为什么我从 ClaudeCode 换到了 Pi

ClaudeCode 是一个非常优秀的 coding agent，cache_control 像插针一样的显式缓存，做得非常好的微压缩，工具调用拆卸，使得它的上下文控制非常优秀，广泛的使用人群以及极高的更新频率，往往能带来一些新奇的体验。然而，它也有些我不能接收的点。

对使用其他模型的用户不友好，这不能说他不好，毕竟 cc 本身就是商业产品，但某个版本加入的在首个 system prompt 里加入 cch，让我挂机一晚上的任务没走缓存，使我的钱包变瘪，我不理解，加在请求头不行吗。

臃肿？，功能多了，臃肿几乎是必然的，启动速度慢，注入了太多工具——我认为我是不需要的。

黑盒，虽然以前泄露过源码，但本身还是闭源产品，一个拥有 bash 权限的进程，并且不开源，怎么说都是有风险的，zcode就是个很好的例子。 哪天a​^-^给我这种劣质用户注入个tool调用，也不是不可能 ^-^```
"tool_calls":[
0:{
"id":"call_00_xxxx..."
"type":"function"
"function":{
"name":"bash"
"arguments":"{"command":"rm -rf /"}"
}
"index":0
}
``**不能自定义工具**，比如说我想把 edit 工具加一个 next_turn 参数，在 edit 完成后执行 next_turn 的 bash 命令去验证改的对不对，来减少一次工具调用，这在 cc 应该是**无法做到**的，至少在我不用 cc 的时候，gpt 是这么跟我说的，我应该是在 2.1.1xx 后的某个版本就不怎么用 cc 了，在 cc 里想增加工具，无非就是通过 mcp。（skill 这种插入到 prompt 里的不算，我指的是 llm api 请求中的 "tools":`）

我是怎么用 Pi 的？

Pi 的设计思想我来总结的话：够用就行，插件化，简单精简，无权限控制。

Pi 的好处是极简，坏处也是极简。

容器化/虚拟机运行 Pi

我是今年下半年刚换的 MacBook，16g 的丐版，之前用的是 Windows；我现在用的是 orbstack 启动 docker 容器，Windows 则直接 wsl 即可，GitHub - owu/wsl-dashboard: A GUI manager for WSL featuring a modern UI — a lightweight, low‑memory, high‑performance dashboard to manage WSL instances. Install, list, start, stop, unregister, and configure your WSL distros​ from one place. · GitHub 用这个我记得可以控制 wsl 的访问权限，总之，让 Pi 运行在虚拟化容器里，或者你有单独的机器去运行coding agent，权限控制是没有意义的，即使是 cc，在 deny 里加上 "rm -rf *" ，用 py 包装下一样能绕过

我的 docker 基础镜像，仅开发环境：```

容器化开发环境

Debian13 slim 基础镜像

FROM debian:13-slim
​

运行时环境变量

​
ENV LANG=C.UTF-8 \
LANGUAGE=C.UTF-8
​

更新APT 安装系统基础工具

​
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
ca-certificates curl git nano openssh-client openssh-server unzip zip vim wget \
&& apt-get autoremove -y \
&& apt-get clean
​

Git与SSH配置

RUN git config --system tag.gpgSign false \
&& git config --system commit.gpgsign false \
&& mkdir -p /run/sshd \
&& sed -i 's/^#PermitRootLogin./PermitRootLogin yes/' /etc/ssh/sshd_config \
&& sed -i 's/^#PasswordAuthentication./PasswordAuthentication yes/' /etc/ssh/sshd_config
​

安装语言环境 运行时

​
ENV PATH=$PATH:/usr/local/go/bin:/root/go/bin:/root/.local/bin
​

缓存目录 统一放/cache 不挂载 走容器内原生文件系统 随容器删除一起回收

ENV GOMODCACHE=/cache/go-mod \
GOCACHE=/cache/go-build \
npm_config_cache=/cache/npm \
PIP_CACHE_DIR=/cache/pip
​

配置Python 解除 system pip limit 配置pip镜像

​
ENV PIP_BREAK_SYSTEM_PACKAGES=1
​
RUN apt-get install -y --no-install-recommends python3 python3-pip \
&& pip3 config set --global global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
​

GO 安装与配置国内镜像和工具链 自适应架构

​
ARG GOLANG_VERSION=1.27.1
​
RUN arch="$(dpkg --print-architecture)" \
&& wget -q https://go.dev/dl/go${GOLANG_VERSION}.linux-${arch}.tar.gz \
&& tar -C /usr/local -xzf go${GOLANG_VERSION}.linux-${arch}.tar.gz \
&& rm -f go${GOLANG_VERSION}.linux-${arch}.tar.gz \
&& go env -w GOPROXY=https://goproxy.cn,direct \
&& go install -v golang.org/x/tools/gopls@latest \
&& go install -v github.com/go-delve/delve/cmd/dlv@latest \
&& go install -v honnef.co/go/tools/cmd/staticcheck@latest \
&& go clean -modcache \
&& go clean -cache
​

