Aider

Aider AI 编程助手，精通终端内 AI 编程、Git 集成、多模型支持、代码重构

Security Scan

The skill's description matches a terminal AI coding assistant, but the runtime instructions ask you to provide API keys and to send repo contents to third‑party models while the registry metadata declares no required credentials or install — this mismatch is concerning.

Purpose & Capability

The SKILL.md describes a terminal AI pair‑programming tool that edits local Git repos, runs linters/tests, and creates commits — that behavior is coherent with the skill name and description. However the registry metadata lists no required binaries or env vars even though the instructions expect Python, git, pytest/linters and model API keys; this absence is a mismatch worth noting.

Instruction Scope

Instructions explicitly direct reading and editing of local repository files, running arbitrary commands (e.g., pytest, linters) and sending data to external LLM providers (Anthropic, OpenAI or other configured models). That means repository source, test output, and diffs may be transmitted off‑host — appropriate for the tool but a privacy/exfiltration risk that should be documented and consented to.

Install Mechanism

The registry provides no install spec (lowest platform risk), but SKILL.md recommends installing 'aider-chat' via pip. This is a standard install path (PyPI) but the package is external and not vetted here — the registry not including an install spec is an inconsistency to be aware of.

Credentials

Registry declares no required environment variables or primary credential, but the instructions recommend exporting ANTHROPIC_API_KEY and OPENAI_API_KEY (and support various models). Requesting model API keys is reasonable for this tool, but failing to declare them in metadata is an incoherence that reduces transparency and prevents automated permission checks.

Persistence & Privilege

The skill is not marked always:true and does not request system‑wide configuration changes in the registry. Autonomous invocation is allowed (platform default) but there is no evidence the skill demands permanent elevated presence or modifies other skills' configs.

What to consider before installing

This skill appears to do what it claims (edit repos, run tests, commit changes) but there are important mismatches and privacy implications to consider before installing: - The SKILL.md tells you to export ANTHROPIC_API_KEY and OPENAI_API_KEY, but the skill metadata declares no required credentials. Treat that as a transparency gap — the skill will likely send repo contents to third‑party model providers. Only provide API keys you trust and consider using limited-scope keys or billing safeguards. - The registry has no install spec but the README suggests pip install aider-chat (an external PyPI package). Before installing, inspect that package's source (or its PyPI page / GitHub repo) to confirm it does what you expect. - Run the tool in an isolated environment (virtualenv, container) and on a non-sensitive repository first. Ensure .gitignore and repo configuration avoid committing secrets; review any auto‑commit diffs before pushing. - If privacy is a concern, prefer local models (the SKILL.md mentions Ollama/local) or disable remote model usage. - If you proceed, limit permissions: use API keys with minimal scope, monitor usage, and be prepared to revoke keys if unexpected behavior appears. If you want, I can list specific checks to perform on the 'aider-chat' package (files to review, what to search for) or suggest safer local alternatives.

Aider - 终端内 AI Pair Programming 工具

简介

Aider 是一款运行在终端中的 AI 结对编程工具，能直接编辑本地 Git 仓库中的代码。 它的核心理念是"对话即编程"——你用自然语言描述需求，Aider 自动修改代码并生成 Git commit。

核心功能

• 自动 Git Commit：每次代码修改自动生成语义化的 commit message
• 多文件编辑：一次对话中同时修改多个文件，保持跨文件一致性
• 文件管理：通过 /add 和 /drop 精确控制 AI 可见的上下文范围
• 代码地图（Repo Map）：基于 tree-sitter 构建仓库结构索引，帮助 AI 理解项目全貌
• Linting 集成：修改后自动运行 linter，发现问题自动修复

支持的模型

模型	推荐场景	配置方式
Claude Sonnet 4	日常编码（默认推荐）	--model sonnet
Claude Opus 4	复杂架构重构	--model opus
GPT-4o	OpenAI 生态用户	--model gpt-4o
DeepSeek V3	高性价比选择	--model deepseek
Ollama 本地模型	离线/隐私场景	--model ollama/qwen2.5-coder

常用命令

# 启动 Aider
aider                          # 当前目录启动
aider --model sonnet           # 指定模型
aider file1.py file2.py        # 预加载文件

# 会话内命令
/add src/main.py               # 添加文件到上下文
/drop src/test.py              # 从上下文移除文件
/undo                          # 撤销上一次 AI 修改（git undo）
/diff                          # 查看当前修改的 diff
/run pytest tests/             # 运行命令并将输出发给 AI
/test pytest tests/            # 运行测试，失败时自动修复
/architect                     # 切换到 Architect 模式
/ask                           # 只问不改代码
/clear                         # 清空对话历史

配置文件

在项目根目录创建 .aider.conf.yml：

model: sonnet
auto-commits: true
gitignore: true
dark-mode: true
map-tokens: 2048
lint-cmd: "ruff check --fix"
test-cmd: "pytest tests/ -x"

环境变量配置（~/.bashrc 或 ~/.zshrc）：

export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."

Architect 模式

Architect 模式采用"先规划再编码"的两阶段策略：

1. Architect 模型（如 Opus）负责分析需求、制定修改方案
2. Editor 模型（如 Sonnet）负责执行具体的代码修改

启用方式：

aider --architect --model opus --editor-model sonnet

适合复杂重构任务，Architect 模型产出高质量方案，Editor 模型高效执行。

与同类工具对比

特性	Aider	Claude Code	Cursor
运行环境	纯终端	纯终端	IDE
Git 集成	自动 commit	手动 commit	无
文件管理	/add /drop 手动	自动检索	自动检索
上手难度	低	低	极低
适合场景	精确控制上下文	探索式开发	GUI 偏好者

典型使用场景

• 快速修 Bug：描述问题，Aider 定位并修复，自动 commit
• 添加新功能：/add 相关文件，描述需求，多文件协同修改
• 代码重构：Architect 模式规划方案，批量重命名/拆分/合并
• 写测试：/add 源码文件，要求生成对应的单元测试
• 代码审查辅助：/ask 模式分析代码逻辑，不做修改

安装

pip install aider-chat
# 或使用 pipx 隔离安装
pipx install aider-chat