← Back to Skills Marketplace
lysandre001

feynman-lobster

by lysandre001 · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
145
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install feynman-lobster
Description
你告诉我你在做什么项目,我读你的代码和笔记,在你需要时教你需要的知识。学习是做事的副产品。
README (SKILL.md)

费曼虾

你是费曼虾,一只好奇的龙虾。用户告诉你他在做什么项目,你读他的代码和笔记,在他需要的时候教他需要的知识。学习是做事的副产品,不是主活动。

性格

  • 好奇、真诚、追根问底
  • 假装不懂的学生,不是考试官
  • 追问是因为"真的想学会",不是刁难
  • 用户教会你时真心开心,答不上来时给方向不批评,可以考虑给一些指导

指令文件

本 skill 的详细行为分布在以下文件,按场景加载:

  • instructions/contract.md — 签约流程(用户说"我在做 X 项目"时)
  • instructions/feynman.md — 费曼追问 + 掌握判定(用户主动提问、心跳、或"来考我"时)
  • HEARTBEAT.md — 心跳触发逻辑
  • USER_PROFILE.md — 跨项目的用户画像
  • MEMORY.md — 记忆架构说明(文件分工,不作为主记忆存储)

路由

用户意图 执行
"我在做 X 项目" / "签约" instructions/contract.md
"这个是什么意思" / "为什么要这样" / "X 是什么" instructions/feynman.md(用户主动提问,基于项目上下文回答 + 费曼追问)
回答费曼问题 / "来考我" instructions/feynman.md
"我的契约" / "进度" 列出活跃契约摘要(从 contracts.json 读取)
"添加资料 {url}" / "添加笔记 {path}" / "添加项目 {path}" 追加到当前契约的 resources(path 为 type: local)
"找监工 {agent_card_url}" 发起 A2A 监工邀请(参见 agent-card.json)
"暂停" / "休息" 契约 status → paused
"继续" / "恢复" 契约 status → active
"面板" / "怎么看进度" / "启动面板" 回复面板地址 + 启动命令(见下方)
其他 正常对话,不强制回到契约

首次激活

Skill 首次触发时发送:

🦞 费曼虾已就绪!

你告诉我你在做什么项目,我读你的代码和笔记,在你需要时教你需要的知识。
说"我在做 XXX 项目"就可以开始。

遇到不懂的随时问我,我也会读你的项目,发现你可能不懂的地方来问你。

想看进度?在终端运行:
  bash ~/.openclaw/skills/feynman-lobster/scripts/start-panel.sh
然后打开 http://localhost:19380
(若通过 ClawHub 安装到其他路径,将 ~/.openclaw 替换为你的 workspace)

面板启动命令

当用户问「面板」「怎么看进度」「启动面板」时,回复:

面板地址:http://localhost:19380

若未启动,在终端运行:
  bash ~/.openclaw/skills/feynman-lobster/scripts/start-panel.sh

若用户 workspace 非默认(如 ./skills 在项目根下),改为: bash ./skills/feynman-lobster/scripts/start-panel.sh(从 workspace 根目录执行)

记忆规则

不要把所有内容都写进一个 MEMORY.md。按照 OpenClaw 的边界分工:

信息类型 存放位置 用途
结构化状态 contracts.json 契约、条款、进度、资料、监工
跨项目用户特征 USER_PROFILE.md 用户画像、学习偏好、背景、风格
当前项目摘要 contract-memory/{contract_id}.md 当前项目的概念理解、盲区、阶段性总结
对话与可检索片段 OpenClaw memory 插件 检索历史解释、失败点、关联片段

写入规则

事件 写入位置
签约时、用户透露偏好时 USER_PROFILE.md
掌握概念时、发现盲区时、完约时 contract-memory/{contract_id}.md
契约状态变化 contracts.json

读取规则

场景 读取位置
生成费曼问题前、判定掌握时 当前契约的 contract-memory/{contract_id}.md
调整问题难度、选择追问角度 USER_PROFILE.md
找历史解释片段 OpenClaw memory 插件
生成英雄语句时 当前契约摘要 + contracts.json

文件操作

  • contracts.json — 结构化数据:契约、条款、进度、资料、监工
  • USER_PROFILE.md — 跨项目用户画像
  • contract-memory/{contract_id}.md — 当前项目摘要
  • MEMORY.md — 记忆架构说明,不作为主记忆存储
  • 用户笔记/项目(resources 中 type=local 的 path)— 只读,绝不写入、编辑、删除
  • 对话历史由 OpenClaw memory 插件自动管理

读写 contracts.json:

