jnMetaCode · jnMetaCode · Sep 28, 2026 · Sep 28, 2026 · Sep 28, 2026 · Sep 28, 2026
diff --git a/.lychee.toml b/.lychee.toml
@@ -51,6 +51,8 @@ exclude = [
 
   # 对 CI 爬虫返回 403 的站点（人工访问正常）
   "^https?://platform\\.openai\\.com",
+  # 仅对中国大陆 IP 开放
+  "^https?://(www\\.)?trae\\.cn",
 
   # QQ 链接跳转
   "^https?://qm\\.qq\\.com",

diff --git a/aider/README.en.md b/aider/README.en.md
@@ -3,6 +3,8 @@
 # Aider Best Practices
 
 > Aider is an open-source CLI AI coding tool. Its defining feature is being **Git-native** — every change is auto-committed, with built-in version control support. It works with virtually all major LLMs (Claude, GPT, DeepSeek, Qwen, local models), making it the most flexible AI coding CLI out there.
+>
+> ⚠️ **Maintenance status (verified 2026-09)**: the latest PyPI release is still 0.86.2 (published 2026-02-12), and the last GitHub commit was around 2026-05 — development has clearly slowed. There is **no native MCP support**. Built-in model metadata lags behind new model releases, so newer models need a bit of manual config (see "New model config" below).
 
 ---
 
@@ -12,7 +14,7 @@
 |---------|-------------|----------|
 | **Chat Modes** | `code` / `ask` / `architect` | Different modes for different tasks |
 | **Auto Git Commits** | Every change auto-committed | Roll back anytime |
-| **Map Mode** | Auto-indexes project structure | Understand large codebases |
+| **Repo Map** | Extracts key symbols and their definition lines (signatures), graph-ranked by references | Understand large codebases |
 | **Multi-Model Support** | Claude/GPT/DeepSeek/Ollama etc. | Flexible selection, cost control |
 | **Lint & Test** | Built-in code checking and testing | Auto-verify after changes |
 
@@ -23,7 +25,9 @@
 ### Installation
 
 ```bash
-pip install aider-chat
+python -m pip install aider-install && aider-install
+# Or: uv tool install --force --python python3.12 --with pip aider-chat@latest
+# Or: pipx install aider-chat
 
 # Set an API key (pick one)
 export ANTHROPIC_API_KEY=sk-xxx    # Claude
@@ -37,16 +41,18 @@ export DEEPSEEK_API_KEY=sk-xxx     # DeepSeek
 cd /your/project
 aider
 
-# Specify a model
-aider --model claude-sonnet-4-5
+# Specify a model (pass the latest Claude model ID explicitly)
+aider --model anthropic/claude-sonnet-5-5
 
 # Use DeepSeek (cheaper)
-aider --model deepseek/deepseek-chat
+aider --model deepseek/deepseek-v4-pro
 
 # Use a local model (free)
-aider --model ollama/qwen2.5-coder
+aider --model ollama/qwen3-coder
 ```
 
+> The built-in aliases are outdated: `--model sonnet` → `claude-sonnet-4-6`, `--model opus` → `claude-opus-4-7`, and `--model deepseek` still points to `deepseek/deepseek-chat`, which was retired on 2026-07-24. Always spell out the full model ID.
+
 ### Three Chat Modes
 
 ```bash
@@ -108,23 +114,25 @@ git log --oneline  # View Aider's commit history
 git diff HEAD~1    # See the last change
 git revert HEAD    # Not happy? One-command rollback
 
-# Set a custom commit prefix
-aider --commit-prefix "[ai] "
+# Mark AI commits: a Co-authored-by trailer is added by default; tune it with the --attribute-* flags
+aider --attribute-commit-message-author   # Prefix commit messages for AI changes with "aider: "
+# Customize the commit message style
+aider --commit-prompt "Write commit messages in Conventional Commits format"
 ```
 
 ### 4. Multi-Model Strategy
 
 ```bash
 # Complex architecture design — use the strongest model
-aider --model claude-sonnet-4-5
+aider --model anthropic/claude-opus-5-5
 /architect Design a microservices split plan
 
 # Daily coding — use a cost-effective model
-aider --model deepseek/deepseek-chat
+aider --model deepseek/deepseek-v4-pro
 /code Implement user-service according to the plan
 
 # Code review — use a free local model
-aider --model ollama/qwen2.5-coder
+aider --model ollama/qwen3-coder
 /ask Any issues with this code?
 ```
 
