sun-praise · Svtter · May 14, 2026 · May 14, 2026 · May 14, 2026 · May 14, 2026
diff --git a/README.md b/README.md
@@ -1,5 +1,7 @@
 # opencode-actions
 
+[中文文档](#中文文档)
+
 Reusable GitHub Actions for installing and running OpenCode in other repositories.
 
 This repository is licensed under Apache 2.0.
@@ -20,7 +22,7 @@ Write this in your CI.yaml
     opencode-go-api-key: ${{ secrets.OPENCODE_GO_API_KEY }}
 ```
 
-You'll get a automatic reviewer in Chinese.
+You'll get an automatic reviewer (Chinese by default, configurable via `language` input).
 
 Or add the skills to your project:
 
@@ -64,6 +66,7 @@ Use this when you want the shortest consumer workflow for `opencode github run`.
 | `reasoning-effort` | `max` | Reasoning effort level for the model agent (`low`, `medium`, `high`, `max`) |
 | `enable-thinking` | `true` | Enable thinking mode for the model agent |
 | `use-github-token` | `true` | Exported as `USE_GITHUB_TOKEN` before `opencode github run` |
+| `language` | `zh` | Response language: `zh` for Chinese, `en` for English |
 | `attempts` | `3` | Total attempts before failing |
 | `retry-profile` | `github-network` | Built-in retry preset for common GitHub failures |
 | `timeout-seconds` | `600` | Maximum execution time for `opencode github run`; `0` disables it |
@@ -108,7 +111,7 @@ Use this alongside `review` and `feature-missing` to audit whether a PR implemen
 Unlike `feature-missing` (which checks PR self-described scope), `spec-coverage` uses the project's own spec files as the **authoritative source of intended scope**. This catches cases where a PR implements part of a larger planned feature but skips critical integration steps.
 
 - auto-discovers spec files in `openspec/changes/*/tasks.md`, `specs/**`, and other common locations
-- intelligently skips bug fixes and minor changes that don't need specs (`无需审计`)
+- intelligently skips bug fixes and minor changes that don't need specs (`SKIP`)
 - reports missing spec files as a CRITICAL gap when a feature PR should have one but doesn't
 - cross-references unchecked task items against the PR diff
 - checks end-to-end integration (models read at runtime, configs consumed, APIs called)
@@ -199,6 +202,26 @@ uses: Svtter/opencode-actions/run-opencode@v2
     opencode-go-api-key: ${{ secrets.OPENCODE_GO_API_KEY }}
 ```
 
+To use English output, set the `language` input:
+
+```yaml
+- name: Run OpenCode review (English)
+  uses: Svtter/opencode-actions/review@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+    language: en
+```
+
+Or configure it from repository variables:
+
+```yaml
+- name: Run OpenCode review
+  uses: Svtter/opencode-actions/review@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+    language: ${{ vars.OPENCODE_LANGUAGE }}
+```
+
 More examples live in `examples/`.
 
