, "review-specs")];
   for (const dir of defaultDirs) {
     if (existsSync(dir)) {
       dirs.push(dir);

package/src/parse-title-options.spec.ts

@@ -35,10 +35,10 @@ describe("parseTitleOptions", () => {
 
   describe("LLM 模式参数", () => {
     it("应该解析 -l 短参数", () => {
-      const title = "fix: bug [/ai-review -l claude-code]";
+      const title = "fix: bug [/ai-review -l openai]";
       const options = parseTitleOptions(title);
 
-      expect(options.llmMode).toBe("claude-code");
+      expect(options.llmMode).toBe("openai");
     });
 
     it("应该解析 --llm-mode 长参数", () => {
@@ -193,10 +193,10 @@ describe("parseTitleOptions", () => {
     });
 
     it("应该解析 --deletion-analysis-mode", () => {
-      const title = "fix: bug [/ai-review --deletion-analysis-mode claude-code]";
+      const title = "fix: bug [/ai-review --deletion-analysis-mode open-code]";
       const options = parseTitleOptions(title);
 
-      expect(options.deletionAnalysisMode).toBe("claude-code");
+      expect(options.deletionAnalysisMode).toBe("open-code");
     });
   });
 

package/src/parse-title-options.ts

@@ -22,7 +22,7 @@ export interface TitleOptions {
  * 支持的格式：标题末尾 [/review -l openai -v 2]
  *
  * 支持的参数：
- * - `-l, --llm-mode <mode>`: LLM 模式 (claude-code, openai, gemini)
+ * - `-l, --llm-mode <mode>`: LLM 模式 (openai, gemini, open-code)
  * - `-v, --verbose [level]`: 详细输出级别 (1 或 2)
  * - `-d, --dry-run`: 仅打印不执行
  * - `-i, --includes <pattern>`: 文件过滤模式
@@ -167,11 +167,11 @@ function parseArgs(argsString: string): string[] {
 }
 
 function isValidLLMMode(value: string): value is LLMMode {
-  return ["claude-code", "openai", "gemini"].includes(value);
+  return ["openai", "gemini", "open-code"].includes(value);
 }
 
 function isValidDeletionAnalysisMode(value: string): value is LLMMode {
-  return ["openai", "claude-code"].includes(value);
+  return ["openai", "open-code"].includes(value);
 }
 
 function isValidAnalyzeDeletionsMode(value: string): boolean {

package/src/review-context.ts

@@ -37,7 +37,7 @@ export interface ReviewContext extends ReviewOptions {
   retryDelay?: number;
   /** 仅执行删除代码分析，跳过常规代码审查 */
   deletionOnly?: boolean;
-  /** 删除代码分析模式：openai 使用标准模式，claude-agent 使用 Agent 模式 */
+  /** 删除代码分析模式：openai 使用标准模式，open-code 使用 Agent 模式 */
   deletionAnalysisMode?: LLMMode;
   /** 输出格式：markdown, terminal, json。不指定则智能选择 */
   outputFormat?: ReportFormat;

package/src/review-llm.spec.ts

@@ -4,9 +4,6 @@ import type { ReviewPrompt } from "./review.service";
 import { ChangedFileCollection } from "./changed-file-collection";
 
 vi.mock("c12");
-vi.mock("@anthropic-ai/claude-agent-sdk", () => ({
-  query: vi.fn(),
-}));
 vi.mock("@opencode-ai/sdk", () => ({
   createOpencodeClient: vi.fn().mockReturnValue({
     session: {
@@ -72,7 +69,7 @@ describe("ReviewLlmProcessor", () => {
       chat: vi.fn(),
       chatStream: vi.fn(),
       createSession: vi.fn(),
-      getAvailableAdapters: vi.fn().mockReturnValue(["claude-code", "openai"]),
+      getAvailableAdapters: vi.fn().mockReturnValue(["openai", "open-code"]),
     };
 
     processor = new ReviewLlmProcessor(mockLlmProxyService as any, mockReviewSpecService as any);
@@ -83,20 +80,6 @@ describe("ReviewLlmProcessor", () => {
   });
 
   describe("runLLMReview", () => {
-    it("should call callLLM when llmMode is claude", async () => {
-      const callLLMSpy = vi
-        .spyOn(processor, "callLLM")
-        .mockResolvedValue({ issues: [], summary: [] } as any);
-
-      const mockPrompt: ReviewPrompt = {
-        filePrompts: [{ filename: "test.ts", systemPrompt: "system", userPrompt: "user" }],
-      };
-
-      await processor.runLLMReview("claude-code", mockPrompt);
-
-      expect(callLLMSpy).toHaveBeenCalledWith("claude-code", mockPrompt, {});
-    });
-
     it("should call callLLM when llmMode is openai", async () => {
       const callLLMSpy = vi
         .spyOn(processor, "callLLM")

package/src/review-result-model.ts

@@ -1069,12 +1069,12 @@ export class ReviewResultModel {
     if (round > 1) {
       const prevIssues = allIssues.filter((i) => i.round === round - 1);
       if (prevIssues.length > 0) {
-        const prevFixed = prevIssues.filter((i) => i.fixed).length;
-        const prevResolved = prevIssues.filter((i) => i.resolved && !i.fixed).length;
-        const prevInvalid = prevIssues.filter(
-          (i) => i.valid === "false" && !i.fixed && !i.resolved,
-        ).length;
-        const prevPending = prevIssues.length - prevFixed - prevResolved - prevInvalid;
+        const {
+          fixed: prevFixed,
+          resolved: prevResolved,
+          invalid: prevInvalid,
+          pending: prevPending,
+        } = calculateIssueStats(prevIssues);
         parts.push("");
         parts.push(
           `<details><summary>📊 Round ${round - 1} 回顾 (${prevIssues.length} 个问题)</summary>\n`,

package/src/review.config.ts

@@ -3,7 +3,7 @@ import type { LLMMode, VerboseLevel, LocalReviewMode } from "@spaceflow/core";
 import type { ReportFormat } from "./review-report";
 
 /** LLM 模式 schema（与 core 中的 LLMMode 保持一致） */
-const llmModeSchema = z.enum(["claude-code", "openai", "gemini", "open-code"]);
+const llmModeSchema = z.enum(["openai", "gemini", "open-code"]);
 
 /** 删除代码分析模式 schema */
 const analyzeDeletionsModeSchema = z.union([z.boolean(), z.enum(["ci", "pr", "terminal"])]);
@@ -69,7 +69,7 @@ export interface ReviewOptions {
   analyzeDeletions?: AnalyzeDeletionsMode;
   /** 仅执行删除代码分析，跳过常规代码审查 */
   deletionOnly?: boolean;
-  /** 删除代码分析模式：openai 使用标准模式，claude-agent 使用 Agent 模式 */
+  /** 删除代码分析模式：openai 使用标准模式，open-code 使用 Agent 模式 */
   deletionAnalysisMode?: LLMMode;
   /** 输出格式：markdown, terminal, json。不指定则智能选择 */
   outputFormat?: ReportFormat;