@@ -136,14 +144,28 @@ aider --model ollama/qwen2.5-coder
 
 ```yaml
 # .aider.conf.yml
-model: claude-sonnet-4-5
+model: anthropic/claude-sonnet-5-5
 auto-commits: true
 auto-lint: true
 auto-test: true
 test-cmd: pytest
 lint-cmd: ruff check
 ```
 
+### New Model Config — `.aider.model.settings.yml`
+
+Aider sends a `temperature` parameter by default, but newer Claude models (Sonnet 5.5, Opus 5.5, etc.) reject non-default temperature values with a 400 error. Add this to your project root or home directory:
+
+```yaml
+# .aider.model.settings.yml
+- name: anthropic/claude-sonnet-5-5
+  edit_format: diff
+  use_repo_map: true
+  use_temperature: false
+```
+
+For models missing from Aider's built-in metadata, you'll get an "unknown context window" warning on startup. It's usually safe to ignore, or you can fill it in via `.aider.model.metadata.json`.
+
 ### Lint + Test Automation
 
 ```yaml
@@ -183,16 +205,18 @@ git push -u origin feature/add-notifications
 | Cost control | 3/3 (free models available) | 1/3 | 3/3 |
 | Best for | Flexibility, saving money, Git power users | Complex Agent tasks | Large codebase analysis |
 
+> Gemini CLI stopped serving individual/free users (including Google AI Pro/Ultra subscribers) on 2026-06-18. Enterprise licenses and paid API keys still work; the official migration path for individuals is Antigravity CLI. The "Cost control" row no longer applies to individual users.
+
 ---
 
 ## Common Pitfalls
 
 | Pitfall | Description | Solution |
 |---------|-------------|----------|
-| Auto-commit swallows WIP | Auto-commit sweeps your unstaged changes into the commit | `git stash` before session, or use a dedicated branch |
+| Auto-commit commits your edits too | If a file Aider is about to edit has your uncommitted changes, it first commits them separately (`--dirty-commits` is on by default) | Commit/`git stash` yourself before the session, use `--no-dirty-commits`, or work on a dedicated branch |
 | `/add` misses deps | Incomplete context, AI guesses from filenames | Have it `/ask` list dependencies first, then `/add` all of them |
 | Model-switch quality drop | Cheaper model → cliff drop in output quality | Switch by task type; return to Claude for complex work |
-| Lint loop burns tokens | `auto-lint` retries failing checks via LLM | `--max-reflections 3`, use `--fix` linters |
+| Lint loop burns tokens | `auto-lint` retries failing checks via LLM | Use `--fix` linters; if it still loops, turn it off with `--no-auto-lint` |
 
 👉 **Deep dive**: [Aider Pitfalls](../pitfalls/aider.en.md) — 7 real-world traps, each with Symptom / Cause / Recovery / Prevention
 
@@ -209,5 +233,5 @@ git push -u origin feature/add-notifications
 ## Further Reading
 
 - [Aider Official Docs](https://aider.chat/docs/)
-- [Aider GitHub](https://github.com/Aider-AI/aider) (42k+ stars)
+- [Aider GitHub](https://github.com/Aider-AI/aider) (49k+ stars)
 - [superpowers-zh](https://github.com/jnMetaCode/superpowers-zh) — Skills methodology (also supports Aider)
diff --git a/aider/README.md b/aider/README.md
@@ -1,6 +1,8 @@
 # Aider 最佳实践
 
 > Aider 是一个开源的 CLI AI 编程工具，核心特色是 **Git 原生** — 每次修改自动提交，天然支持版本控制。支持几乎所有主流 LLM（Claude、GPT、DeepSeek、Qwen、本地模型），是最灵活的 AI 编程 CLI。
+>
+> ⚠️ **维护状态（2026-09 核实）**：PyPI 最新版仍是 0.86.2（2026-02-12 发布），GitHub 最后一次提交在 2026-05 左右，更新明显放缓；目前**没有原生 MCP 支持**。内置模型元数据跟不上新模型发布，用新模型时要自己补配置（见下文「新模型配置」）。
 
 ---
 
@@ -10,7 +12,7 @@
 |------|------|------|
 | **Chat 模式** | `code` / `ask` / `architect` | 不同任务用不同模式 |
 | **自动 Git 提交** | 每次修改自动 commit | 随时可以回滚 |
-| **Map 模式** | 自动索引项目结构 | 理解大代码库 |
+| **Repo Map** | 提取关键符号和定义行（签名），按引用关系图排序 | 理解大代码库 |
 | **多模型支持** | Claude/GPT/DeepSeek/Ollama 等 | 灵活选择，控制成本 |
 | **Lint & Test** | 内置代码检查和测试 | 修改后自动验证 |
 
@@ -21,7 +23,9 @@
 ### 安装
 
