fix(repo): handle missing model cache and localize templates

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -1,78 +1,77 @@
-name: Bug Report
-description: Report a confirmed defect or regression
-title: "bug: "
+name: 问题反馈
+description: 报告已确认的缺陷或回归问题
+title: "问题："
 labels:
   - bug
 body:
   - type: markdown
     attributes:
       value: |
-        Provide a reproducible report. Include exact commands, config, and observed behavior.
+        请尽量提供可复现的信息，包括命令、配置、实际表现和预期结果。
 
   - type: input
     id: summary
     attributes:
-      label: Summary
-      description: One-sentence description of the bug
-      placeholder: codex_workdir is ignored by /home and the directory browser Home action
+      label: 问题概述
+      description: 用一句话描述你遇到的问题
+      placeholder: codex_workdir 在 /home 和目录浏览器的 Home 按钮中没有生效
     validations:
       required: true
 
   - type: textarea
     id: environment
     attributes:
-      label: Environment
-      description: Python, PDM, NoneBot, adapter, OS, and any relevant Codex setup details
+      label: 运行环境
+      description: 填写 Python、PDM、NoneBot、适配器、操作系统以及相关 Codex 环境信息
       placeholder: |
-        Python:
-        PDM:
-        NoneBot:
-        Adapter:
-        OS:
-        Codex CLI:
+        Python 版本：
+        PDM 版本：
+        NoneBot 版本：
+        Adapter 版本：
+        操作系统：
+        Codex CLI：
     validations:
       required: true
 
   - type: textarea
     id: reproduce
     attributes:
-      label: Steps To Reproduce
-      description: Provide exact steps or commands
+      label: 复现步骤
+      description: 提供可以稳定复现问题的步骤或命令
       placeholder: |
-        1. Configure codex_workdir = "/tmp/work"
-        2. Start the bot
-        3. Run /home
-        4. Observe the reported workdir
+        1. 配置 codex_workdir = "/tmp/work"
+        2. 启动机器人
+        3. 执行 /home
+        4. 观察返回的工作目录
     validations:
       required: true
 
   - type: textarea
     id: expected
     attributes:
-      label: Expected Behavior
+      label: 预期行为
     validations:
       required: true
 
   - type: textarea
     id: actual
     attributes:
-      label: Actual Behavior
+      label: 实际行为
     validations:
       required: true
 
   - type: textarea
     id: evidence
     attributes:
-      label: Verification Evidence
-      description: Include failing tests, logs, screenshots, or command output summaries
+      label: 补充信息
+      description: 可填写失败测试、日志、截图或命令输出摘要
 
   - type: textarea
     id: affected
     attributes:
-      label: Affected Files Or Commands
+      label: 相关文件或命令
       placeholder: |
         src/nonebot_plugin_codex/service.py
         src/nonebot_plugin_codex/telegram.py
         /home
         /cd
-

.github/ISSUE_TEMPLATE/config.yml

@@ -1,6 +1,5 @@
 blank_issues_enabled: false
 contact_links:
-  - name: Repository Plans
+  - name: 仓库规划文档
     url: https://github.com/ttiee/nonebot-plugin-codex/tree/main/docs/plans
-    about: Review existing design and implementation planning documents before opening broad-scoped requests.
-
+    about: 在提交范围较大的请求前，请先查看现有设计文档和实现计划。

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -1,51 +1,50 @@
-name: Feature Request
-description: Propose an improvement or new capability
-title: "feat: "
+name: 功能请求
+description: 提议一个改进或新能力
+title: "功能："
 labels:
   - enhancement
 body:
   - type: markdown
     attributes:
       value: |
-        Describe the user problem first, then the proposed solution.
+        请先描述用户问题，再说明你希望的解决方案。
 
   - type: input
     id: summary
     attributes:
-      label: Summary
-      placeholder: Add button panels for mode/model/effort/permission commands
+      label: 功能概述
+      placeholder: 为 /mode、/model、/effort、/permission 增加按钮面板
     validations:
       required: true
 
   - type: textarea
     id: problem
     attributes:
-      label: Problem Statement
-      description: What is hard, missing, or error-prone today?
+      label: 问题背景
+      description: 当前使用上有什么困难、缺失或容易出错的地方？
     validations:
       required: true
 
   - type: textarea
     id: proposal
     attributes:
-      label: Proposed Solution
+      label: 建议方案
     validations:
       required: true
 
   - type: textarea
     id: alternatives
     attributes:
-      label: Alternatives Considered
+      label: 备选方案
 
   - type: textarea
     id: scope
     attributes:
-      label: Scope And Constraints
-      description: Compatibility, migration, UX, or testing constraints that matter
+      label: 范围与约束
+      description: 这里填写兼容性、迁移、交互或测试方面的约束
 
