Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add jianshuo/claude-skills --skill wjs-eating-and-growinggit clone --depth 1 https://github.com/jianshuo/claude-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/jianshuo/claude-skills/wjs-eating-and-growing)<a href="https://agentmods.dev/skills/jianshuo/claude-skills/wjs-eating-and-growing"><img src="https://agentmods.dev/badge/skills/jianshuo/claude-skills/wjs-eating-and-growing/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/jianshuo/claude-skills/wjs-eating-and-growing"><img src="https://agentmods.dev/badge/skills/jianshuo/claude-skills/wjs-eating-and-growing.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00168 | $0.02976 |
| Opus 5 | $0.00084 | $0.01488 |
| Sonnet 5 | $0.00034 | $0.00595 |
| Haiku 4.5 | $0.00017 | $0.00298 |
Grade A, and why
wjs-eating-and-growing scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 196 lines — stays where its author put it; the contents beside it link to each section on GitHub.
wjs-eating-and-growing(吃一堑长一智)
这个 skill 在干什么
帮王建硕在每次"吃堑"之后,按 5 步走一遍:
堑 → 自动输出 → 旧权重 → 新参数 → 替代动作
不开练习菜单,不排打卡频次,不要他每天冥想 10 分钟。最终只交付一条:下次这件事再发生的那一秒,他具体做什么不一样。
但在"下次做什么"之前,必须先穿过一层:这次该更新的是哪一条旧解释模式 / 想训练的新参数是什么。这是从「情绪复盘」推进到「行为训练」的关键——不更新解释模式,再多 next step 也只是症状管理。
底层框架(来自 Mars 任鑫《庄子,业力和大模型》):
| 层 | 类比 | 改它靠什么 | 改完会变吗 |
|---|---|---|---|
| L1 外挂文档 | 笔记、读一篇文章 | 读 / 记 | ❌ 不变 |
| L2 推理链路 | 临场慢思考 | 外部触发器 | 状态好时变,状态差时被冲垮 |
| L3 底层权重 | 不假思索的第一反应 | 高梯度小数据练习 | ✅ 默认输出真的变 |
这个 skill 默认目标永远是 L3 —— 不假思索的第一反应。
The Iron Law:知道道理 ≠ 临场用上 ≠ 默认反应。光把"道理"加到笔记里,是在重复作者吐槽过的状态——"Obsidian 两千条笔记,下次被怼还是先暴怒三秒"。
怎么用 — 5 步交互式
一步一问。每一步只问一个问题,等用户回答,再走下一步。不可跳,不可合并,不可一次性把五个问题罗列出来要他一起答。
用户在情绪里走出来的速度是慢的,跳步等于回到 L1 笔记。
Step 1 — 这次"堑"是什么?
"用一句话写清楚发生了什么。
不要解读,不要分析,先把事实落地。比如:
- 朋友临时爽约了
- 对方一句话让我瞬间破防
- 我又拖延到最后一刻才做事"
为什么:先把"事件"从"对事件的解释"里剥出来。后面要更新的是解释,不是事件——必须先看清这两层不是同一个东西。
如果用户一上来就给的是解释("他不尊重我"),追问一次:"这是你的解读。先退回一步——纯事实层面、外面装个摄像头能拍到的那件事,是什么?"
Step 2 — 我当时的自动输出是什么?
"不是你事后的解释,是第一反应。
那个瞬间脑子里蹦出来的原话——比如:
- "他不尊重我。"
- "他就是看不起我。"
- "我完了,我总是这样。"
- "我必须马上回击。"
如果一时想不起,就先回到身体——哪里紧了、热了?心跳起来了吗?拳头握住了吗?身体先反应,话紧随其后。把那句话抓住。"
为什么:自动输出是 L3 权重的可观测投影。文章原话——"这个瞬间的 learning rate 远高于事后复盘"。事后再聪明的总结,梯度都已经凉了,改不动 L3。
如果用户给的是事后视角("我应该当时冷静一点"、"我那时候不该…"),追问一次:"这是事后视角。回到那一秒——脑子里实际蹦出来的那句话是什么?"
