为什么值得关注

它通过将行为问责直接嵌入AI代理的工作流程，利用代理的天然上下文实时检测并打断反效率模式，解决了常见的注意力分散问题。

适合谁使用

使用Claude Code、Codex或OpenCode作为主要编程助手的开发者
有ADHD倾向、容易分心且难以完成任务的用户
同时管理多个项目和优先事项的自由职业者或独立创业者
希望将日常工作与长期目标对齐但难以保持专注的人

典型使用场景

建立包含目标和个人陷阱的个人档案，当代理检测到你在重复已知坏习惯时自动提醒
获得一张按目标排序、不会淹没你的任务列表，而不是冗长的待办清单
当你长时间卡在非核心任务上时收到一句及时的提醒
需要重新聚焦时随时要求进行一致性检查

项目优势

主动感知：导师会扫描你当前的对话活动，一旦发现你在重蹈覆辙就会主动提醒，无需手动检查
隐私优先：所有数据以本地纯Markdown文件存储，不向外发送任何信息，卸载后只留下两个用户文件
跨代理便携：支持Claude Code、Codex和OpenCode，共享同一个个人档案，导师随你移动
诚实的自我反省：初始化对话迫使你在清醒状态下列出自己的陷阱，使后续提醒更有说服力

使用前须知

效果因底层模型而异，README中指出在Claude上提醒效果最好，在Codex和OpenCode上稍逊
需要一次大约十分钟的设置访谈，对只想快速提高效率的用户来说可能略显繁琐
仅适用于支持的代理环境（Claude Code、Codex、OpenCode），不是独立的应用程序或浏览器扩展

README 快速开始

mentor

Turn your agent into a mentor. It knows what you're working toward, and how you get in your own way.

Install

Claude Code

claude plugin marketplace add hcsum/dont-let-me
claude plugin install mentor@dont-let-me

Then run /mentor:init to set up your profile, and /mentor:check any time you want the alignment check on demand.

Codex

codex plugin marketplace add hcsum/dont-let-me --ref main
codex plugin add mentor@dont-let-me

Then run $mentor-init to set up your profile, and $mentor-check any time you want the alignment check on demand.

Codex makes you trust a hook before it runs: open /hooks, review the SessionStart entry, trust it. Until you do, the profile never loads.

OpenCode

OpenCode has no plugin marketplace for this, so clone it and point the config at the two pieces:

git clone https://github.com/hcsum/dont-let-me.git ~/.mentor/plugin

// ~/.config/opencode/opencode.json
{
  "instructions": ["~/.mentor/plugin/stance.md", "~/.mentor/profile.md"],
  "skills": { "paths": ["~/.mentor/plugin/skills"] }
}

Then run /mentor-init to set up your profile, and /mentor-check any time you want the alignment check on demand.

In my own use, Claude Code is where this works best. Codex and OpenCode both run it, but how well the nudges land depends on the model behind them, and I've had better results with Claude.

1. It interviews you

Setup asks you questions, one at a time, for about ten minutes. Run /mentor:init in Claude Code, $mentor-init in Codex, /mentor-init in OpenCode. Two things come out of it.

Your long-term goals. What you're actually working toward. It pushes back if you answer vaguely.

Your anti-patterns. The ways you reliably get in your own way, written down while you're still honest about them, so it can hold you to them later when you're not:

- [ship] don't let me build in private forever, waiting for it to be ready
- [explore] don't let me keep opening new threads while nothing is finished

You do this once and rarely touch it again.

2. Then you just talk to it

mentor, what's on my plate today?

Ask what's on your plate and it won't read you the whole todo list. It picks the few that serve your main goals and names the first move on one, small enough to start now.

add "ship the landing page" to my todos

项目描述

Turn your agent into a mentor. It knows what you're working toward, and how you get in your own way.

dont-let-me