 If you need more control, you can still use `setup-opencode` and `run-opencode` directly. For example, pass `opencode-path` explicitly when reusing a binary from another job or a custom location.
@@ -243,3 +266,187 @@ The initial release-notes template lives at `docs/releases/v1.0.0.md`.
 - `args` is parsed as a space-delimited string, so keep complex text in environment variables like `PROMPT`
 - the installer is external, so cache invalidation uses `cache-key` instead of a guaranteed version pin
 - first version is optimized for GitHub-hosted or Linux self-hosted runners
+
+---
+
+# 中文文档
+
+[English Documentation](#opencode-actions)
+
+可复用的 GitHub Actions，用于在其他仓库中安装和运行 OpenCode。
+
+本仓库基于 Apache 2.0 许可证。
+
+## 快速开始
+
+在你的 CI.yaml 中写入：
+
+```yaml
+- name: Run OpenCode review
+  uses: Svtter/opencode-actions/review@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+
+    # 以下密钥只需提供一个即可
+    deepseek-api-key: ${{ secrets.DEEPSEEK_API_KEY }}
+    zhipu-api-key: ${{ secrets.ZHIPU_API_KEY }}
+    opencode-go-api-key: ${{ secrets.OPENCODE_GO_API_KEY }}
+```
+
+你将获得一个自动化的 PR 审查机器人（默认中文回复，可通过 `language` 输入切换语言）。
+
+或将技能添加到你的项目：
+
+```bash
+npx skills add sun-praise/opencode-actions
+```
+
+## 包含的 Actions
+
+- `review`：内置 prompt 和模型默认值的 PR 审查封装
+- `feature-missing`：对照关联 issue 的需求，审计 PR 实现的完整性
+- `spec-coverage`：对照项目规格/任务文件，交叉检查 PR 实现的覆盖度
+- `github-run-opencode`：`opencode github run` 常用工作流的一步封装
+- `setup-opencode`：安装 OpenCode，恢复缓存，导出二进制路径
+- `run-opencode`：运行 `opencode`，支持 GitHub 网络抖动的重试逻辑
+
+## 语言配置
+
+所有 action 均支持 `language` 输入参数，用于控制 agent 的回复语言：
+
+- `zh`（默认）：中文回复
+- `en`：英文回复
+
+```yaml
+- name: Run OpenCode review (英文)
+  uses: Svtter/opencode-actions/review@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+    language: en
+```
+
+也可以从仓库变量中读取：
+
+```yaml
+- name: Run OpenCode review
+  uses: Svtter/opencode-actions/review@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+    language: ${{ vars.OPENCODE_LANGUAGE }}
+```
+
+## 三个审查 Action 的区别
+
+| Action | 审查范围 | 发现的问题 |
+| --- | --- | --- |
+| `review` | PR diff | 代码质量、安全问题、Bug |
+| `feature-missing` | PR 标题/正文 + 关联 issue | PR 自述需求的完整性 |
+| `spec-coverage` | 项目规格/任务文件 | 计划范围 vs 实际实现的差距 |
+
+## review
+
+最简单的 PR 审查方案。
+
+- 内置审查 prompt 模板
+- 内置 `MODEL` 解析：显式 `model` 输入 > `MODEL_NAME` 环境变量 > `zhipuai-coding-plan/glm-5.1`
+- 内置 `timeout-seconds` 默认值：`600`（10 分钟）
+- 可选的 `fallback-models` 支持超时驱动的模型轮换
+
+## feature-missing
+
+配合 `review` 使用，审计 PR 是否完整实现了关联 issue 中的需求。
+
+- 自动通过 `gh pr view` 读取关联 issue 作为需求规格
+- 如无关联 issue，从 PR 标题和正文提取需求
+- 按严重程度分类：CRITICAL / MEDIUM / LOW
+
+```yaml
+- name: Run feature missing audit
+  uses: Svtter/opencode-actions/feature-missing@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+    zhipu-api-key: ${{ secrets.ZHIPU_API_KEY }}
+    opencode-go-api-key: ${{ secrets.OPENCODE_GO_API_KEY }}
+```
+
+## spec-coverage
+