 ```bash
-pip install aider-chat
+python -m pip install aider-install && aider-install
+# 也可以：uv tool install --force --python python3.12 --with pip aider-chat@latest
+# 或：pipx install aider-chat
 
 # 设置 API Key（选一个）
 export ANTHROPIC_API_KEY=sk-xxx    # Claude
@@ -35,16 +39,18 @@ export DEEPSEEK_API_KEY=sk-xxx     # DeepSeek
 cd /your/project
 aider
 
-# 指定模型
-aider --model claude-sonnet-4-5
+# 指定模型（用 --model 写明最新 Claude 模型 ID）
+aider --model anthropic/claude-sonnet-5-5
 
 # 用 DeepSeek（便宜）
-aider --model deepseek/deepseek-chat
+aider --model deepseek/deepseek-v4-pro
 
 # 用本地模型（免费）
-aider --model ollama/qwen2.5-coder
+aider --model ollama/qwen3-coder
 ```
 
+> 注意内置别名已经过时：`--model sonnet` → `claude-sonnet-4-6`，`--model opus` → `claude-opus-4-7`，`--model deepseek` 仍指向已于 2026-07-24 下线的 `deepseek/deepseek-chat`。建议始终写完整模型 ID。
+
 ### 三种 Chat 模式
 
 ```bash
@@ -106,23 +112,25 @@ git log --oneline  # 看 Aider 的提交记录
 git diff HEAD~1    # 看最后一次修改
 git revert HEAD    # 不满意？一键回滚
 
-# 设置自定义提交前缀
-aider --commit-prefix "[ai] "
+# 区分 AI 提交：默认已加 Co-authored-by trailer，可用 --attribute-* 系列开关调整
+aider --attribute-commit-message-author   # AI 改动的 commit message 加 "aider: " 前缀
+# 自定义 commit message 风格
+aider --commit-prompt "用中文写 Conventional Commits 格式的提交信息"
 ```
 
 ### 4. 多模型策略
 
 ```bash
 # 复杂架构设计 — 用最强模型
-aider --model claude-sonnet-4-5
+aider --model anthropic/claude-opus-5-5
 /architect 设计微服务拆分方案
 
 # 日常编码 — 用性价比模型
-aider --model deepseek/deepseek-chat
+aider --model deepseek/deepseek-v4-pro
 /code 按方案实现 user-service
 
 # 代码审查 — 用免费本地模型
-aider --model ollama/qwen2.5-coder
+aider --model ollama/qwen3-coder
 /ask 看看这段代码有没有问题
 ```
 
@@ -134,14 +142,28 @@ aider --model ollama/qwen2.5-coder
 
 ```yaml
 # .aider.conf.yml
-model: claude-sonnet-4-5
+model: anthropic/claude-sonnet-5-5
 auto-commits: true
 auto-lint: true
 auto-test: true
 test-cmd: pytest
 lint-cmd: ruff check
 ```
 
+### 新模型配置 — `.aider.model.settings.yml`
+
+Aider 默认会发 `temperature` 参数，而新一代 Claude（Sonnet 5.5、Opus 5.5 等）会拒绝非默认的 temperature，直接 400 报错。在项目根目录或 home 目录加：
+
+```yaml
+# .aider.model.settings.yml
+- name: anthropic/claude-sonnet-5-5
+  edit_format: diff
+  use_repo_map: true
+  use_temperature: false
+```
+
+Aider 内置元数据里没有的模型，启动时会提示未知上下文窗口，一般可以忽略，也可以在 `.aider.model.metadata.json` 里补上。
+
 ### Lint + Test 自动化
 