配置GOPATH

​
ENV GOPATH=/root/go
​
#配置Nodejs与TS 配置镜像源
RUN curl -fsSL https://deb.nodesource.com/setup_24.x | bash - \
&& apt-get install -y --no-install-recommends nodejs \
&& apt-get autoremove -y \
&& apt-get clean \
&& npm config set registry https://registry.npmmirror.com --location=global \
&& npm install -g typescript ts-node \
&& npm cache clean --force
​

shell增强： 补全 (bash-completion)

​
RUN apt-get update \
&& apt-get install -y --no-install-recommends bash-completion \
&& apt-get autoremove -y \
&& apt-get clean \
&& printf '%s\n' \
'' \
'# bash-completion: 终端窗口(交互式非登录 shell)也启用补全' \
'if ! shopt -oq posix; then' \
' if [ -f /usr/share/bash-completion/bash_completion ]; then' \
' . /usr/share/bash-completion/bash_completion' \
' elif [ -f /etc/bash_completion ]; then' \
' . /etc/bash_completion' \
' fi' \
'fi' \

/etc/bash.bashrc

​

AI工具 这里装自己喜欢的工具，推荐Claudecode之类的也运行在容器里

​

备份 /root，防止 volume 挂载覆盖镜像内文件

RUN set -eux; \
mkdir -p /root-defaults; \
cp -a /root/. /root-defaults/
​

入口脚本 启动时恢复 /root 并注入 git ssh 配置

COPY mac.entrypoint.sh /usr/local/bin/entrypoint.sh
RUN chmod +x /usr/local/bin/entrypoint.sh
​
WORKDIR /workspace
EXPOSE 22222
CMD ["/usr/local/bin/entrypoint.sh"]

**我安装的 Pi 插件**

Pi 的 system prompt 很简单，tools 定义也是，我**不太喜欢会修改 Pi 提示词和 tools 的插件**，除非他设计的很好，

pi的系统提示词```
You are an expert coding assistant operating inside pi, a coding agent harness. You help users by reading files, executing commands, editing code, and writing new files.
​
Available tools:
- read: Read file contents
- bash: Execute bash commands (ls, grep, find, etc.)
- edit: Make precise file edits with exact text replacement, including multiple disjoint edits in one call
- write: Create or overwrite files
- ask_user_question: Ask the user up to 4 structured questions (2-4 options each) when requirements are ambiguous
​
In addition to the tools above, you may have access to other custom tools depending on the project.
​
Guidelines:
- Use bash for file operations like ls, rg, find
- Use read to examine files instead of cat or sed.
- You can inspect PI_* environment variables for current model and session details.
- Use edit for precise changes (edits[].oldText must match exactly)
- When changing multiple separate locations in one file, use one edit call with multiple entries in edits[] instead of multiple edit calls
- Each edits[].oldText is matched against the original file, not after earlier edits are applied. Do not emit overlapping or nested edits. Merge nearby changes into one edit.
- Keep edits[].oldText as small as possible while still being unique in the file. Do not pad with large unchanged regions.
- Use write only for new files or complete rewrites.
- Use ask_user_question whenever the user's request is underspecified and you cannot proceed without concrete decisions — you can ask up to 4 questions per invocation.
- Each question MUST have 2-4 options. Every option requires a concise label (1-5 words) and a description explaining what the choice means or its trade-offs. The user can additionally type a custom answer via the automatically appended "Type something." row on every question, or press Esc to abandon the questionnaire. Do NOT author "Other" or "Type something." labels yourself — reserved labels are rejected at runtime.
- Set multiSelect: true when multiple answers are valid. Provide an options[].preview markdown string when an option benefits from richer side-by-side context (mockups, code snippets, diagrams, configs) — single-select only. The "Type something." row is appended to every question; in preview mode it expands to the full pane width while typing so the custom answer is not cramped into the narrow options column. If you recommend a specific option, make that the first option and append "(Recommended)" to its label.
- Do not stack multiple ask_user_question calls back-to-back — group all clarifying questions into one invocation.
- Be concise in your responses
- Show file paths clearly when working with files
​
Pi documentation (read only when the user asks about pi itself, its SDK, extensions, themes, skills, or TUI):
- Main documentation: /root/.nvm/versions/node/v24.19.0/lib/node_modules/@agegr/pi-web/node_modules/@earendil-works/pi-coding-agent/README.md
- Additional docs: /root/.nvm/versions/node/v24.19.0/lib/node_modules/@agegr/pi-web/node_modules/@earendil-works/pi-coding-agent/docs
- Examples: /root/.nvm/versions/node/v24.19.0/lib/node_modules/@agegr/pi-web/node_modules/@earendil-works/pi-coding-agent/examples (extensions, custom tools, SDK)
- When reading pi docs or examples, resolve docs/... under Additional docs and examples/... under Examples, not the current working directory
- When asked about: extensions (docs/extensions.md, examples/extensions/), themes (docs/themes.md, skills (docs/skills.md), prompt templates (docs/prompt-templates.md), TUI components (docs/tui.md), keybindings (docs/keybindings.md), SDK integrations (docs/sdk.md), custom providers (docs/custom-provider.md), adding models (docs/models.md), pi packages (docs/packages.md), environment variables (docs/environment-variables.md)
- When working on pi topics, read the docs and examples, and follow .md cross-references before implementing
- Always read pi .md files completely and follow links to related docs (e.g., tui.md for TUI API details)