+配合 `review` 和 `feature-missing` 使用，审计 PR 是否实现了项目规格/任务文件中的全部计划。
+
+与 `feature-missing`（检查 PR 自述范围）不同，`spec-coverage` 使用项目自身的规格文件作为**预期范围的权威来源**。可以发现 PR 实现了大型计划功能的一部分但跳过了关键集成步骤的情况。
+
+- 自动发现 `openspec/changes/*/tasks.md`、`specs/**` 等位置的规格文件
+- 智能跳过不需要规格的 bug 修复和小改动
+- 当功能型 PR 缺少规格文件时报告为 CRITICAL 级别的遗漏
+- 按严重程度分类：CRITICAL / MEDIUM / LOW
+
+```yaml
+- name: Run spec coverage audit
+  uses: Svtter/opencode-actions/spec-coverage@v2
+  with:
+    github-token: ${{ secrets.GITHUB_TOKEN }}
+    zhipu-api-key: ${{ secrets.ZHIPU_API_KEY }}
+    opencode-go-api-key: ${{ secrets.OPENCODE_GO_API_KEY }}
+```
+
+## github-run-opencode
+
+当你需要最简短的 `opencode github run` 工作流时使用。
+
+### 常用输入
+
+| 输入 | 默认值 | 描述 |
+| --- | --- | --- |
+| `model` | `MODEL_NAME`，否则 `zhipuai-coding-plan/glm-5.1` | 在 `opencode github run` 前导出为 `MODEL` |
+| `fallback-models` | 空 | 可选的有序回退模型列表，按可用 API 密钥过滤 |
+| `prompt` | 内置 PR 审查模板 | 在 `opencode github run` 前导出为 `PROMPT` |
+| `language` | `zh` | 回复语言：`zh` 中文，`en` 英文 |
+| `github-token` | 空 | 在 `opencode github run` 前导出为 `GITHUB_TOKEN` |
+| `zhipu-api-key` | 空 | 在 `opencode github run` 前导出为 `ZHIPU_API_KEY` |
+| `deepseek-api-key` | 空 | 在 `opencode github run` 前导出为 `DEEPSEEK_API_KEY` |
+| `opencode-go-api-key` | 空 | 在 `opencode github run` 前导出为 `OPENCODE_GO_API_KEY` |
+| `reasoning-effort` | `max` | 模型推理力度（`low`、`medium`、`high`、`max`） |
+| `enable-thinking` | `true` | 启用模型的 thinking 模式 |
+| `attempts` | `3` | 失败前的总尝试次数 |
+| `timeout-seconds` | `600` | `opencode github run` 的最大执行时间；`0` 禁用 |
+
+## setup-opencode
+
+### 输入
+
+| 输入 | 默认值 | 描述 |
+| --- | --- | --- |
+| `install-url` | `https://opencode.ai/install` | 引导安装 OpenCode 的安装器 URL |
+| `install-dir` | 自动计算 | `opencode` 安装目录 |
+| `xdg-cache-home` | 自动计算 | XDG 缓存目录 |
+| `cache` | `true` | 启用 `actions/cache` 缓存 |
+| `cache-key` | `v1` | 用于使安装器缓存失效的键后缀 |
+| `install-attempts` | `3` | 安装器总尝试次数 |
+| `allow-preinstalled` | `false` | 复用 PATH 上已有的可信 `opencode` |
+
+## 使用方式
+
+```yaml
+uses: Svtter/opencode-actions/review@v2
+uses: Svtter/opencode-actions/feature-missing@v2
+uses: Svtter/opencode-actions/spec-coverage@v2
+uses: Svtter/opencode-actions/github-run-opencode@v2
+uses: Svtter/opencode-actions/setup-opencode@v2
+uses: Svtter/opencode-actions/run-opencode@v2
+```
+
+更多示例请参见 `examples/` 目录。
+
+## 本地验证
+
+```bash
+bash tests/test.sh
+```
+
+## 发布清单
+
+1. 公开推送仓库。
+2. 确认 `main` 分支的 CI 通过。
+3. 创建带有 semver 标签（如 `v1.0.0`）的 GitHub Release。
+4. 确认 `Update Major Tag` 工作流将 `v1` 标签移至该 release。
+5. 使用 `owner/repo/review@v2` 进行审查，`owner/repo/feature-missing@v2` 进行 PR 范围审计，`owner/repo/spec-coverage@v2` 进行规格覆盖审计。
diff --git a/feature-missing/action.yml b/feature-missing/action.yml
@@ -92,26 +92,26 @@ inputs:
       Do not repeat issues from earlier revisions unless they are still reproducible in the current files.
       Verify any gap claim against the current code before listing it.
 