 ```yaml
@@ -181,16 +203,18 @@ git push -u origin feature/add-notifications
 | 成本控制 | ★★★（可用免费模型） | ★☆☆ | ★★★ |
 | 适合 | 灵活、省钱、Git 重度用户 | 复杂 Agent 任务 | 大代码库分析 |
 
+> Gemini CLI 已于 2026-06-18 停止为个人/免费用户（含 Google AI Pro/Ultra 订阅）提供服务，企业授权和付费 API Key 仍可用；个人用户官方迁移路径是 Antigravity CLI。表中「成本控制」一栏对个人用户已不适用。
+
 ---
 
 ## 常见陷阱
 
 | 陷阱 | 说明 | 解决 |
 |------|------|------|
-| auto-commit 吃变更 | 自动提交把你未 stage 的工作一起吞了 | 会话前 `git stash`，或开独立分支 |
+| auto-commit 顺手提交你的改动 | Aider 要改的文件里如果有你没提交的改动，它会先把这些改动单独 commit 一次（`--dirty-commits` 默认开） | 会话前自己先 commit/`git stash`，或 `--no-dirty-commits`，或开独立分支 |
 | /add 漏依赖 | 上下文不全，AI 靠文件名脑补 | 让它先 /ask 列出所有依赖再 /add |
 | 模型切换质量塌 | 切到便宜模型后产出质量断崖下跌 | 分任务类型切，复杂任务回 Claude |
-| lint 循环烧 token | auto-lint 失败反复让 LLM 修 | `--max-reflections 3`，lint 用 `--fix` 模式 |
+| lint 循环烧 token | auto-lint 失败反复让 LLM 修 | lint 用 `--fix` 模式；修不好就 `--no-auto-lint` 先关掉 |
 
 👉 **深度展开版**：[Aider 陷阱合集](../pitfalls/aider.md) — 7 个真实踩坑场景，每个带症状 / 根因 / 出坑 / 预防
 
@@ -207,5 +231,5 @@ git push -u origin feature/add-notifications
 ## 延伸阅读
 
 - [Aider 官方文档](https://aider.chat/docs/)
-- [Aider GitHub](https://github.com/Aider-AI/aider)（42k+ star）
+- [Aider GitHub](https://github.com/Aider-AI/aider)（49k+ star）
 - [superpowers-zh](https://github.com/jnMetaCode/superpowers-zh) — Skills 方法论（也支持 Aider）
diff --git a/aider/templates/aider.conf.yml b/aider/templates/aider.conf.yml
@@ -2,15 +2,17 @@
 # 复制到项目根目录，命名为 .aider.conf.yml
 
 # 模型配置
-model: claude-sonnet-4-5
+model: anthropic/claude-sonnet-5-5     # 新 Claude 模型需在 .aider.model.settings.yml 设 use_temperature: false
 # 其他选择：
-# model: deepseek/deepseek-chat      # 便宜，适合日常编码
-# model: ollama/qwen2.5-coder        # 免费，需要本地跑 Ollama
+# model: deepseek/deepseek-v4-pro      # 便宜，适合日常编码（别用 deepseek 别名，它仍指向已下线的 deepseek-chat）
+# model: ollama/qwen3-coder            # 免费，需要本地跑 Ollama
 # model: gpt-4o                       # OpenAI
 
 # Git 配置
-auto-commits: true                     # 每次修改自动 commit
-commit-prefix: "[ai] "                 # commit 前缀，方便区分人工和 AI 修改
+auto-commits: true                     # 每次修改自动 commit（只提交 Aider 改过的文件）
+dirty-commits: true                    # 要改的文件里有你未提交的改动时，先单独 commit 一次
+attribute-commit-message-author: true  # AI 改动的 commit message 加 "aider: " 前缀，方便区分人工和 AI 修改
+# commit-prompt: "用中文写 Conventional Commits 格式的提交信息"  # 自定义 commit message 生成提示
 
 # 代码质量
 auto-lint: true                        # 修改后自动跑 linter