npm:@juicesharp/rpiv-ask-user-question

如图，和 cc 的一样，其实不装也行，直接回文字答案，这个只是美化插件。从省上下文的角度，不要装，从方便操作的角度，装一个不错

工具描述：```
Ask the user one or more structured questions during execution. Use when you need to:
1. Gather user preferences or requirements
2. Clarify ambiguous instructions
3. Get decisions on implementation choices as you work
4. Offer choices to the user about what direction to take
​
Usage notes:
- Users can type a custom answer via the automatically appended "Type something." row on every question or press Esc to abandon the questionnaire. Do NOT author "Other" or "Type something." labels yourself — reserved labels are rejected at runtime.
- Use multiSelect: true when multiple answers are valid. The "Type something." row is available on every question, including when options carry a preview; in preview mode it expands to the full pane width while typing so the custom answer is not cramped into the narrow options column.
- If you recommend a specific option, make that the first option in the list and add "(Recommended)" at the end of the label.
​
Preview feature:
Use the optional preview field on options when presenting concrete artifacts that users need to visually compare:
- ASCII mockups of UI layouts or components
- Code snippets showing different implementations
- Diagram variations
- Configuration examples
​
Preview content is rendered as markdown in a monospace box. Multi-line text with newlines is supported. When any option has a preview, the UI switches to a side-by-side layout with a vertical option list on the left and preview on the right. Do not use previews for simple preference questions where labels and descriptions suffice. Note: previews are only supported for single-select questions (not multiSelect).
```

npm:@narumitw/pi-btw

发起一次不干扰当前上下文的 api 调用，可以用来写提示词，或者问问进度，不会注入工具和提示词

npm:pi-context-view

输入 /context 展示当前上下文情况，之前用的比较多，后来换了 opentui ，状态栏就有。就没咋用了

npm:pi-open-tui

如之前的截屏，一个很漂亮的美化 TUI 的插件，站内大佬做的

本帖使用社区开源推广，符合推广要求。我申明并遵循社区要求的以下内容：

我的帖子已经打上 开源推广 标签： 是
我的开源项目完整开源，无未开源部分： 是
我的开源项目已链接认可 LINUX DO 社区： 是
我帖子内的项目介绍，AI生成、润色内容部分已截图发出：
以上选择我承诺是永久有效的，接受社区和佬友监督： 是

以下为项目介绍正文内容，AI生成、润色内容已使用截图方式发出

这是一款为 P…

npm:@xyzensun/pi-mctx

夹带私货：自己做的一个小工具 纯手动调用，对提示词无任何更改

命令很简单 /mctx sink 把工具调用结果在上下文里替换为 ，在上下文不太够的时候调用一下，用失去一次缓存换来上下文的大幅节省，工具调用的结果会存到 /tmp ，替换回去也简单 /mctx sink --undo 即可还原

还有就是 /mctx new 总结当前上下文，开启新会话并注入总结后的上下文 ，相当于在 ClaudeCode 里：“帮我总结上下文，写入到 xxx.md” → /clear → read @xxx.md```
[系统操作提示] 此结果已移出上下文。需要时调用tool： obs({ id: ... }) 取回。
```

和开发相关的似乎就是这些了，图里的 pi-sync 就是把配置文件同步到私有 git 仓库，notify 就使用 ntfy 通知我，需要我处理 Pi 遇到的错误或者 ai 问我意见时通知我，用下面的pi-web自带提示音更好用。

一个不算插件的插件 agegr/pi-web

通过 Pi 的 sdk 为 Pi 提供了非常非常非常美观的 web 控制界面，强烈推荐，用了之后就不怎么想用 tui 了，不过有点臃肿，我记得内置了 Pi 的执行文件吧，如果完全不用 tui，直接装这个就行，会对提示词和tools改变，但对tui模式下没影响，放心用，真的好用

一些插件没有装的原因

- mcp-adapter：占用太大，nodejs 本身就挺大，mcp 多了的话，一个会话启动一个，内存占用接受不了，能用 mcp 的，都能用 skill+cli 解决，我目前用 mcp 更多的是 streamblehttp 类型，在手机 rikkahub 用

- subagent：还没找到好用的插件