Step 3 — 这个输出背后的旧权重是什么?
"也就是:你为什么总往这个方向解释?
不是这一次为什么,是你这个人为什么会对这类事情默认这样反应。比如:
- 我很怕被忽视
- 我对失控特别敏感
- 我从小对批评高度警觉
- 我习惯把别人的状态解释成针对我
- 我遇到压力就会逃避
用一句话写。"
为什么:这一步是关键。真正该更新的不是事件本身,而是你解释事件的旧模式。
如果用户停在症状层("我又急了"、"我意志力不够"、"我就是这样"),追问一次:"这是症状。再下一层——为什么这个症状会发生在你身上、不发生在别人身上?底下那个一直没解决的旧情绪、一直被违反的假设、一直没补的能力,是什么?"
Step 4 — 这次我想训练哪个新参数?
"不要太大,不要空泛,要具体。
不要写:
- "以后我要情绪稳定。"
- "我要更成熟。"
- "我要开悟。"
要写成针对一类具体情境的新响应模式:
- 遇到别人冷淡时,不立刻理解成否定我
- 愤怒起来时,先停三秒再说话
- 被放鸽子时,先确认事实,不脑补动机
- 焦虑时,先描述身体感觉,不马上相信想法
你这次想训练的那一条是?"
好的新参数的 3 条标准(用户答完核对一遍):
- 指向一类触发情境 —— "遇到 X 时…",不是"这次以后…",也不是"我整个人要变成 Z"
- 是针对那一类触发的具体新响应 —— "不立刻 Y" / "先做 W 再 V",不是"我要更成熟 / 更冷静 / 想得开"
- 小到能落地一类情境 —— 一条参数对应一类触发,不是包打天下的人生修行目标
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 196 lines · 168 tokens per session scan A ba6a4e9b5486
wjs-eating-and-growing is a skill published in the GitHub repository jianshuo/claude-skills (129 stars, last pushed 22d ago), licensed MIT. It adds 168 tokens to every session and 2,976 once invoked, about $0.0008 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
domain
When you want to brainstorm and check available .com domains for a new project — brand naming, aftermarket pricing (HugeDomains / Afternic / Sedo / Dan), USPTO trademark screening, and social handle availability. Built on Laura Roeder's "work backwards from availability, not from a name you fell in love with"…
company-brain
Your team's shared, AI-ready knowledge base — people, companies, meetings, SOPs, and decisions structured so Claude can answer questions on your team's behalf. Team-scope sibling to second-brain (which is personal-scope). Seven modes — capture (drop something into the right structured dir), compile (process into wiki…
slide-deck
When you want to draft, update, convert, or export a slide deck for a React/Next.js slide system (${SLIDEDECKREPO:-$HOME/code/your-slide-deck-site}/src/app/slides/). Writes TypeScript Slide[] arrays using your primitives (Eyebrow, Heading, Accent, Body, BulletList, Divider, TwoCol, GradientText), 12 cycling brand…
google-calendar
Interact with Google Calendar - list calendars, view events, create/update/delete events, and find free time. Use when user asks to: check calendar, schedule a meeting, create an event, find available time, list upcoming events, delete or update a calendar event, or respond to meeting invitations. Lightweight…
service-desk
Runs the IT service desk — intake, triage, prioritization, escalation, knowledge, and the metrics that improve service rather than distort it. Use this to set up or fix a service desk, design ticket priority and escalation, reduce repeat contacts, structure a knowledge base, or work out why a desk hitting its targets…
collaboration-platform-administration
Administers the email, chat, meeting and file-sharing platform the organization runs on — tenant and domain configuration, mail authentication and routing, phishing and spam controls, shared mailboxes and distribution groups, external sharing and guest access, permission sprawl in file storage, and retention and legal…