   - type: textarea
     id: verification
     attributes:
-      label: Verification Plan
-      description: Tests, manual checks, or rollout validation you expect
-
+      label: 验证计划
+      description: 希望如何验证这个功能，例如测试、手工检查或上线确认

.github/pull_request_template.md

@@ -1,21 +1,20 @@
-## Summary
+## 变更摘要
 
 - 
 
-## Linked Issue
+## 关联 Issue
 
 Closes #
 
-## What Changed
+## 具体改动
 
 - 
 
-## Verification
+## 验证
 
 - [ ] `pdm run pytest -q`
 - [ ] `pdm run ruff check .`
 
-## Risks / Follow-Ups
+## 风险 / 后续项
 
 - 
-

docs/plans/2026-03-12-model-cache-and-cn-templates-design.md

@@ -0,0 +1,99 @@
+# Model Cache Fallback And Chinese Templates Design
+
+## Background
+
+The repository still has one confirmed runtime defect: if the configured model cache file is missing, `get_preferences()` currently tries to build default preferences via `load_models()`, which makes non-model commands fail early. At the same time, the repository's new GitHub issue templates and PR template are still written in English, while the project's user-facing language and maintainer preference are Chinese.
+
+## Goals
+
+- Fix the remaining bug where non-model commands fail when the model cache file is missing.
+- Keep model-specific commands honest: they may still require model metadata, but generic commands should not.
+- Localize the GitHub issue templates into Chinese.
+- Localize the PR template into Chinese as well for consistency.
+
+## Non-Goals
+
+- Do not redesign the Telegram command flow again.
+- Do not add external configuration discovery beyond the current Codex config and existing settings.
+- Do not change issue labels or broader repo governance.
+
+## Root Cause
+
+`CodexBridgeService.get_preferences()` lazily creates default preferences through `_default_preferences()`. That method currently loads the model cache first, even though many call sites only need a summary of current settings or the current workdir. As a result, commands like `/pwd`, `/codex` without a prompt, and `/new` can fail with `FileNotFoundError` before the user reaches any model-specific operation.
+
+## Approach Options
+
+### Option 1: Catch FileNotFoundError in handlers
+
+Wrap more handlers in `FileNotFoundError` handling and keep the service logic unchanged.
+
+- Pros: very small code diff
+- Cons: symptom fix only; the service still couples generic preference creation to model metadata
+
+### Option 2: Decouple default preferences from model cache
+
+Make `_default_preferences()` prefer `~/.codex/config.toml` values first, then safe fallback defaults, and only consult the model cache when available for normalization.
+
+- Pros: fixes the root cause at the service layer, keeps generic commands usable, preserves existing behavior when metadata exists
+- Cons: needs careful tests so model-specific commands still behave intentionally
+
+### Option 3: Allow missing model in preferences
+
+Let preferences hold `model=None` and push the requirement down to execution-time paths only.
+
+- Pros: semantically explicit
+- Cons: much wider type and formatting churn than necessary
+
+## Recommended Design
+
+Use Option 2.
+
+### Service Behavior
+
+- `_default_preferences()` should stop hard-requiring `load_models()`.
+- It should:
+  - read `codex_config_path` first
+  - if the model cache is available, preserve the current normalization behavior
+  - if the model cache is missing or invalid, fall back to the configured model/effort from `config.toml`
+  - if even that is missing, use a small internal default such as `gpt-5` and `high`
+
+This keeps generic preference creation stable while avoiding a large `None`-propagation refactor.
+
+### Command Expectations
+
+- These should work even if the model cache file is absent:
+  - `/pwd`
+  - `/codex` without prompt
+  - `/new`
+- These may still require model metadata and can continue to error clearly if it is missing:
+  - `/models`
+  - `/model`
+  - `/effort`
+  - model and effort selector panels
+
+### Template Localization
+
+Translate the current GitHub-facing templates to Chinese:
+
+- `.github/ISSUE_TEMPLATE/bug_report.yml`
+- `.github/ISSUE_TEMPLATE/feature_request.yml`
+- `.github/ISSUE_TEMPLATE/config.yml`
+- `.github/pull_request_template.md`
+
+The structure should remain the same; only the repository-facing language and helper copy need localization.
+
+### Testing
+
+Add failing tests first for:
+
+- service default preferences without a model cache
+- handler paths that should stay usable without a model cache
+- presence of Chinese wording in the issue and PR templates
+
+## Risks And Mitigations
+
+- Risk: fallback defaults may diverge from an individual user's actual Codex defaults.
+  Mitigation: prefer `codex_config_path` values before any hardcoded fallback.
+
+- Risk: a broad fallback could accidentally hide model-metadata errors everywhere.
+  Mitigation: keep `load_models()`-dependent commands unchanged; only decouple generic preference creation.