-      Please respond in Chinese. DO NOT modify any code, only provide analysis.
+      DO NOT modify any code, only provide analysis.
 
       Verdict rule (MUST follow):
       - After completing your analysis, set the first line to exactly one of:
-        - "无遗漏" if and only if ALL spec items are fully implemented (no gaps at all)
-        - "发现遗漏" if and only if at least one gap exists
-      - The first line MUST match your actual finding. NEVER write "发现遗漏" when no gaps exist.
+        - "无遗漏" / "NO GAPS" if and only if ALL spec items are fully implemented (no gaps at all)
+        - "发现遗漏" / "GAPS FOUND" if and only if at least one gap exists
+      - The first line MUST match your actual finding. NEVER write "发现遗漏" / "GAPS FOUND" when no gaps exist.
 
       Output format:
-      - First line: the verdict ("无遗漏" or "发现遗漏") — see verdict rule above
+      - First line: the verdict ("无遗漏" or "发现遗漏" for Chinese; "NO GAPS" or "GAPS FOUND" for English) — see verdict rule above
       - Then a short summary of your analysis
-      - If verdict is "发现遗漏", classify each gap by severity:
+      - If verdict is "发现遗漏" / "GAPS FOUND", classify each gap by severity:
         ### CRITICAL
         Core functionality described in the spec is missing or completely non-functional.
         ### MEDIUM
         Partial implementation, integration gaps, or important edge cases not handled.
         ### LOW
         Minor gaps, nice-to-have features, or documentation discrepancies.
         For each gap: describe what's missing, reference the relevant spec section, and suggest where the implementation should be added.
-      - If verdict is "无遗漏", briefly confirm all spec items are covered — do NOT include any severity sections.
+      - If verdict is "无遗漏" / "NO GAPS", briefly confirm all spec items are covered — do NOT include any severity sections.
   use-github-token:
     description: Value exported as USE_GITHUB_TOKEN before running `opencode github run`.
     required: false
@@ -140,6 +140,15 @@ inputs:
     description: Enable thinking mode for the model agent. When true, generates opencode.json with thinking.type=enabled.
     required: false
     default: "true"
+  extra-env:
+    description: >-
+      Extra environment variables to pass to opencode runtime.
+      Multi-line KEY=VALUE pairs, one per line. Empty lines and lines
+      starting with '#' are ignored. Note: these variables are set after
+      internal configuration and may override sensitive runtime settings
+      such as MODEL, GITHUB_TOKEN, or API keys.
+    required: false
+    default: ""
   fallback-models:
     description: Optional comma- or newline-delimited fallback models tried after timeout or timeout-like failures.
     required: false
@@ -152,6 +161,11 @@ inputs:
     description: Rotate to the next fallback model when output matches this regex.
     required: false
     default: timed out|timeout|deadline exceeded|context deadline exceeded|operation timed out|connection timed out
+  language:
+    description: >-
+      Response language for the agent. `zh` (default) for Chinese, `en` for English.
+    required: false
+    default: "zh"
 
 runs:
   using: composite
@@ -255,8 +269,10 @@ runs:
         GITHUB_RUN_OPENCODE_FALLBACK_ON_REGEX: ${{ inputs.fallback-on-regex }}
         GITHUB_RUN_OPENCODE_PERMISSION: >-
           {"edit":"deny","bash":{"git commit *":"deny","git push *":"deny","git add *":"deny","git stash *":"deny","git reset *":"deny","git checkout *":"deny"}}
+        GITHUB_RUN_OPENCODE_LANGUAGE: ${{ inputs.language }}
+        GITHUB_RUN_OPENCODE_EXTRA_ENV: ${{ inputs.extra-env }}
       run: |
-        if ! command -v python3 >\/dev/null 2>&1; then
+        if ! command -v python3 >/dev/null 2>&1; then
           printf 'python3 is required but not installed on this runner\n' >&2
           exit 1
         fi

diff --git a/github-run-opencode/action.yml b/github-run-opencode/action.yml
@@ -113,6 +113,12 @@ inputs:
       Example: '{"edit":"deny","bash":{"git commit *":"deny"}}'
     required: false
     default: ""
+  language:
+    description: >-
+      Response language for the agent. `zh` (default) for Chinese, `en` for English.
+      Appends a language override instruction to the prompt.
+    required: false
+    default: "zh"
 
 runs:
   using: composite
@@ -215,6 +221,7 @@ runs:
         GITHUB_RUN_OPENCODE_REASONING_EFFORT: ${{ inputs.reasoning-effort }}
         GITHUB_RUN_OPENCODE_ENABLE_THINKING: ${{ inputs.enable-thinking }}
         GITHUB_RUN_OPENCODE_PERMISSION: ${{ inputs.permission }}
+        GITHUB_RUN_OPENCODE_LANGUAGE: ${{ inputs.language }}
       run: |
         if ! command -v python3 >/dev/null 2>&1; then
           printf 'python3 is required but not installed on this runner\n' >&2