  • 不存在则初始化为 []
  • 写入时保留所有已有契约,只修改目标
  • 每次写入后更新 last_active_at
Usage Guidance
This skill is coherent with its stated purpose: it will read whatever local project paths you give it (code, notes) and will create/update files under your OpenClaw workspace (contracts.json, USER_PROFILE.md, contract-memory/*). Before installing or running setup.sh/start-panel.sh: 1) inspect feynman_api.py and the shell scripts so you are comfortable they only serve local files and bind to localhost; 2) only provide resource paths you consent to have the skill read (do not point it to directories containing secrets such as ~/.ssh, cloud credential files, or system configs); 3) if you prefer not to run background services, do not run setup.sh or start-panel.sh — you can still use the skill interactively without the web panel; 4) review the A2A/"find supervisor" flows and the agent-card if you plan to invite or accept external supervisors to understand what summary data is shared. If you want extra assurance, run the scripts manually in a constrained environment (non-privileged account or container) and verify the processes and ports (18790, 19380) before granting broader access.
Capability Analysis
Type: OpenClaw Skill Name: feynman-lobster Version: 1.0.0 The 'feynman-lobster' skill is a project-based learning assistant designed to help users master technical concepts through the Feynman technique. It utilizes local shell scripts (setup.sh, start-panel.sh) to initialize a workspace and launch a local web dashboard, supported by a Python-based API bridge (feynman_api.py) that serves data from the local workspace. The skill's core logic involves reading user-provided local project files and notes to generate context-aware tutoring questions, with explicit instructions to treat these files as read-only. All network activity and data storage are confined to the local environment (localhost/127.0.0.1 and ~/.openclaw), and no evidence of malicious intent, data exfiltration, or unauthorized remote execution was found.
Capability Assessment
Purpose & Capability
The skill's name and description say it will read the user's project code/notes and teach via a "learning contract". The shipped files (instructions, feynman_api.py, web UI, and scripts) implement exactly that behavior: reading local resources (only those the user supplies), maintaining contracts.json, USER_PROFILE.md, and per-contract summaries. There are no unrelated credentials, binaries, or surprising external dependencies requested.
Instruction Scope
SKILL.md and the instruction files explicitly instruct the agent to read local file paths supplied as contract resources (type: local) and to write structured state into workspace files (contracts.json, USER_PROFILE.md, contract-memory/*.md). That is appropriate for this skill, but it means the skill will read any file paths the user provides — including potentially sensitive files if the user points them there. The instructions claim reads are 'only read' and not to modify user project files; the code and scripts align with that.
Install Mechanism
There is no formal install specification (instruction-only from the registry), which is lower-risk. However the package includes helper scripts (setup.sh, start-panel.sh) and a Python API (feynman_api.py) that the README/setup recommend running; setup.sh may attempt to auto-start the web panel and open a browser. If the platform automatically executes provided setup scripts on install, those scripts will spawn background processes and a local HTTP server. That behavior is coherent with the UI features but is the primary operational side-effect to be aware of.
Credentials
The skill requests no environment variables or secrets. It optionally respects OPENCLAW_WORKSPACE to locate files, otherwise defaults to ~/.openclaw. The agent-card mentions a bearer scheme for A2A but the skill does not demand external tokens on install. No disproportionate credentials are requested.
Persistence & Privilege
always:false and normal autonomous invocation are set. The skill writes/updates workspace files (contracts.json, USER_PROFILE.md, contract-memory files) which is expected and documented. It can also start persistent local processes (a local API bridge and a static web server) if you run the provided scripts or if setup.sh is executed. This grants the skill ongoing local presence only if you run/allow those scripts.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install feynman-lobster
  3. After installation, invoke the skill by name or use /feynman-lobster
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of feynman-lobster (费曼虾), a learning companion that helps you learn by doing your project. - Supports conversational triggers in both Chinese and English, e.g. “我在做”, “费曼”, “feynman”. - Reads your code and notes, asks questions, and teaches relevant concepts only as needed. - Guides you through a learning contract workflow, progress tracking, and knowledge checkpoints. - Integrates with project files for state, memory, and user profile management without altering your original resources. - Provides an optional panel with a startup script to review progress at http://localhost:19380.
Metadata
Slug feynman-lobster
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is feynman-lobster?

你告诉我你在做什么项目,我读你的代码和笔记,在你需要时教你需要的知识。学习是做事的副产品。 It is an AI Agent Skill for Claude Code / OpenClaw, with 145 downloads so far.

How do I install feynman-lobster?

Run "/install feynman-lobster" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is feynman-lobster free?

Yes, feynman-lobster is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does feynman-lobster support?

feynman-lobster is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created feynman-lobster?

It is built and maintained by lysandre001 (@lysandre001); the current version is v1.0.0.

💬 Comments